./000077500000000000000000000000001504515377700112315ustar00rootroot00000000000000./BreakingChanges.adoc000066400000000000000000000317371504515377700151070ustar00rootroot00000000000000= Upcoming breaking changes The Git project aims to ensure backwards compatibility to the best extent possible. Minor releases will not break backwards compatibility unless there is a very strong reason to do so, like for example a security vulnerability. Regardless of that, due to the age of the Git project, it is only natural to accumulate a backlog of backwards-incompatible changes that will eventually be required to keep the project aligned with a changing world. These changes fall into several categories: * Changes to long established defaults. * Concepts that have been replaced with a superior design. * Concepts, commands, configuration or options that have been lacking in major ways and that cannot be fixed and which will thus be removed without any replacement. Explicitly not included in this list are fixes to minor bugs that may cause a change in user-visible behavior. The Git project irregularly releases breaking versions that deliberately break backwards compatibility with older versions. This is done to ensure that Git remains relevant, safe and maintainable going forward. The release cadence of breaking versions is typically measured in multiple years. We had the following major breaking releases in the past: * Git 1.6.0, released in August 2008. * Git 2.0, released in May 2014. We use . release numbers these days, starting from Git 2.0. For future releases, our plan is to increment in the release number when we make the next breaking release. Before Git 2.0, the release numbers were 1.. with the intention to increment for "usual" breaking releases, reserving the jump to Git 2.0 for really large backward-compatibility breaking changes. The intent of this document is to track upcoming deprecations for future breaking releases. Furthermore, this document also tracks what will _not_ be deprecated. This is done such that the outcome of discussions document both when the discussion favors deprecation, but also when it rejects a deprecation. Items should have a clear summary of the reasons why we do or do not want to make the described change that can be easily understood without having to read the mailing list discussions. If there are alternatives to the changed feature, those alternatives should be pointed out to our users. All items should be accompanied by references to relevant mailing list threads where the deprecation was discussed. These references use message-IDs, which can visited via https://lore.kernel.org/git/$message_id/ to see the message and its surrounding discussion. Such a reference is there to make it easier for you to find how the project reached consensus on the described item back then. This is a living document as the environment surrounding the project changes over time. If circumstances change, an earlier decision to deprecate or change something may need to be revisited from time to time. So do not take items on this list to mean "it is settled, do not waste our time bringing it up again". == Procedure Discussing the desire to make breaking changes, declaring that breaking changes are made at a certain version boundary, and recording these decisions in this document, are necessary but not sufficient. Because such changes are expected to be numerous, and the design and implementation of them are expected to span over time, they have to be deployable trivially at such a version boundary, prepared over long time. The breaking changes MUST be guarded with the a compile-time switch, WITH_BREAKING_CHANGES, to help this process. When built with it, the resulting Git binary together with its documentation would behave as if these breaking changes slated for the next big version boundary are already in effect. We also have a CI job to exercise the work-in-progress version of Git with these breaking changes. == Git 3.0 The following subsections document upcoming breaking changes for Git 3.0. There is no planned release date for this breaking version yet. Proposed changes and removals only include items which are "ready" to be done. In other words, this is not supposed to be a wishlist of features that should be changed to or replaced in case the alternative was implemented already. === Changes * The default hash function for new repositories will be changed from "sha1" to "sha256". SHA-1 has been deprecated by NIST in 2011 and is nowadays recommended against in FIPS 140-2 and similar certifications. Furthermore, there are practical attacks on SHA-1 that weaken its cryptographic properties: + ** The SHAppening (2015). The first demonstration of a practical attack against SHA-1 with 2^57 operations. ** SHAttered (2017). Generation of two valid PDF files with 2^63 operations. ** Birthday-Near-Collision (2019). This attack allows for chosen prefix attacks with 2^68 operations. ** Shambles (2020). This attack allows for chosen prefix attacks with 2^63 operations. + While we have protections in place against known attacks, it is expected that more attacks against SHA-1 will be found by future research. Paired with the ever-growing capability of hardware, it is only a matter of time before SHA-1 will be considered broken completely. We want to be prepared and will thus change the default hash algorithm to "sha256" for newly initialized repositories. + An important requirement for this change is that the ecosystem is ready to support the "sha256" object format. This includes popular Git libraries, applications and forges. + There is no plan to deprecate the "sha1" object format at this point in time. + Cf. <2f5de416-04ba-c23d-1e0b-83bb655829a7@zombino.com>, <20170223155046.e7nxivfwqqoprsqj@LykOS.localdomain>, . * The default storage format for references in newly created repositories will be changed from "files" to "reftable". The "reftable" format provides multiple advantages over the "files" format: + ** It is impossible to store two references that only differ in casing on case-insensitive filesystems with the "files" format. This issue is common on Windows and macOS platforms. As the "reftable" backend does not use filesystem paths to encode reference names this problem goes away. ** Similarly, macOS normalizes path names that contain unicode characters, which has the consequence that you cannot store two names with unicode characters that are encoded differently with the "files" backend. Again, this is not an issue with the "reftable" backend. ** Deleting references with the "files" backend requires Git to rewrite the complete "packed-refs" file. In large repositories with many references this file can easily be dozens of megabytes in size, in extreme cases it may be gigabytes. The "reftable" backend uses tombstone markers for deleted references and thus does not have to rewrite all of its data. ** Repository housekeeping with the "files" backend typically performs all-into-one repacks of references. This can be quite expensive, and consequently housekeeping is a tradeoff between the number of loose references that accumulate and slow down operations that read references, and compressing those loose references into the "packed-refs" file. The "reftable" backend uses geometric compaction after every write, which amortizes costs and ensures that the backend is always in a well-maintained state. ** Operations that write multiple references at once are not atomic with the "files" backend. Consequently, Git may see in-between states when it reads references while a reference transaction is in the process of being committed to disk. ** Writing many references at once is slow with the "files" backend because every reference is created as a separate file. The "reftable" backend significantly outperforms the "files" backend by multiple orders of magnitude. ** The reftable backend uses a binary format with prefix compression for reference names. As a result, the format uses less space compared to the "packed-refs" file. + Users that get immediate benefit from the "reftable" backend could continue to opt-in to the "reftable" format manually by setting the "init.defaultRefFormat" config. But defaults matter, and we think that overall users will have a better experience with less platform-specific quirks when they use the new backend by default. + A prerequisite for this change is that the ecosystem is ready to support the "reftable" format. Most importantly, alternative implementations of Git like JGit, libgit2 and Gitoxide need to support it. === Removals * Support for grafting commits has long been superseded by git-replace(1). Grafts are inferior to replacement refs: + ** Grafts are a local-only mechanism and cannot be shared across repositories. ** Grafts can lead to hard-to-diagnose problems when transferring objects between repositories. + The grafting mechanism has been marked as outdated since e650d0643b (docs: mark info/grafts as outdated, 2014-03-05) and will be removed. + Cf. <20140304174806.GA11561@sigill.intra.peff.net>. * The git-pack-redundant(1) command can be used to remove redundant pack files. The subcommand is unusably slow and the reason why nobody reports it as a performance bug is suspected to be the absence of users. We have nominated the command for removal and have started to emit a user-visible warning in c3b58472be (pack-redundant: gauge the usage before proposing its removal, 2020-08-25) whenever the command is executed. + So far there was a single complaint about somebody still using the command, but that complaint did not cause us to reverse course. On the contrary, we have doubled down on the deprecation and starting with 4406522b76 (pack-redundant: escalate deprecation warning to an error, 2023-03-23), the command dies unless the user passes the `--i-still-use-this` option. + There have not been any subsequent complaints, so this command will finally be removed. + Cf. , , <20230323204047.GA9290@coredump.intra.peff.net>, * Support for storing shorthands for remote URLs in "$GIT_COMMON_DIR/branches/" and "$GIT_COMMON_DIR/remotes/" has been long superseded by storing remotes in the repository configuration. + The mechanism has originally been introduced in f170e4b39d ([PATCH] fetch/pull: short-hand notation for remote repositories., 2005-07-16) and was superseded by 6687f8fea2 ([PATCH] Use .git/remote/origin, not .git/branches/origin., 2005-08-20), where we switched from ".git/branches/" to ".git/remotes/". That commit already mentions an upcoming deprecation of the ".git/branches/" directory, and starting with a1d4aa7424 (Add repository-layout document., 2005-09-01) we have also marked this layout as deprecated. Eventually we also started to migrate away from ".git/remotes/" in favor of config-based remotes, and we have marked the directory as legacy in 3d3d282146 (Documentation: Grammar correction, wording fixes and cleanup, 2011-08-23) + As our documentation mentions, these directories are unlikely to be used in modern repositories and most users aren't even aware of these mechanisms. They have been deprecated for almost 20 years and 14 years respectively, and we are not aware of any active users that have complained about this deprecation. Furthermore, the ".git/branches/" directory is nowadays misleadingly named and may cause confusion as "branches" are almost exclusively used in the context of references. + These features will be removed. * Support for "--stdin" option in the "name-rev" command was deprecated (and hidden from the documentation) in the Git 2.40 timeframe, in preference to its synonym "--annotate-stdin". Git 3.0 removes the support for "--stdin" altogether. * The git-whatchanged(1) command has outlived its usefulness more than 10 years ago, and takes more keystrokes to type than its rough equivalent `git log --raw`. We have nominated the command for removal, have changed the command to refuse to work unless the `--i-still-use-this` option is given, and asked the users to report when they do so. So far there hasn't been a single complaint. + The command will be removed. == Superseded features that will not be deprecated Some features have gained newer replacements that aim to improve the design in certain ways. The fact that there is a replacement does not automatically mean that the old way of doing things will eventually be removed. This section tracks those features with newer alternatives. * The features git-checkout(1) offers are covered by the pair of commands git-restore(1) and git-switch(1). Because the use of git-checkout(1) is still widespread, and it is not expected that this will change anytime soon, all three commands will stay. + This decision may get revisited in case we ever figure out that there are almost no users of any of the commands anymore. + Cf. , , <112b6568912a6de6672bf5592c3a718e@manjaro.org>. ./BreakingChanges.html000066400000000000000000001315701504515377700151410ustar00rootroot00000000000000 Upcoming breaking changes

The Git project aims to ensure backwards compatibility to the best extent possible. Minor releases will not break backwards compatibility unless there is a very strong reason to do so, like for example a security vulnerability.

Regardless of that, due to the age of the Git project, it is only natural to accumulate a backlog of backwards-incompatible changes that will eventually be required to keep the project aligned with a changing world. These changes fall into several categories:

  • Changes to long established defaults.

  • Concepts that have been replaced with a superior design.

  • Concepts, commands, configuration or options that have been lacking in major ways and that cannot be fixed and which will thus be removed without any replacement.

Explicitly not included in this list are fixes to minor bugs that may cause a change in user-visible behavior.

The Git project irregularly releases breaking versions that deliberately break backwards compatibility with older versions. This is done to ensure that Git remains relevant, safe and maintainable going forward. The release cadence of breaking versions is typically measured in multiple years. We had the following major breaking releases in the past:

  • Git 1.6.0, released in August 2008.

  • Git 2.0, released in May 2014.

We use <major>.<minor> release numbers these days, starting from Git 2.0. For future releases, our plan is to increment <major> in the release number when we make the next breaking release. Before Git 2.0, the release numbers were 1.<major>.<minor> with the intention to increment <major> for "usual" breaking releases, reserving the jump to Git 2.0 for really large backward-compatibility breaking changes.

The intent of this document is to track upcoming deprecations for future breaking releases. Furthermore, this document also tracks what will not be deprecated. This is done such that the outcome of discussions document both when the discussion favors deprecation, but also when it rejects a deprecation.

Items should have a clear summary of the reasons why we do or do not want to make the described change that can be easily understood without having to read the mailing list discussions. If there are alternatives to the changed feature, those alternatives should be pointed out to our users.

All items should be accompanied by references to relevant mailing list threads where the deprecation was discussed. These references use message-IDs, which can visited via

https://lore.kernel.org/git/$message_id/

to see the message and its surrounding discussion. Such a reference is there to make it easier for you to find how the project reached consensus on the described item back then.

This is a living document as the environment surrounding the project changes over time. If circumstances change, an earlier decision to deprecate or change something may need to be revisited from time to time. So do not take items on this list to mean "it is settled, do not waste our time bringing it up again".

Procedure

Discussing the desire to make breaking changes, declaring that breaking changes are made at a certain version boundary, and recording these decisions in this document, are necessary but not sufficient. Because such changes are expected to be numerous, and the design and implementation of them are expected to span over time, they have to be deployable trivially at such a version boundary, prepared over long time.

The breaking changes MUST be guarded with the a compile-time switch, WITH_BREAKING_CHANGES, to help this process. When built with it, the resulting Git binary together with its documentation would behave as if these breaking changes slated for the next big version boundary are already in effect. We also have a CI job to exercise the work-in-progress version of Git with these breaking changes.

Git 3.0

The following subsections document upcoming breaking changes for Git 3.0. There is no planned release date for this breaking version yet.

Proposed changes and removals only include items which are "ready" to be done. In other words, this is not supposed to be a wishlist of features that should be changed to or replaced in case the alternative was implemented already.

Changes

  • The default hash function for new repositories will be changed from "sha1" to "sha256". SHA-1 has been deprecated by NIST in 2011 and is nowadays recommended against in FIPS 140-2 and similar certifications. Furthermore, there are practical attacks on SHA-1 that weaken its cryptographic properties:

    • The SHAppening (2015). The first demonstration of a practical attack against SHA-1 with 2^57 operations.

    • SHAttered (2017). Generation of two valid PDF files with 2^63 operations.

    • Birthday-Near-Collision (2019). This attack allows for chosen prefix attacks with 2^68 operations.

    • Shambles (2020). This attack allows for chosen prefix attacks with 2^63 operations.

      While we have protections in place against known attacks, it is expected that more attacks against SHA-1 will be found by future research. Paired with the ever-growing capability of hardware, it is only a matter of time before SHA-1 will be considered broken completely. We want to be prepared and will thus change the default hash algorithm to "sha256" for newly initialized repositories.

      An important requirement for this change is that the ecosystem is ready to support the "sha256" object format. This includes popular Git libraries, applications and forges.

      There is no plan to deprecate the "sha1" object format at this point in time.

      Cf. <2f5de416-04ba-c23d-1e0b-83bb655829a7@zombino.com>, <20170223155046.e7nxivfwqqoprsqj@LykOS.localdomain>, <CA+EOSBncr=4a4d8n9xS4FNehyebpmX8JiUwCsXD47EQDE+DiUQ@mail.gmail.com>.

  • The default storage format for references in newly created repositories will be changed from "files" to "reftable". The "reftable" format provides multiple advantages over the "files" format:

    • It is impossible to store two references that only differ in casing on case-insensitive filesystems with the "files" format. This issue is common on Windows and macOS platforms. As the "reftable" backend does not use filesystem paths to encode reference names this problem goes away.

    • Similarly, macOS normalizes path names that contain unicode characters, which has the consequence that you cannot store two names with unicode characters that are encoded differently with the "files" backend. Again, this is not an issue with the "reftable" backend.

    • Deleting references with the "files" backend requires Git to rewrite the complete "packed-refs" file. In large repositories with many references this file can easily be dozens of megabytes in size, in extreme cases it may be gigabytes. The "reftable" backend uses tombstone markers for deleted references and thus does not have to rewrite all of its data.

    • Repository housekeeping with the "files" backend typically performs all-into-one repacks of references. This can be quite expensive, and consequently housekeeping is a tradeoff between the number of loose references that accumulate and slow down operations that read references, and compressing those loose references into the "packed-refs" file. The "reftable" backend uses geometric compaction after every write, which amortizes costs and ensures that the backend is always in a well-maintained state.

    • Operations that write multiple references at once are not atomic with the "files" backend. Consequently, Git may see in-between states when it reads references while a reference transaction is in the process of being committed to disk.

    • Writing many references at once is slow with the "files" backend because every reference is created as a separate file. The "reftable" backend significantly outperforms the "files" backend by multiple orders of magnitude.

    • The reftable backend uses a binary format with prefix compression for reference names. As a result, the format uses less space compared to the "packed-refs" file.

      Users that get immediate benefit from the "reftable" backend could continue to opt-in to the "reftable" format manually by setting the "init.defaultRefFormat" config. But defaults matter, and we think that overall users will have a better experience with less platform-specific quirks when they use the new backend by default.

      A prerequisite for this change is that the ecosystem is ready to support the "reftable" format. Most importantly, alternative implementations of Git like JGit, libgit2 and Gitoxide need to support it.

Removals

  • Support for grafting commits has long been superseded by git-replace(1). Grafts are inferior to replacement refs:

    • Grafts are a local-only mechanism and cannot be shared across repositories.

    • Grafts can lead to hard-to-diagnose problems when transferring objects between repositories.

      The grafting mechanism has been marked as outdated since e650d0643b (docs: mark info/grafts as outdated, 2014-03-05) and will be removed.

      Cf. <20140304174806.GA11561@sigill.intra.peff.net>.

  • The git-pack-redundant(1) command can be used to remove redundant pack files. The subcommand is unusably slow and the reason why nobody reports it as a performance bug is suspected to be the absence of users. We have nominated the command for removal and have started to emit a user-visible warning in c3b58472be (pack-redundant: gauge the usage before proposing its removal, 2020-08-25) whenever the command is executed.

    So far there was a single complaint about somebody still using the command, but that complaint did not cause us to reverse course. On the contrary, we have doubled down on the deprecation and starting with 4406522b76 (pack-redundant: escalate deprecation warning to an error, 2023-03-23), the command dies unless the user passes the --i-still-use-this option.

    There have not been any subsequent complaints, so this command will finally be removed.

    Cf. <xmqq1rjuz6n3.fsf_-_@gitster.c.googlers.com>, <CAKvOHKAFXQwt4D8yUCCkf_TQL79mYaJ=KAKhtpDNTvHJFuX1NA@mail.gmail.com>, <20230323204047.GA9290@coredump.intra.peff.net>,

  • Support for storing shorthands for remote URLs in "$GIT_COMMON_DIR/branches/" and "$GIT_COMMON_DIR/remotes/" has been long superseded by storing remotes in the repository configuration.

    The mechanism has originally been introduced in f170e4b39d ([PATCH] fetch/pull: short-hand notation for remote repositories., 2005-07-16) and was superseded by 6687f8fea2 ([PATCH] Use .git/remote/origin, not .git/branches/origin., 2005-08-20), where we switched from ".git/branches/" to ".git/remotes/". That commit already mentions an upcoming deprecation of the ".git/branches/" directory, and starting with a1d4aa7424 (Add repository-layout document., 2005-09-01) we have also marked this layout as deprecated. Eventually we also started to migrate away from ".git/remotes/" in favor of config-based remotes, and we have marked the directory as legacy in 3d3d282146 (Documentation: Grammar correction, wording fixes and cleanup, 2011-08-23)

    As our documentation mentions, these directories are unlikely to be used in modern repositories and most users aren’t even aware of these mechanisms. They have been deprecated for almost 20 years and 14 years respectively, and we are not aware of any active users that have complained about this deprecation. Furthermore, the ".git/branches/" directory is nowadays misleadingly named and may cause confusion as "branches" are almost exclusively used in the context of references.

    These features will be removed.

  • Support for "--stdin" option in the "name-rev" command was deprecated (and hidden from the documentation) in the Git 2.40 timeframe, in preference to its synonym "--annotate-stdin". Git 3.0 removes the support for "--stdin" altogether.

  • The git-whatchanged(1) command has outlived its usefulness more than 10 years ago, and takes more keystrokes to type than its rough equivalent git log --raw. We have nominated the command for removal, have changed the command to refuse to work unless the --i-still-use-this option is given, and asked the users to report when they do so. So far there hasn’t been a single complaint.

    The command will be removed.

Superseded features that will not be deprecated

Some features have gained newer replacements that aim to improve the design in certain ways. The fact that there is a replacement does not automatically mean that the old way of doing things will eventually be removed. This section tracks those features with newer alternatives.

  • The features git-checkout(1) offers are covered by the pair of commands git-restore(1) and git-switch(1). Because the use of git-checkout(1) is still widespread, and it is not expected that this will change anytime soon, all three commands will stay.

    This decision may get revisited in case we ever figure out that there are almost no users of any of the commands anymore.

    Cf. <xmqqttjazwwa.fsf@gitster.g>, <xmqqleeubork.fsf@gitster.g>, <112b6568912a6de6672bf5592c3a718e@manjaro.org>.

./DecisionMaking.adoc000066400000000000000000000074201504515377700147500ustar00rootroot00000000000000Decision-Making Process in the Git Project ========================================== Introduction ------------ This document describes the current decision-making process in the Git project. It is a descriptive rather than prescriptive doc; that is, we want to describe how things work in practice rather than explicitly recommending any particular process or changes to the current process. Here we document how the project makes decisions for discussions (with or without patches), in scale larger than an individual patch series (which is fully covered by the SubmittingPatches document). Larger Discussions (with patches) --------------------------------- As with discussions on an individual patch series, starting a larger-scale discussion often begins by sending a patch or series to the list. This might take the form of an initial design doc, with implementation following in later iterations of the series (for example, link:https://lore.kernel.org/git/0169ce6fb9ccafc089b74ae406db0d1a8ff8ac65.1688165272.git.steadmon@google.com/[adding unit tests] or link:https://lore.kernel.org/git/20200420235310.94493-1-emilyshaffer@google.com/[config-based hooks]), or it might include a full implementation from the beginning. In either case, discussion progresses the same way for an individual patch series, until consensus is reached or the topic is dropped. Larger Discussions (without patches) ------------------------------------ Occasionally, larger discussions might occur without an associated patch series. These may be very large-scale technical decisions that are beyond the scope of even a single large patch series, or they may be more open-ended, policy-oriented discussions (examples: link:https://lore.kernel.org/git/ZZ77NQkSuiRxRDwt@nand.local/[introducing Rust] or link:https://lore.kernel.org/git/YHofmWcIAidkvJiD@google.com/[improving submodule UX]). In either case, discussion progresses as described above for general patch series. For larger discussions without a patch series or other concrete implementation, it may be hard to judge when consensus has been reached, as there are not any official guidelines. If discussion stalls at this point, it may be helpful to restart discussion with an RFC patch series (such as a partial, unfinished implementation or proof of concept) that can be more easily debated. When consensus is reached that it is a good idea, the original proposer is expected to coordinate the effort to make it happen, with help from others who were involved in the discussion, as needed. For decisions that require code changes, it is often the case that the original proposer will follow up with a patch series, although it is also common for other interested parties to provide an implementation (or parts of the implementation, for very large changes). For non-technical decisions such as community norms or processes, it is up to the community as a whole to implement and sustain agreed-upon changes. The project leadership committee (PLC) may help the implementation of policy decisions. Other Discussion Venues ----------------------- Occasionally decision proposals are presented off-list, e.g. at the semi-regular Contributors' Summit. While higher-bandwidth face-to-face discussion is often useful for quickly reaching consensus among attendees, generally we expect to summarize the discussion in notes that can later be presented on-list. For an example, see the thread link:https://lore.kernel.org/git/AC2EB721-2979-43FD-922D-C5076A57F24B@jramsay.com.au/[Notes from Git Contributor Summit, Los Angeles (April 5, 2020)] by James Ramsay. We prefer that "official" discussion happens on the list so that the full community has opportunity to engage in discussion. This also means that the mailing list archives contain a more-or-less complete history of project discussions and decisions. ./DecisionMaking.html000066400000000000000000001031771504515377700150140ustar00rootroot00000000000000 Decision-Making Process in the Git Project

Introduction

This document describes the current decision-making process in the Git project. It is a descriptive rather than prescriptive doc; that is, we want to describe how things work in practice rather than explicitly recommending any particular process or changes to the current process.

Here we document how the project makes decisions for discussions (with or without patches), in scale larger than an individual patch series (which is fully covered by the SubmittingPatches document).

Larger Discussions (with patches)

As with discussions on an individual patch series, starting a larger-scale discussion often begins by sending a patch or series to the list. This might take the form of an initial design doc, with implementation following in later iterations of the series (for example, adding unit tests or config-based hooks), or it might include a full implementation from the beginning. In either case, discussion progresses the same way for an individual patch series, until consensus is reached or the topic is dropped.

Larger Discussions (without patches)

Occasionally, larger discussions might occur without an associated patch series. These may be very large-scale technical decisions that are beyond the scope of even a single large patch series, or they may be more open-ended, policy-oriented discussions (examples: introducing Rust or improving submodule UX). In either case, discussion progresses as described above for general patch series.

For larger discussions without a patch series or other concrete implementation, it may be hard to judge when consensus has been reached, as there are not any official guidelines. If discussion stalls at this point, it may be helpful to restart discussion with an RFC patch series (such as a partial, unfinished implementation or proof of concept) that can be more easily debated.

When consensus is reached that it is a good idea, the original proposer is expected to coordinate the effort to make it happen, with help from others who were involved in the discussion, as needed.

For decisions that require code changes, it is often the case that the original proposer will follow up with a patch series, although it is also common for other interested parties to provide an implementation (or parts of the implementation, for very large changes).

For non-technical decisions such as community norms or processes, it is up to the community as a whole to implement and sustain agreed-upon changes. The project leadership committee (PLC) may help the implementation of policy decisions.

Other Discussion Venues

Occasionally decision proposals are presented off-list, e.g. at the semi-regular Contributors' Summit. While higher-bandwidth face-to-face discussion is often useful for quickly reaching consensus among attendees, generally we expect to summarize the discussion in notes that can later be presented on-list. For an example, see the thread Notes from Git Contributor Summit, Los Angeles (April 5, 2020) by James Ramsay.

We prefer that "official" discussion happens on the list so that the full community has opportunity to engage in discussion. This also means that the mailing list archives contain a more-or-less complete history of project discussions and decisions.

./MyFirstContribution.adoc000066400000000000000000001543241504515377700160670ustar00rootroot00000000000000My First Contribution to the Git Project ======================================== :sectanchors: [[summary]] == Summary This is a tutorial demonstrating the end-to-end workflow of creating a change to the Git tree, sending it for review, and making changes based on comments. [[prerequisites]] === Prerequisites This tutorial assumes you're already fairly familiar with using Git to manage source code. The Git workflow steps will largely remain unexplained. [[related-reading]] === Related Reading This tutorial aims to summarize the following documents, but the reader may find useful additional context: - `Documentation/SubmittingPatches` - `Documentation/howto/new-command.adoc` [[getting-help]] === Getting Help If you get stuck, you can seek help in the following places. ==== git@vger.kernel.org This is the main Git project mailing list where code reviews, version announcements, design discussions, and more take place. Those interested in contributing are welcome to post questions here. The Git list requires plain-text-only emails and prefers inline and bottom-posting when replying to mail; you will be CC'd in all replies to you. Optionally, you can subscribe to the list by sending an email to (see https://subspace.kernel.org/subscribing.html for details). The https://lore.kernel.org/git[archive] of this mailing list is available to view in a browser. ==== https://web.libera.chat/#git-devel[#git-devel] on Libera Chat This IRC channel is for conversations between Git contributors. If someone is currently online and knows the answer to your question, you can receive help in real time. Otherwise, you can read the https://colabti.org/irclogger/irclogger_logs/git-devel[scrollback] to see whether someone answered you. IRC does not allow offline private messaging, so if you try to private message someone and then log out of IRC, they cannot respond to you. It's better to ask your questions in the channel so that you can be answered if you disconnect and so that others can learn from the conversation. [[getting-started]] == Getting Started [[cloning]] === Clone the Git Repository Git is mirrored in a number of locations. Clone the repository from one of them; https://git-scm.com/downloads suggests one of the best places to clone from is the mirror on GitHub. ---- $ git clone https://github.com/git/git git $ cd git ---- [[dependencies]] === Installing Dependencies To build Git from source, you need to have a handful of dependencies installed on your system. For a hint of what's needed, you can take a look at `INSTALL`, paying close attention to the section about Git's dependencies on external programs and libraries. That document mentions a way to "test-drive" our freshly built Git without installing; that's the method we'll be using in this tutorial. Make sure that your environment has everything you need by building your brand new clone of Git from the above step: ---- $ make ---- NOTE: The Git build is parallelizable. `-j#` is not included above but you can use it as you prefer, here and elsewhere. [[identify-problem]] === Identify Problem to Solve //// Use + to indicate fixed-width here; couldn't get ` to work nicely with the quotes around "Pony Saying 'Um, Hello'". //// In this tutorial, we will add a new command, +git psuh+, short for ``Pony Saying `Um, Hello''' - a feature which has gone unimplemented despite a high frequency of invocation during users' typical daily workflow. (We've seen some other effort in this space with the implementation of popular commands such as `sl`.) [[setup-workspace]] === Set Up Your Workspace Let's start by making a development branch to work on our changes. Per `Documentation/SubmittingPatches`, since a brand new command is a new feature, it's fine to base your work on `master`. However, in the future for bugfixes, etc., you should check that document and base it on the appropriate branch. For the purposes of this document, we will base all our work on the `master` branch of the upstream project. Create the `psuh` branch you will use for development like so: ---- $ git checkout -b psuh origin/master ---- We'll make a number of commits here in order to demonstrate how to send a topic with multiple patches up for review simultaneously. [[code-it-up]] == Code It Up! NOTE: A reference implementation can be found at https://github.com/nasamuffin/git/tree/psuh. [[add-new-command]] === Adding a New Command Lots of the subcommands are written as builtins, which means they are implemented in C and compiled into the main `git` executable. Implementing the very simple `psuh` command as a built-in will demonstrate the structure of the codebase, the internal API, and the process of working together as a contributor with the reviewers and maintainer to integrate this change into the system. Built-in subcommands are typically implemented in a function named "cmd_" followed by the name of the subcommand, in a source file named after the subcommand and contained within `builtin/`. So it makes sense to implement your command in `builtin/psuh.c`. Create that file, and within it, write the entry point for your command in a function matching the style and signature: ---- int cmd_psuh(int argc UNUSED, const char **argv UNUSED, const char *prefix UNUSED, struct repository *repo UNUSED) ---- A few things to note: * A subcommand implementation takes its command line arguments in `int argc` + `const char **argv`, like `main()` would. * It also takes two extra parameters, `prefix` and `repo`. What they mean will not be discussed until much later. * Because this first example will not use any of the parameters, your compiler will give warnings on unused parameters. As the list of these four parameters is mandated by the API to add new built-in commands, you cannot omit them. Instead, you add `UNUSED` to each of them to tell the compiler that you *know* you are not (yet) using it. We'll also need to add the declaration of psuh; open up `builtin.h`, find the declaration for `cmd_pull`, and add a new line for `psuh` immediately before it, in order to keep the declarations alphabetically sorted: ---- int cmd_psuh(int argc, const char **argv, const char *prefix, struct repository *repo); ---- Be sure to `#include "builtin.h"` in your `psuh.c`. You'll also need to `#include "gettext.h"` to use functions related to printing output text. Go ahead and add some throwaway printf to the `cmd_psuh` function. This is a decent starting point as we can now add build rules and register the command. NOTE: Your throwaway text, as well as much of the text you will be adding over the course of this tutorial, is user-facing. That means it needs to be localizable. Take a look at `po/README` under "Marking strings for translation". Throughout the tutorial, we will mark strings for translation as necessary; you should also do so when writing your user-facing commands in the future. ---- int cmd_psuh(int argc UNUSED, const char **argv UNUSED, const char *prefix UNUSED, struct repository *repo UNUSED) { printf(_("Pony saying hello goes here.\n")); return 0; } ---- Let's try to build it. Open `Makefile`, find where `builtin/pull.o` is added to `BUILTIN_OBJS`, and add `builtin/psuh.o` in the same way next to it in alphabetical order. Once you've done so, move to the top-level directory and build simply with `make`. Also add the `DEVELOPER=1` variable to turn on some additional warnings: ---- $ echo DEVELOPER=1 >config.mak $ make ---- NOTE: When you are developing the Git project, it's preferred that you use the `DEVELOPER` flag; if there's some reason it doesn't work for you, you can turn it off, but it's a good idea to mention the problem to the mailing list. Great, now your new command builds happily on its own. But nobody invokes it. Let's change that. The list of commands lives in `git.c`. We can register a new command by adding a `cmd_struct` to the `commands[]` array. `struct cmd_struct` takes a string with the command name, a function pointer to the command implementation, and a setup option flag. For now, let's keep mimicking `push`. Find the line where `cmd_push` is registered, copy it, and modify it for `cmd_psuh`, placing the new line in alphabetical order (immediately before `cmd_pull`). The options are documented in `builtin.h` under "Adding a new built-in." Since we hope to print some data about the user's current workspace context later, we need a Git directory, so choose `RUN_SETUP` as your only option. Go ahead and build again. You should see a clean build, so let's kick the tires and see if it works. There's a binary you can use to test with in the `bin-wrappers` directory. ---- $ ./bin-wrappers/git psuh ---- Check it out! You've got a command! Nice work! Let's commit this. `git status` reveals modified `Makefile`, `builtin.h`, and `git.c` as well as untracked `builtin/psuh.c` and `git-psuh`. First, let's take care of the binary, which should be ignored. Open `.gitignore` in your editor, find `/git-pull`, and add an entry for your new command in alphabetical order: ---- ... /git-prune-packed /git-psuh /git-pull /git-push /git-quiltimport /git-range-diff ... ---- Checking `git status` again should show that `git-psuh` has been removed from the untracked list and `.gitignore` has been added to the modified list. Now we can stage and commit: ---- $ git add Makefile builtin.h builtin/psuh.c git.c .gitignore $ git commit -s ---- You will be presented with your editor in order to write a commit message. Start the commit with a 50-column or less subject line, including the name of the component you're working on, followed by a blank line (always required) and then the body of your commit message, which should provide the bulk of the context. Remember to be explicit and provide the "Why" of your change, especially if it couldn't easily be understood from your diff. When editing your commit message, don't remove the `Signed-off-by` trailer which was added by `-s` above. ---- psuh: add a built-in by popular demand Internal metrics indicate this is a command many users expect to be present. So here's an implementation to help drive customer satisfaction and engagement: a pony which doubtfully greets the user, or, a Pony Saying "Um, Hello" (PSUH). This commit message is intentionally formatted to 72 columns per line, starts with a single line as "commit message subject" that is written as if to command the codebase to do something (add this, teach a command that). The body of the message is designed to add information about the commit that is not readily deduced from reading the associated diff, such as answering the question "why?". Signed-off-by: A U Thor ---- Go ahead and inspect your new commit with `git show`. "psuh:" indicates you have modified mainly the `psuh` command. The subject line gives readers an idea of what you've changed. The sign-off line (`-s`) indicates that you agree to the Developer's Certificate of Origin 1.1 (see the `Documentation/SubmittingPatches` +++[[dco]]+++ header). For the remainder of the tutorial, the subject line only will be listed for the sake of brevity. However, fully-fleshed example commit messages are available on the reference implementation linked at the top of this document. [[implementation]] === Implementation It's probably useful to do at least something besides printing out a string. Let's start by having a look at everything we get. Modify your `cmd_psuh` implementation to dump the args you're passed, keeping existing `printf()` calls in place; because the args are now used, remove the `UNUSED` macro from them: ---- int i; ... printf(Q_("Your args (there is %d):\n", "Your args (there are %d):\n", argc), argc); for (i = 0; i < argc; i++) printf("%d: %s\n", i, argv[i]); printf(_("Your current working directory:\n%s%s\n"), prefix ? "/" : "", prefix ? prefix : ""); ---- Build and try it. As you may expect, there's pretty much just whatever we give on the command line, including the name of our command. (If `prefix` is empty for you, try `cd Documentation/ && ../bin-wrappers/git psuh`). That's not so helpful. So what other context can we get? Add a line to `#include "config.h"` and `#include "repository.h"`. Then, add the following bits to the function body: function body: ---- const char *cfg_name; ... repo_config(repo, git_default_config, NULL); if (repo_config_get_string_tmp(repo, "user.name", &cfg_name)) printf(_("No name is found in config\n")); else printf(_("Your name: %s\n"), cfg_name); ---- `repo_config()` will grab the configuration from config files known to Git and apply standard precedence rules. `repo_config_get_string_tmp()` will look up a specific key ("user.name") and give you the value. There are a number of single-key lookup functions like this one; you can see them all (and more info about how to use `repo_config()`) in `Documentation/technical/api-config.adoc`. You should see that the name printed matches the one you see when you run: ---- $ git config --get user.name ---- Great! Now we know how to check for values in the Git config. Let's commit this too, so we don't lose our progress. ---- $ git add builtin/psuh.c $ git commit -sm "psuh: show parameters & config opts" ---- NOTE: Again, the above is for sake of brevity in this tutorial. In a real change you should not use `-m` but instead use the editor to write a meaningful message. Still, it'd be nice to know what the user's working context is like. Let's see if we can print the name of the user's current branch. We can mimic the `git status` implementation; the printer is located in `wt-status.c` and we can see that the branch is held in a `struct wt_status`. `wt_status_print()` gets invoked by `cmd_status()` in `builtin/commit.c`. Looking at that implementation we see the status config being populated like so: ---- status_init_config(&s, git_status_config); ---- But as we drill down, we can find that `status_init_config()` wraps a call to `repo_config()`. Let's modify the code we wrote in the previous commit. Be sure to include the header to allow you to use `struct wt_status`: ---- #include "wt-status.h" ---- Then modify your `cmd_psuh` implementation to declare your `struct wt_status`, prepare it, and print its contents: ---- struct wt_status status; ... wt_status_prepare(repo, &status); repo_config(repo, git_default_config, &status); ... printf(_("Your current branch: %s\n"), status.branch); ---- Run it again. Check it out - here's the (verbose) name of your current branch! Let's commit this as well. ---- $ git add builtin/psuh.c $ git commit -sm "psuh: print the current branch" ---- Now let's see if we can get some info about a specific commit. Luckily, there are some helpers for us here. `commit.h` has a function called `lookup_commit_reference_by_name` to which we can simply provide a hardcoded string; `pretty.h` has an extremely handy `pp_commit_easy()` call which doesn't require a full format object to be passed. Add the following includes: ---- #include "commit.h" #include "pretty.h" ---- Then, add the following lines within your implementation of `cmd_psuh()` near the declarations and the logic, respectively. ---- struct commit *c = NULL; struct strbuf commitline = STRBUF_INIT; ... c = lookup_commit_reference_by_name("origin/master"); if (c != NULL) { pp_commit_easy(CMIT_FMT_ONELINE, c, &commitline); printf(_("Current commit: %s\n"), commitline.buf); } ---- The `struct strbuf` provides some safety belts to your basic `char*`, one of which is a length member to prevent buffer overruns. It needs to be initialized nicely with `STRBUF_INIT`. Keep it in mind when you need to pass around `char*`. `lookup_commit_reference_by_name` resolves the name you pass it, so you can play with the value there and see what kind of things you can come up with. `pp_commit_easy` is a convenience wrapper in `pretty.h` that takes a single format enum shorthand, rather than an entire format struct. It then pretty-prints the commit according to that shorthand. These are similar to the formats available with `--pretty=FOO` in many Git commands. Build it and run, and if you're using the same name in the example, you should see the subject line of the most recent commit in `origin/master` that you know about. Neat! Let's commit that as well. ---- $ git add builtin/psuh.c $ git commit -sm "psuh: display the top of origin/master" ---- [[add-documentation]] === Adding Documentation Awesome! You've got a fantastic new command that you're ready to share with the community. But hang on just a minute - this isn't very user-friendly. Run the following: ---- $ ./bin-wrappers/git help psuh ---- Your new command is undocumented! Let's fix that. Take a look at `Documentation/git-*.adoc`. These are the manpages for the subcommands that Git knows about. You can open these up and take a look to get acquainted with the format, but then go ahead and make a new file `Documentation/git-psuh.adoc`. Like with most of the documentation in the Git project, help pages are written with AsciiDoc (see CodingGuidelines, "Writing Documentation" section). Use the following template to fill out your own manpage: // Surprisingly difficult to embed AsciiDoc source within AsciiDoc. [listing] .... git-psuh(1) =========== NAME ---- git-psuh - Delight users' typo with a shy horse SYNOPSIS -------- [verse] 'git-psuh [...]' DESCRIPTION ----------- ... OPTIONS[[OPTIONS]] ------------------ ... OUTPUT ------ ... GIT --- Part of the linkgit:git[1] suite .... The most important pieces of this to note are the file header, underlined by =, the NAME section, and the SYNOPSIS, which would normally contain the grammar if your command took arguments. Try to use well-established manpage headers so your documentation is consistent with other Git and UNIX manpages; this makes life easier for your user, who can skip to the section they know contains the information they need. NOTE: Before trying to build the docs, make sure you have the package `asciidoc` installed. Now that you've written your manpage, you'll need to build it explicitly. We convert your AsciiDoc to troff which is man-readable like so: ---- $ make all doc $ man Documentation/git-psuh.1 ---- or ---- $ make -C Documentation/ git-psuh.1 $ man Documentation/git-psuh.1 ---- While this isn't as satisfying as running through `git help`, you can at least check that your help page looks right. You can also check that the documentation coverage is good (that is, the project sees that your command has been implemented as well as documented) by running `make check-docs` from the top-level. Go ahead and commit your new documentation change. [[add-usage]] === Adding Usage Text Try and run `./bin-wrappers/git psuh -h`. Your command should crash at the end. That's because `-h` is a special case which your command should handle by printing usage. Take a look at `Documentation/technical/api-parse-options.adoc`. This is a handy tool for pulling out options you need to be able to handle, and it takes a usage string. In order to use it, we'll need to prepare a NULL-terminated array of usage strings and a `builtin_psuh_options` array. Add a line to `#include "parse-options.h"`. At global scope, add your array of usage strings: ---- static const char * const psuh_usage[] = { N_("git psuh [...]"), NULL, }; ---- Then, within your `cmd_psuh()` implementation, we can declare and populate our `option` struct. Ours is pretty boring but you can add more to it if you want to explore `parse_options()` in more detail: ---- struct option options[] = { OPT_END() }; ---- Finally, before you print your args and prefix, add the call to `parse-options()`: ---- argc = parse_options(argc, argv, prefix, options, psuh_usage, 0); ---- This call will modify your `argv` parameter. It will strip the options you specified in `options` from `argv` and the locations pointed to from `options` entries will be updated. Be sure to replace your `argc` with the result from `parse_options()`, or you will be confused if you try to parse `argv` later. It's worth noting the special argument `--`. As you may be aware, many Unix commands use `--` to indicate "end of named parameters" - all parameters after the `--` are interpreted merely as positional arguments. (This can be handy if you want to pass as a parameter something which would usually be interpreted as a flag.) `parse_options()` will terminate parsing when it reaches `--` and give you the rest of the options afterwards, untouched. Now that you have a usage hint, you can teach Git how to show it in the general command list shown by `git help git` or `git help -a`, which is generated from `command-list.txt`. Find the line for 'git-pull' so you can add your 'git-psuh' line above it in alphabetical order. Now, we can add some attributes about the command which impacts where it shows up in the aforementioned help commands. The top of `command-list.txt` shares some information about what each attribute means; in those help pages, the commands are sorted according to these attributes. `git psuh` is user-facing, or porcelain - so we will mark it as "mainporcelain". For "mainporcelain" commands, the comments at the top of `command-list.txt` indicate we can also optionally add an attribute from another list; since `git psuh` shows some information about the user's workspace but doesn't modify anything, let's mark it as "info". Make sure to keep your attributes in the same style as the rest of `command-list.txt` using spaces to align and delineate them: ---- git-prune-packed plumbingmanipulators git-psuh mainporcelain info git-pull mainporcelain remote git-push mainporcelain remote ---- Build again. Now, when you run with `-h`, you should see your usage printed and your command terminated before anything else interesting happens. Great! Go ahead and commit this one, too. [[testing]] == Testing It's important to test your code - even for a little toy command like this one. Moreover, your patch won't be accepted into the Git tree without tests. Your tests should: * Illustrate the current behavior of the feature * Prove the current behavior matches the expected behavior * Ensure the externally-visible behavior isn't broken in later changes So let's write some tests. Related reading: `t/README` [[overview-test-structure]] === Overview of Testing Structure The tests in Git live in `t/` and are named with a 4-digit decimal number using the schema shown in the Naming Tests section of `t/README`. [[write-new-test]] === Writing Your Test Since this a toy command, let's go ahead and name the test with t9999. However, as many of the family/subcmd combinations are full, best practice seems to be to find a command close enough to the one you've added and share its naming space. Create a new file `t/t9999-psuh-tutorial.sh`. Begin with the header as so (see "Writing Tests" and "Source 'test-lib.sh'" in `t/README`): ---- #!/bin/sh test_description='git-psuh test This test runs git-psuh and makes sure it does not crash.' . ./test-lib.sh ---- Tests are framed inside of a `test_expect_success` in order to output TAP formatted results. Let's make sure that `git psuh` doesn't exit poorly and does mention the right animal somewhere: ---- test_expect_success 'runs correctly with no args and good output' ' git psuh >actual && grep Pony actual ' ---- Indicate that you've run everything you wanted by adding the following at the bottom of your script: ---- test_done ---- Make sure you mark your test script executable: ---- $ chmod +x t/t9999-psuh-tutorial.sh ---- You can get an idea of whether you created your new test script successfully by running `make -C t test-lint`, which will check for things like test number uniqueness, executable bit, and so on. [[local-test]] === Running Locally Let's try and run locally: ---- $ make $ cd t/ && prove t9999-psuh-tutorial.sh ---- You can run the full test suite and ensure `git-psuh` didn't break anything: ---- $ cd t/ $ prove -j$(nproc) --shuffle t[0-9]*.sh ---- NOTE: You can also do this with `make test` or use any testing harness which can speak TAP. `prove` can run concurrently. `shuffle` randomizes the order the tests are run in, which makes them resilient against unwanted inter-test dependencies. `prove` also makes the output nicer. Go ahead and commit this change, as well. [[ready-to-share]] == Getting Ready to Share: Anatomy of a Patch Series You may have noticed already that the Git project performs its code reviews via emailed patches, which are then applied by the maintainer when they are ready and approved by the community. The Git project does not accept contributions from pull requests, and the patches emailed for review need to be formatted a specific way. :patch-series: https://lore.kernel.org/git/pull.1218.git.git.1645209647.gitgitgadget@gmail.com/ :lore: https://lore.kernel.org/git/ Before taking a look at how to convert your commits into emailed patches, let's analyze what the end result, a "patch series", looks like. Here is an {patch-series}[example] of the summary view for a patch series on the web interface of the {lore}[Git mailing list archive]: ---- 2022-02-18 18:40 [PATCH 0/3] libify reflog John Cai via GitGitGadget 2022-02-18 18:40 ` [PATCH 1/3] reflog: libify delete reflog function and helpers John Cai via GitGitGadget 2022-02-18 19:10 ` Ævar Arnfjörð Bjarmason [this message] 2022-02-18 19:39 ` Taylor Blau 2022-02-18 19:48 ` Ævar Arnfjörð Bjarmason 2022-02-18 19:35 ` Taylor Blau 2022-02-21 1:43 ` John Cai 2022-02-21 1:50 ` Taylor Blau 2022-02-23 19:50 ` John Cai 2022-02-18 20:00 ` // other replies elided 2022-02-18 18:40 ` [PATCH 2/3] reflog: call reflog_delete from reflog.c John Cai via GitGitGadget 2022-02-18 19:15 ` Ævar Arnfjörð Bjarmason 2022-02-18 20:26 ` Junio C Hamano 2022-02-18 18:40 ` [PATCH 3/3] stash: call reflog_delete from reflog.c John Cai via GitGitGadget 2022-02-18 19:20 ` Ævar Arnfjörð Bjarmason 2022-02-19 0:21 ` Taylor Blau 2022-02-22 2:36 ` John Cai 2022-02-22 10:51 ` Ævar Arnfjörð Bjarmason 2022-02-18 19:29 ` [PATCH 0/3] libify reflog Ævar Arnfjörð Bjarmason 2022-02-22 18:30 ` [PATCH v2 0/3] libify reflog John Cai via GitGitGadget 2022-02-22 18:30 ` [PATCH v2 1/3] stash: add test to ensure reflog --rewrite --updatref behavior John Cai via GitGitGadget 2022-02-23 8:54 ` Ævar Arnfjörð Bjarmason 2022-02-23 21:27 ` Junio C Hamano // continued ---- We can note a few things: - Each commit is sent as a separate email, with the commit message title as subject, prefixed with "[PATCH _i_/_n_]" for the _i_-th commit of an _n_-commit series. - Each patch is sent as a reply to an introductory email called the _cover letter_ of the series, prefixed "[PATCH 0/_n_]". - Subsequent iterations of the patch series are labelled "PATCH v2", "PATCH v3", etc. in place of "PATCH". For example, "[PATCH v2 1/3]" would be the first of three patches in the second iteration. Each iteration is sent with a new cover letter (like "[PATCH v2 0/3]" above), itself a reply to the cover letter of the previous iteration (more on that below). NOTE: A single-patch topic is sent with "[PATCH]", "[PATCH v2]", etc. without _i_/_n_ numbering (in the above thread overview, no single-patch topic appears, though). [[cover-letter]] === The cover letter In addition to an email per patch, the Git community also expects your patches to come with a cover letter. This is an important component of change submission as it explains to the community from a high level what you're trying to do, and why, in a way that's more apparent than just looking at your patches. The title of your cover letter should be something which succinctly covers the purpose of your entire topic branch. It's often in the imperative mood, just like our commit message titles. Here is how we'll title our series: --- Add the 'psuh' command --- The body of the cover letter is used to give additional context to reviewers. Be sure to explain anything your patches don't make clear on their own, but remember that since the cover letter is not recorded in the commit history, anything that might be useful to future readers of the repository's history should also be in your commit messages. Here's an example body for `psuh`: ---- Our internal metrics indicate widespread interest in the command git-psuh - that is, many users are trying to use it, but finding it is unavailable, using some unknown workaround instead. The following handful of patches add the psuh command and implement some handy features on top of it. This patchset is part of the MyFirstContribution tutorial and should not be merged. ---- At this point the tutorial diverges, in order to demonstrate two different methods of formatting your patchset and getting it reviewed. The first method to be covered is GitGitGadget, which is useful for those already familiar with GitHub's common pull request workflow. This method requires a GitHub account. The second method to be covered is `git send-email`, which can give slightly more fine-grained control over the emails to be sent. This method requires some setup which can change depending on your system and will not be covered in this tutorial. Regardless of which method you choose, your engagement with reviewers will be the same; the review process will be covered after the sections on GitGitGadget and `git send-email`. [[howto-ggg]] == Sending Patches via GitGitGadget One option for sending patches is to follow a typical pull request workflow and send your patches out via GitGitGadget. GitGitGadget is a tool created by Johannes Schindelin to make life as a Git contributor easier for those used to the GitHub PR workflow. It allows contributors to open pull requests against its mirror of the Git project, and does some magic to turn the PR into a set of emails and send them out for you. It also runs the Git continuous integration suite for you. It's documented at https://gitgitgadget.github.io/. [[create-fork]] === Forking `git/git` on GitHub Before you can send your patch off to be reviewed using GitGitGadget, you will need to fork the Git project and upload your changes. First thing - make sure you have a GitHub account. Head to the https://github.com/git/git[GitHub mirror] and look for the Fork button. Place your fork wherever you deem appropriate and create it. [[upload-to-fork]] === Uploading to Your Own Fork To upload your branch to your own fork, you'll need to add the new fork as a remote. You can use `git remote -v` to show the remotes you have added already. From your new fork's page on GitHub, you can press "Clone or download" to get the URL; then you need to run the following to add, replacing your own URL and remote name for the examples provided: ---- $ git remote add remotename git@github.com:remotename/git.git ---- or to use the HTTPS URL: ---- $ git remote add remotename https://github.com/remotename/git/.git ---- Run `git remote -v` again and you should see the new remote showing up. `git fetch remotename` (with the real name of your remote replaced) in order to get ready to push. Next, double-check that you've been doing all your development in a new branch by running `git branch`. If you didn't, now is a good time to move your new commits to their own branch. As mentioned briefly at the beginning of this document, we are basing our work on `master`, so go ahead and update as shown below, or using your preferred workflow. ---- $ git checkout master $ git pull -r $ git rebase master psuh ---- Finally, you're ready to push your new topic branch! (Due to our branch and command name choices, be careful when you type the command below.) ---- $ git push remotename psuh ---- Now you should be able to go and check out your newly created branch on GitHub. [[send-pr-ggg]] === Sending a PR to GitGitGadget In order to have your code tested and formatted for review, you need to start by opening a Pull Request against `gitgitgadget/git`. Head to https://github.com/gitgitgadget/git and open a PR either with the "New pull request" button or the convenient "Compare & pull request" button that may appear with the name of your newly pushed branch. Review the PR's title and description, as they're used by GitGitGadget respectively as the subject and body of the cover letter for your change. Refer to <> above for advice on how to title your submission and what content to include in the description. NOTE: For single-patch contributions, your commit message should already be meaningful and explain at a high level the purpose (what is happening and why) of your patch, so you usually do not need any additional context. In that case, remove the PR description that GitHub automatically generates from your commit message (your PR description should be empty). If you do need to supply even more context, you can do so in that space and it will be appended to the email that GitGitGadget will send, between the three-dash line and the diffstat (see <> for how this looks once submitted). When you're happy, submit your pull request. [[run-ci-ggg]] === Running CI and Getting Ready to Send If it's your first time using GitGitGadget (which is likely, as you're using this tutorial) then someone will need to give you permission to use the tool. As mentioned in the GitGitGadget documentation, you just need someone who already uses it to comment on your PR with `/allow `. GitGitGadget will automatically run your PRs through the CI even without the permission given but you will not be able to `/submit` your changes until someone allows you to use the tool. NOTE: You can typically find someone who can `/allow` you on GitGitGadget by either examining recent pull requests where someone has been granted `/allow` (https://github.com/gitgitgadget/git/pulls?utf8=%E2%9C%93&q=is%3Apr+is%3Aopen+%22%2Fallow%22[Search: is:pr is:open "/allow"]), in which case both the author and the person who granted the `/allow` can now `/allow` you, or by inquiring on the https://web.libera.chat/#git-devel[#git-devel] IRC channel on Libera Chat linking your pull request and asking for someone to `/allow` you. If the CI fails, you can update your changes with `git rebase -i` and push your branch again: ---- $ git push -f remotename psuh ---- In fact, you should continue to make changes this way up until the point when your patch is accepted into `next`. //// TODO https://github.com/gitgitgadget/gitgitgadget/issues/83 It'd be nice to be able to verify that the patch looks good before sending it to everyone on Git mailing list. [[check-work-ggg]] === Check Your Work //// [[send-mail-ggg]] === Sending Your Patches Now that your CI is passing and someone has granted you permission to use GitGitGadget with the `/allow` command, sending out for review is as simple as commenting on your PR with `/submit`. [[responding-ggg]] === Updating With Comments Skip ahead to <> for information on how to reply to review comments you will receive on the mailing list. Once you have your branch again in the shape you want following all review comments, you can submit again: ---- $ git push -f remotename psuh ---- Next, go look at your pull request against GitGitGadget; you should see the CI has been kicked off again. Now while the CI is running is a good time for you to modify your description at the top of the pull request thread; it will be used again as the cover letter. You should use this space to describe what has changed since your previous version, so that your reviewers have some idea of what they're looking at. When the CI is done running, you can comment once more with `/submit` - GitGitGadget will automatically add a v2 mark to your changes. [[howto-git-send-email]] == Sending Patches with `git send-email` If you don't want to use GitGitGadget, you can also use Git itself to mail your patches. Some benefits of using Git this way include finer grained control of subject line (for example, being able to use the tag [RFC PATCH] in the subject) and being able to send a ``dry run'' mail to yourself to ensure it all looks good before going out to the list. [[setup-git-send-email]] === Prerequisite: Setting Up `git send-email` Configuration for `send-email` can vary based on your operating system and email provider, and so will not be covered in this tutorial, beyond stating that in many distributions of Linux, `git-send-email` is not packaged alongside the typical `git` install. You may need to install this additional package; there are a number of resources online to help you do so. You will also need to determine the right way to configure it to use your SMTP server; again, as this configuration can change significantly based on your system and email setup, it is out of scope for the context of this tutorial. [[format-patch]] === Preparing Initial Patchset Sending emails with Git is a two-part process; before you can prepare the emails themselves, you'll need to prepare the patches. Luckily, this is pretty simple: ---- $ git format-patch --cover-letter -o psuh/ --base=auto psuh@{u}..psuh ---- . The `--cover-letter` option tells `format-patch` to create a cover letter template for you. You will need to fill in the template before you're ready to send - but for now, the template will be next to your other patches. . The `-o psuh/` option tells `format-patch` to place the patch files into a directory. This is useful because `git send-email` can take a directory and send out all the patches from there. . The `--base=auto` option tells the command to record the "base commit", on which the recipient is expected to apply the patch series. The `auto` value will cause `format-patch` to compute the base commit automatically, which is the merge base of tip commit of the remote-tracking branch and the specified revision range. . The `psuh@{u}..psuh` option tells `format-patch` to generate patches for the commits you created on the `psuh` branch since it forked from its upstream (which is `origin/master` if you followed the example in the "Set up your workspace" section). If you are already on the `psuh` branch, you can just say `@{u}`, which means "commits on the current branch since it forked from its upstream", which is the same thing. The command will make one patch file per commit. After you run, you can go have a look at each of the patches with your favorite text editor and make sure everything looks alright; however, it's not recommended to make code fixups via the patch file. It's a better idea to make the change the normal way using `git rebase -i` or by adding a new commit than by modifying a patch. NOTE: Optionally, you can also use the `--rfc` flag to prefix your patch subject with ``[RFC PATCH]'' instead of ``[PATCH]''. RFC stands for ``request for comments'' and indicates that while your code isn't quite ready for submission, you'd like to begin the code review process. This can also be used when your patch is a proposal, but you aren't sure whether the community wants to solve the problem with that approach or not - to conduct a sort of design review. You may also see on the list patches marked ``WIP'' - this means they are incomplete but want reviewers to look at what they have so far. You can add this flag with `--subject-prefix=WIP`. Check and make sure that your patches and cover letter template exist in the directory you specified - you're nearly ready to send out your review! [[preparing-cover-letter]] === Preparing Email Since you invoked `format-patch` with `--cover-letter`, you've already got a cover letter template ready. Open it up in your favorite editor. You should see a number of headers present already. Check that your `From:` header is correct. Then modify your `Subject:` (see <> for how to choose good title for your patch series): ---- Subject: [PATCH 0/7] Add the 'psuh' command ---- Make sure you retain the ``[PATCH 0/X]'' part; that's what indicates to the Git community that this email is the beginning of a patch series, and many reviewers filter their email for this type of flag. You'll need to add some extra parameters when you invoke `git send-email` to add the cover letter. Next you'll have to fill out the body of your cover letter. Again, see <> for what content to include. The template created by `git format-patch --cover-letter` includes a diffstat. This gives reviewers a summary of what they're in for when reviewing your topic. The one generated for `psuh` from the sample implementation looks like this: ---- Documentation/git-psuh.adoc | 40 +++++++++++++++++++++ Makefile | 1 + builtin.h | 1 + builtin/psuh.c | 73 ++++++++++++++++++++++++++++++++++++++ git.c | 1 + t/t9999-psuh-tutorial.sh | 12 +++++++ 6 files changed, 128 insertions(+) create mode 100644 Documentation/git-psuh.adoc create mode 100644 builtin/psuh.c create mode 100755 t/t9999-psuh-tutorial.sh ---- Finally, the letter will include the version of Git used to generate the patches. You can leave that string alone. [[sending-git-send-email]] === Sending Email At this point you should have a directory `psuh/` which is filled with your patches and a cover letter. Time to mail it out! You can send it like this: ---- $ git send-email --to=target@example.com psuh/*.patch ---- NOTE: Check `git help send-email` for some other options which you may find valuable, such as changing the Reply-to address or adding more CC and BCC lines. :contrib-scripts: footnoteref:[contrib-scripts,Scripts under `contrib/` are + not part of the core `git` binary and must be called directly. Clone the Git + codebase and run `perl contrib/contacts/git-contacts`.] NOTE: If you're not sure whom to CC, running `contrib/contacts/git-contacts` can list potential reviewers. In addition, you can do `git send-email --cc-cmd='perl contrib/contacts/git-contacts' feature/*.patch`{contrib-scripts} to automatically pass this list of emails to `send-email`. NOTE: When you are sending a real patch, it will go to git@vger.kernel.org - but please don't send your patchset from the tutorial to the real mailing list! For now, you can send it to yourself, to make sure you understand how it will look. After you run the command above, you will be presented with an interactive prompt for each patch that's about to go out. This gives you one last chance to edit or quit sending something (but again, don't edit code this way). Once you press `y` or `a` at these prompts your emails will be sent! Congratulations! Awesome, now the community will drop everything and review your changes. (Just kidding - be patient!) [[v2-git-send-email]] === Sending v2 This section will focus on how to send a v2 of your patchset. To learn what should go into v2, skip ahead to <> for information on how to handle comments from reviewers. We'll reuse our `psuh` topic branch for v2. Before we make any changes, we'll mark the tip of our v1 branch for easy reference: ---- $ git checkout psuh $ git branch psuh-v1 ---- Refine your patch series by using `git rebase -i` to adjust commits based upon reviewer comments. Once the patch series is ready for submission, generate your patches again, but with some new flags: ---- $ git format-patch -v2 --cover-letter -o psuh/ --range-diff master..psuh-v1 master.. ---- The `--range-diff master..psuh-v1` parameter tells `format-patch` to include a range-diff between `psuh-v1` and `psuh` in the cover letter (see linkgit:git-range-diff[1]). This helps tell reviewers about the differences between your v1 and v2 patches. The `-v2` parameter tells `format-patch` to output your patches as version "2". For instance, you may notice that your v2 patches are all named like `v2-000n-my-commit-subject.patch`. `-v2` will also format your patches by prefixing them with "[PATCH v2]" instead of "[PATCH]", and your range-diff will be prefaced with "Range-diff against v1". After you run this command, `format-patch` will output the patches to the `psuh/` directory, alongside the v1 patches. Using a single directory makes it easy to refer to the old v1 patches while proofreading the v2 patches, but you will need to be careful to send out only the v2 patches. We will use a pattern like `psuh/v2-*.patch` (not `psuh/*.patch`, which would match v1 and v2 patches). Edit your cover letter again. Now is a good time to mention what's different between your last version and now, if it's something significant. You do not need the exact same body in your second cover letter; focus on explaining to reviewers the changes you've made that may not be as visible. You will also need to go and find the Message-ID of your previous cover letter. You can either note it when you send the first series, from the output of `git send-email`, or you can look it up on the https://lore.kernel.org/git[mailing list]. Find your cover letter in the archives, click on it, then click "permalink" or "raw" to reveal the Message-ID header. It should match: ---- Message-ID: ---- Your Message-ID is ``. This example will be used below as well; make sure to replace it with the correct Message-ID for your **previous cover letter** - that is, if you're sending v2, use the Message-ID from v1; if you're sending v3, use the Message-ID from v2. While you're looking at the email, you should also note who is CC'd, as it's common practice in the mailing list to keep all CCs on a thread. You can add these CC lines directly to your cover letter with a line like so in the header (before the Subject line): ---- CC: author@example.com, Othe R ---- Now send the emails again, paying close attention to which messages you pass in to the command: ---- $ git send-email --to=target@example.com --in-reply-to="" psuh/v2-*.patch ---- [[single-patch]] === Bonus Chapter: One-Patch Changes In some cases, your very small change may consist of only one patch. When that happens, you only need to send one email. Your commit message should already be meaningful and explain at a high level the purpose (what is happening and why) of your patch, but if you need to supply even more context, you can do so below the `---` in your patch. Take the example below, which was generated with `git format-patch` on a single commit, and then edited to add the content between the `---` and the diffstat. ---- From 1345bbb3f7ac74abde040c12e737204689a72723 Mon Sep 17 00:00:00 2001 From: A U Thor Date: Thu, 18 Apr 2019 15:11:02 -0700 Subject: [PATCH] README: change the grammar I think it looks better this way. This part of the commit message will end up in the commit-log. Signed-off-by: A U Thor --- Let's have a wild discussion about grammar on the mailing list. This part of my email will never end up in the commit log. Here is where I can add additional context to the mailing list about my intent, outside of the context of the commit log. This section was added after `git format-patch` was run, by editing the patch file in a text editor. README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 88f126184c..38da593a60 100644 --- a/README.md +++ b/README.md @@ -3,7 +3,7 @@ Git - fast, scalable, distributed revision control system ========================================================= -Git is a fast, scalable, distributed revision control system with an +Git is a fast, scalable, and distributed revision control system with an unusually rich command set that provides both high-level operations and full access to internals. -- 2.21.0.392.gf8f6787159e-goog ---- [[now-what]] == My Patch Got Emailed - Now What? Please give reviewers enough time to process your initial patch before sending an updated version. That is, resist the temptation to send a new version immediately, because others may have already started reviewing your initial version. While waiting for review comments, you may find mistakes in your initial patch, or perhaps realize a different and better way to achieve the goal of the patch. In this case you may communicate your findings to other reviewers as follows: - If the mistakes you found are minor, send a reply to your patch as if you were a reviewer and mention that you will fix them in an updated version. - On the other hand, if you think you want to change the course so drastically that reviews on the initial patch would be a waste of time (for everyone involved), retract the patch immediately with a reply like "I am working on a much better approach, so please ignore this patch and wait for the updated version." Now, the above is a good practice if you sent your initial patch prematurely without polish. But a better approach of course is to avoid sending your patch prematurely in the first place. Please be considerate of the time needed by reviewers to examine each new version of your patch. Rather than seeing the initial version right now (followed by several "oops, I like this version better than the previous one" patches over 2 days), reviewers would strongly prefer if a single polished version came 2 days later instead, and that version with fewer mistakes were the only one they would need to review. [[reviewing]] === Responding to Reviews After a few days, you will hopefully receive a reply to your patchset with some comments. Woohoo! Now you can get back to work. It's good manners to reply to each comment, notifying the reviewer that you have made the change suggested, feel the original is better, or that the comment inspired you to do something a new way which is superior to both the original and the suggested change. This way reviewers don't need to inspect your v2 to figure out whether you implemented their comment or not. Reviewers may ask you about what you wrote in the patchset, either in the proposed commit log message or in the changes themselves. You should answer these questions in your response messages, but often the reason why reviewers asked these questions to understand what you meant to write is because your patchset needed clarification to be understood. Do not be satisfied by just answering their questions in your response and hear them say that they now understand what you wanted to say. Update your patches to clarify the points reviewers had trouble with, and prepare your v2; the words you used to explain your v1 to answer reviewers' questions may be useful thing to use. Your goal is to make your v2 clear enough so that it becomes unnecessary for you to give the same explanation to the next person who reads it. If you are going to push back on a comment, be polite and explain why you feel your original is better; be prepared that the reviewer may still disagree with you, and the rest of the community may weigh in on one side or the other. As with all code reviews, it's important to keep an open mind to doing something a different way than you originally planned; other reviewers have a different perspective on the project than you do, and may be thinking of a valid side effect which had not occurred to you. It is always okay to ask for clarification if you aren't sure why a change was suggested, or what the reviewer is asking you to do. Make sure your email client has a plaintext email mode and it is turned on; the Git list rejects HTML email. Please also follow the mailing list etiquette outlined in the https://kernel.googlesource.com/pub/scm/git/git/+/todo/MaintNotes[Maintainer's Note], which are similar to etiquette rules in most open source communities surrounding bottom-posting and inline replies. When you're making changes to your code, it is cleanest - that is, the resulting commits are easiest to look at - if you use `git rebase -i` (interactive rebase). Take a look at this https://www.oreilly.com/library/view/git-pocket-guide/9781449327507/ch10.html[overview] from O'Reilly. The general idea is to modify each commit which requires changes; this way, instead of having a patch A with a mistake, a patch B which was fine and required no upstream reviews in v1, and a patch C which fixes patch A for v2, you can just ship a v2 with a correct patch A and correct patch B. This is changing history, but since it's local history which you haven't shared with anyone, that is okay for now! (Later, it may not make sense to do this; take a look at the section below this one for some context.) [[after-approval]] === After Review Approval The Git project has four integration branches: `seen`, `next`, `master`, and `maint`. Your change will be placed into `seen` fairly early on by the maintainer while it is still in the review process; from there, when it is ready for wider testing, it will be merged into `next`. Plenty of early testers use `next` and may report issues. Eventually, changes in `next` will make it to `master`, which is typically considered stable. Finally, when a new release is cut, `maint` is used to base bugfixes onto. As mentioned at the beginning of this document, you can read `Documents/SubmittingPatches` for some more info about the use of the various integration branches. Back to now: your code has been lauded by the upstream reviewers. It is perfect. It is ready to be accepted. You don't need to do anything else; the maintainer will merge your topic branch to `next` and life is good. However, if you discover it isn't so perfect after this point, you may need to take some special steps depending on where you are in the process. If the maintainer has announced in the "What's cooking in git.git" email that your topic is marked for `next` - that is, that they plan to merge it to `next` but have not yet done so - you should send an email asking the maintainer to wait a little longer: "I've sent v4 of my series and you marked it for `next`, but I need to change this and that - please wait for v5 before you merge it." If the topic has already been merged to `next`, rather than modifying your patches with `git rebase -i`, you should make further changes incrementally - that is, with another commit, based on top of the maintainer's topic branch as detailed in https://github.com/gitster/git. Your work is still in the same topic but is now incremental, rather than a wholesale rewrite of the topic branch. The topic branches in the maintainer's GitHub are mirrored in GitGitGadget, so if you're sending your reviews out that way, you should be sure to open your PR against the appropriate GitGitGadget/Git branch. If you're using `git send-email`, you can use it the same way as before, but you should generate your diffs from `..` and base your work on `` instead of `master`. ./MyFirstContribution.html000066400000000000000000003174541504515377700161320ustar00rootroot00000000000000 My First Contribution to the Git Project

Summary

This is a tutorial demonstrating the end-to-end workflow of creating a change to the Git tree, sending it for review, and making changes based on comments.

Prerequisites

This tutorial assumes you’re already fairly familiar with using Git to manage source code. The Git workflow steps will largely remain unexplained.

This tutorial aims to summarize the following documents, but the reader may find useful additional context:

  • Documentation/SubmittingPatches

  • Documentation/howto/new-command.adoc

Getting Help

If you get stuck, you can seek help in the following places.

git@vger.kernel.org

This is the main Git project mailing list where code reviews, version announcements, design discussions, and more take place. Those interested in contributing are welcome to post questions here. The Git list requires plain-text-only emails and prefers inline and bottom-posting when replying to mail; you will be CC’d in all replies to you. Optionally, you can subscribe to the list by sending an email to <git+subscribe@vger.kernel.org> (see https://subspace.kernel.org/subscribing.html for details). The archive of this mailing list is available to view in a browser.

#git-devel on Libera Chat

This IRC channel is for conversations between Git contributors. If someone is currently online and knows the answer to your question, you can receive help in real time. Otherwise, you can read the scrollback to see whether someone answered you. IRC does not allow offline private messaging, so if you try to private message someone and then log out of IRC, they cannot respond to you. It’s better to ask your questions in the channel so that you can be answered if you disconnect and so that others can learn from the conversation.

Getting Started

Clone the Git Repository

Git is mirrored in a number of locations. Clone the repository from one of them; https://git-scm.com/downloads suggests one of the best places to clone from is the mirror on GitHub.

$ git clone https://github.com/git/git git
$ cd git

Installing Dependencies

To build Git from source, you need to have a handful of dependencies installed on your system. For a hint of what’s needed, you can take a look at INSTALL, paying close attention to the section about Git’s dependencies on external programs and libraries. That document mentions a way to "test-drive" our freshly built Git without installing; that’s the method we’ll be using in this tutorial.

Make sure that your environment has everything you need by building your brand new clone of Git from the above step:

$ make
Note
 The Git build is parallelizable. -j# is not included above but you can use it as you prefer, here and elsewhere.

Identify Problem to Solve

In this tutorial, we will add a new command, git psuh, short for “Pony Saying ‘Um, Hello”’ - a feature which has gone unimplemented despite a high frequency of invocation during users' typical daily workflow.

(We’ve seen some other effort in this space with the implementation of popular commands such as sl.)

Set Up Your Workspace

Let’s start by making a development branch to work on our changes. Per Documentation/SubmittingPatches, since a brand new command is a new feature, it’s fine to base your work on master. However, in the future for bugfixes, etc., you should check that document and base it on the appropriate branch.

For the purposes of this document, we will base all our work on the master branch of the upstream project. Create the psuh branch you will use for development like so:

$ git checkout -b psuh origin/master

We’ll make a number of commits here in order to demonstrate how to send a topic with multiple patches up for review simultaneously.

Code It Up!

Note
 A reference implementation can be found at https://github.com/nasamuffin/git/tree/psuh.

Adding a New Command

Lots of the subcommands are written as builtins, which means they are implemented in C and compiled into the main git executable. Implementing the very simple psuh command as a built-in will demonstrate the structure of the codebase, the internal API, and the process of working together as a contributor with the reviewers and maintainer to integrate this change into the system.

Built-in subcommands are typically implemented in a function named "cmd_" followed by the name of the subcommand, in a source file named after the subcommand and contained within builtin/. So it makes sense to implement your command in builtin/psuh.c. Create that file, and within it, write the entry point for your command in a function matching the style and signature:

int cmd_psuh(int argc UNUSED, const char **argv UNUSED,
             const char *prefix UNUSED, struct repository *repo UNUSED)

A few things to note:

  • A subcommand implementation takes its command line arguments in int argc + const char **argv, like main() would.

  • It also takes two extra parameters, prefix and repo. What they mean will not be discussed until much later.

  • Because this first example will not use any of the parameters, your compiler will give warnings on unused parameters. As the list of these four parameters is mandated by the API to add new built-in commands, you cannot omit them. Instead, you add UNUSED to each of them to tell the compiler that you know you are not (yet) using it.

We’ll also need to add the declaration of psuh; open up builtin.h, find the declaration for cmd_pull, and add a new line for psuh immediately before it, in order to keep the declarations alphabetically sorted:

int cmd_psuh(int argc, const char **argv, const char *prefix, struct repository *repo);

Be sure to #include "builtin.h" in your psuh.c. You’ll also need to #include "gettext.h" to use functions related to printing output text.

Go ahead and add some throwaway printf to the cmd_psuh function. This is a decent starting point as we can now add build rules and register the command.

Note
 Your throwaway text, as well as much of the text you will be adding over the course of this tutorial, is user-facing. That means it needs to be localizable. Take a look at po/README under "Marking strings for translation". Throughout the tutorial, we will mark strings for translation as necessary; you should also do so when writing your user-facing commands in the future. 
int cmd_psuh(int argc UNUSED, const char **argv UNUSED,
             const char *prefix UNUSED, struct repository *repo UNUSED)
{
        printf(_("Pony saying hello goes here.\n"));
        return 0;
}

Let’s try to build it. Open Makefile, find where builtin/pull.o is added to BUILTIN_OBJS, and add builtin/psuh.o in the same way next to it in alphabetical order. Once you’ve done so, move to the top-level directory and build simply with make. Also add the DEVELOPER=1 variable to turn on some additional warnings:

$ echo DEVELOPER=1 >config.mak
$ make
Note
 When you are developing the Git project, it’s preferred that you use the DEVELOPER flag; if there’s some reason it doesn’t work for you, you can turn it off, but it’s a good idea to mention the problem to the mailing list.

Great, now your new command builds happily on its own. But nobody invokes it. Let’s change that.

The list of commands lives in git.c. We can register a new command by adding a cmd_struct to the commands[] array. struct cmd_struct takes a string with the command name, a function pointer to the command implementation, and a setup option flag. For now, let’s keep mimicking push. Find the line where cmd_push is registered, copy it, and modify it for cmd_psuh, placing the new line in alphabetical order (immediately before cmd_pull).

The options are documented in builtin.h under "Adding a new built-in." Since we hope to print some data about the user’s current workspace context later, we need a Git directory, so choose RUN_SETUP as your only option.

Go ahead and build again. You should see a clean build, so let’s kick the tires and see if it works. There’s a binary you can use to test with in the bin-wrappers directory.

$ ./bin-wrappers/git psuh

Check it out! You’ve got a command! Nice work! Let’s commit this.

git status reveals modified Makefile, builtin.h, and git.c as well as untracked builtin/psuh.c and git-psuh. First, let’s take care of the binary, which should be ignored. Open .gitignore in your editor, find /git-pull, and add an entry for your new command in alphabetical order:

...
/git-prune-packed
/git-psuh
/git-pull
/git-push
/git-quiltimport
/git-range-diff
...

Checking git status again should show that git-psuh has been removed from the untracked list and .gitignore has been added to the modified list. Now we can stage and commit:

$ git add Makefile builtin.h builtin/psuh.c git.c .gitignore
$ git commit -s

You will be presented with your editor in order to write a commit message. Start the commit with a 50-column or less subject line, including the name of the component you’re working on, followed by a blank line (always required) and then the body of your commit message, which should provide the bulk of the context. Remember to be explicit and provide the "Why" of your change, especially if it couldn’t easily be understood from your diff. When editing your commit message, don’t remove the Signed-off-by trailer which was added by -s above.

psuh: add a built-in by popular demand

Internal metrics indicate this is a command many users expect to be
present. So here's an implementation to help drive customer
satisfaction and engagement: a pony which doubtfully greets the user,
or, a Pony Saying "Um, Hello" (PSUH).

This commit message is intentionally formatted to 72 columns per line,
starts with a single line as "commit message subject" that is written as
if to command the codebase to do something (add this, teach a command
that). The body of the message is designed to add information about the
commit that is not readily deduced from reading the associated diff,
such as answering the question "why?".

Signed-off-by: A U Thor <author@example.com>

Go ahead and inspect your new commit with git show. "psuh:" indicates you have modified mainly the psuh command. The subject line gives readers an idea of what you’ve changed. The sign-off line (-s) indicates that you agree to the Developer’s Certificate of Origin 1.1 (see the Documentation/SubmittingPatches [[dco]] header).

For the remainder of the tutorial, the subject line only will be listed for the sake of brevity. However, fully-fleshed example commit messages are available on the reference implementation linked at the top of this document.

Implementation

It’s probably useful to do at least something besides printing out a string. Let’s start by having a look at everything we get.

Modify your cmd_psuh implementation to dump the args you’re passed, keeping existing printf() calls in place; because the args are now used, remove the UNUSED macro from them:

        int i;

        ...

        printf(Q_("Your args (there is %d):\n",
                  "Your args (there are %d):\n",
                  argc),
               argc);
        for (i = 0; i < argc; i++)
                printf("%d: %s\n", i, argv[i]);

        printf(_("Your current working directory:\n<top-level>%s%s\n"),
               prefix ? "/" : "", prefix ? prefix : "");

Build and try it. As you may expect, there’s pretty much just whatever we give on the command line, including the name of our command. (If prefix is empty for you, try cd Documentation/ && ../bin-wrappers/git psuh). That’s not so helpful. So what other context can we get?

Add a line to #include "config.h" and #include "repository.h". Then, add the following bits to the function body: function body:

        const char *cfg_name;

...

        repo_config(repo, git_default_config, NULL);
        if (repo_config_get_string_tmp(repo, "user.name", &cfg_name))
                printf(_("No name is found in config\n"));
        else
                printf(_("Your name: %s\n"), cfg_name);

repo_config() will grab the configuration from config files known to Git and apply standard precedence rules. repo_config_get_string_tmp() will look up a specific key ("user.name") and give you the value. There are a number of single-key lookup functions like this one; you can see them all (and more info about how to use repo_config()) in Documentation/technical/api-config.adoc.

You should see that the name printed matches the one you see when you run:

$ git config --get user.name

Great! Now we know how to check for values in the Git config. Let’s commit this too, so we don’t lose our progress.

$ git add builtin/psuh.c
$ git commit -sm "psuh: show parameters & config opts"
Note
 Again, the above is for sake of brevity in this tutorial. In a real change you should not use -m but instead use the editor to write a meaningful message.

Still, it’d be nice to know what the user’s working context is like. Let’s see if we can print the name of the user’s current branch. We can mimic the git status implementation; the printer is located in wt-status.c and we can see that the branch is held in a struct wt_status.

wt_status_print() gets invoked by cmd_status() in builtin/commit.c. Looking at that implementation we see the status config being populated like so:

status_init_config(&s, git_status_config);

But as we drill down, we can find that status_init_config() wraps a call to repo_config(). Let’s modify the code we wrote in the previous commit.

Be sure to include the header to allow you to use struct wt_status:

#include "wt-status.h"

Then modify your cmd_psuh implementation to declare your struct wt_status, prepare it, and print its contents:

        struct wt_status status;

...

        wt_status_prepare(repo, &status);
        repo_config(repo, git_default_config, &status);

...

        printf(_("Your current branch: %s\n"), status.branch);

Run it again. Check it out - here’s the (verbose) name of your current branch!

Let’s commit this as well.

$ git add builtin/psuh.c
$ git commit -sm "psuh: print the current branch"

Now let’s see if we can get some info about a specific commit.

Luckily, there are some helpers for us here. commit.h has a function called lookup_commit_reference_by_name to which we can simply provide a hardcoded string; pretty.h has an extremely handy pp_commit_easy() call which doesn’t require a full format object to be passed.

Add the following includes:

#include "commit.h"
#include "pretty.h"

Then, add the following lines within your implementation of cmd_psuh() near the declarations and the logic, respectively.

        struct commit *c = NULL;
        struct strbuf commitline = STRBUF_INIT;

...

        c = lookup_commit_reference_by_name("origin/master");

        if (c != NULL) {
                pp_commit_easy(CMIT_FMT_ONELINE, c, &commitline);
                printf(_("Current commit: %s\n"), commitline.buf);
        }

The struct strbuf provides some safety belts to your basic char*, one of which is a length member to prevent buffer overruns. It needs to be initialized nicely with STRBUF_INIT. Keep it in mind when you need to pass around char*.

lookup_commit_reference_by_name resolves the name you pass it, so you can play with the value there and see what kind of things you can come up with.

pp_commit_easy is a convenience wrapper in pretty.h that takes a single format enum shorthand, rather than an entire format struct. It then pretty-prints the commit according to that shorthand. These are similar to the formats available with --pretty=FOO in many Git commands.

Build it and run, and if you’re using the same name in the example, you should see the subject line of the most recent commit in origin/master that you know about. Neat! Let’s commit that as well.

$ git add builtin/psuh.c
$ git commit -sm "psuh: display the top of origin/master"

Adding Documentation

Awesome! You’ve got a fantastic new command that you’re ready to share with the community. But hang on just a minute - this isn’t very user-friendly. Run the following:

$ ./bin-wrappers/git help psuh

Your new command is undocumented! Let’s fix that.

Take a look at Documentation/git-*.adoc. These are the manpages for the subcommands that Git knows about. You can open these up and take a look to get acquainted with the format, but then go ahead and make a new file Documentation/git-psuh.adoc. Like with most of the documentation in the Git project, help pages are written with AsciiDoc (see CodingGuidelines, "Writing Documentation" section). Use the following template to fill out your own manpage:

git-psuh(1)
===========

NAME
----
git-psuh - Delight users' typo with a shy horse


SYNOPSIS
--------
[verse]
'git-psuh [<arg>...]'

DESCRIPTION
-----------
...

OPTIONS[[OPTIONS]]
------------------
...

OUTPUT
------
...

GIT
---
Part of the linkgit:git[1] suite

The most important pieces of this to note are the file header, underlined by =, the NAME section, and the SYNOPSIS, which would normally contain the grammar if your command took arguments. Try to use well-established manpage headers so your documentation is consistent with other Git and UNIX manpages; this makes life easier for your user, who can skip to the section they know contains the information they need.

Note
 Before trying to build the docs, make sure you have the package asciidoc installed.

Now that you’ve written your manpage, you’ll need to build it explicitly. We convert your AsciiDoc to troff which is man-readable like so:

$ make all doc
$ man Documentation/git-psuh.1

or

$ make -C Documentation/ git-psuh.1
$ man Documentation/git-psuh.1

While this isn’t as satisfying as running through git help, you can at least check that your help page looks right.

You can also check that the documentation coverage is good (that is, the project sees that your command has been implemented as well as documented) by running make check-docs from the top-level.

Go ahead and commit your new documentation change.

Adding Usage Text

Try and run ./bin-wrappers/git psuh -h. Your command should crash at the end. That’s because -h is a special case which your command should handle by printing usage.

Take a look at Documentation/technical/api-parse-options.adoc. This is a handy tool for pulling out options you need to be able to handle, and it takes a usage string.

In order to use it, we’ll need to prepare a NULL-terminated array of usage strings and a builtin_psuh_options array.

Add a line to #include "parse-options.h".

At global scope, add your array of usage strings:

static const char * const psuh_usage[] = {
        N_("git psuh [<arg>...]"),
        NULL,
};

Then, within your cmd_psuh() implementation, we can declare and populate our option struct. Ours is pretty boring but you can add more to it if you want to explore parse_options() in more detail:

        struct option options[] = {
                OPT_END()
        };

Finally, before you print your args and prefix, add the call to parse-options():

        argc = parse_options(argc, argv, prefix, options, psuh_usage, 0);

This call will modify your argv parameter. It will strip the options you specified in options from argv and the locations pointed to from options entries will be updated. Be sure to replace your argc with the result from parse_options(), or you will be confused if you try to parse argv later.

It’s worth noting the special argument --. As you may be aware, many Unix commands use -- to indicate "end of named parameters" - all parameters after the -- are interpreted merely as positional arguments. (This can be handy if you want to pass as a parameter something which would usually be interpreted as a flag.) parse_options() will terminate parsing when it reaches -- and give you the rest of the options afterwards, untouched.

Now that you have a usage hint, you can teach Git how to show it in the general command list shown by git help git or git help -a, which is generated from command-list.txt. Find the line for git-pull so you can add your git-psuh line above it in alphabetical order. Now, we can add some attributes about the command which impacts where it shows up in the aforementioned help commands. The top of command-list.txt shares some information about what each attribute means; in those help pages, the commands are sorted according to these attributes. git psuh is user-facing, or porcelain - so we will mark it as "mainporcelain". For "mainporcelain" commands, the comments at the top of command-list.txt indicate we can also optionally add an attribute from another list; since git psuh shows some information about the user’s workspace but doesn’t modify anything, let’s mark it as "info". Make sure to keep your attributes in the same style as the rest of command-list.txt using spaces to align and delineate them:

git-prune-packed                        plumbingmanipulators
git-psuh                                mainporcelain           info
git-pull                                mainporcelain           remote
git-push                                mainporcelain           remote

Build again. Now, when you run with -h, you should see your usage printed and your command terminated before anything else interesting happens. Great!

Go ahead and commit this one, too.

Testing

It’s important to test your code - even for a little toy command like this one. Moreover, your patch won’t be accepted into the Git tree without tests. Your tests should:

  • Illustrate the current behavior of the feature

  • Prove the current behavior matches the expected behavior

  • Ensure the externally-visible behavior isn’t broken in later changes

So let’s write some tests.

Related reading: t/README

Overview of Testing Structure

The tests in Git live in t/ and are named with a 4-digit decimal number using the schema shown in the Naming Tests section of t/README.

Writing Your Test

Since this a toy command, let’s go ahead and name the test with t9999. However, as many of the family/subcmd combinations are full, best practice seems to be to find a command close enough to the one you’ve added and share its naming space.

Create a new file t/t9999-psuh-tutorial.sh. Begin with the header as so (see "Writing Tests" and "Source test-lib.sh" in t/README):

#!/bin/sh

test_description='git-psuh test

This test runs git-psuh and makes sure it does not crash.'

. ./test-lib.sh

Tests are framed inside of a test_expect_success in order to output TAP formatted results. Let’s make sure that git psuh doesn’t exit poorly and does mention the right animal somewhere:

test_expect_success 'runs correctly with no args and good output' '
        git psuh >actual &&
        grep Pony actual
'

Indicate that you’ve run everything you wanted by adding the following at the bottom of your script:

test_done

Make sure you mark your test script executable:

$ chmod +x t/t9999-psuh-tutorial.sh

You can get an idea of whether you created your new test script successfully by running make -C t test-lint, which will check for things like test number uniqueness, executable bit, and so on.

Running Locally

Let’s try and run locally:

$ make
$ cd t/ && prove t9999-psuh-tutorial.sh

You can run the full test suite and ensure git-psuh didn’t break anything:

$ cd t/
$ prove -j$(nproc) --shuffle t[0-9]*.sh
Note
 You can also do this with make test or use any testing harness which can speak TAP. prove can run concurrently. shuffle randomizes the order the tests are run in, which makes them resilient against unwanted inter-test dependencies. prove also makes the output nicer.

Go ahead and commit this change, as well.

Getting Ready to Share: Anatomy of a Patch Series

You may have noticed already that the Git project performs its code reviews via emailed patches, which are then applied by the maintainer when they are ready and approved by the community. The Git project does not accept contributions from pull requests, and the patches emailed for review need to be formatted a specific way.

Before taking a look at how to convert your commits into emailed patches, let’s analyze what the end result, a "patch series", looks like. Here is an example of the summary view for a patch series on the web interface of the Git mailing list archive:

2022-02-18 18:40 [PATCH 0/3] libify reflog John Cai via GitGitGadget
2022-02-18 18:40 ` [PATCH 1/3] reflog: libify delete reflog function and helpers John Cai via GitGitGadget
2022-02-18 19:10   ` Ævar Arnfjörð Bjarmason [this message]
2022-02-18 19:39     ` Taylor Blau
2022-02-18 19:48       ` Ævar Arnfjörð Bjarmason
2022-02-18 19:35   ` Taylor Blau
2022-02-21  1:43     ` John Cai
2022-02-21  1:50       ` Taylor Blau
2022-02-23 19:50         ` John Cai
2022-02-18 20:00   ` // other replies elided
2022-02-18 18:40 ` [PATCH 2/3] reflog: call reflog_delete from reflog.c John Cai via GitGitGadget
2022-02-18 19:15   ` Ævar Arnfjörð Bjarmason
2022-02-18 20:26     ` Junio C Hamano
2022-02-18 18:40 ` [PATCH 3/3] stash: call reflog_delete from reflog.c John Cai via GitGitGadget
2022-02-18 19:20   ` Ævar Arnfjörð Bjarmason
2022-02-19  0:21     ` Taylor Blau
2022-02-22  2:36     ` John Cai
2022-02-22 10:51       ` Ævar Arnfjörð Bjarmason
2022-02-18 19:29 ` [PATCH 0/3] libify reflog Ævar Arnfjörð Bjarmason
2022-02-22 18:30 ` [PATCH v2 0/3] libify reflog John Cai via GitGitGadget
2022-02-22 18:30   ` [PATCH v2 1/3] stash: add test to ensure reflog --rewrite --updatref behavior John Cai via GitGitGadget
2022-02-23  8:54     ` Ævar Arnfjörð Bjarmason
2022-02-23 21:27       ` Junio C Hamano
// continued

We can note a few things:

  • Each commit is sent as a separate email, with the commit message title as subject, prefixed with "[PATCH i/n]" for the i-th commit of an n-commit series.

  • Each patch is sent as a reply to an introductory email called the cover letter of the series, prefixed "[PATCH 0/n]".

  • Subsequent iterations of the patch series are labelled "PATCH v2", "PATCH v3", etc. in place of "PATCH". For example, "[PATCH v2 1/3]" would be the first of three patches in the second iteration. Each iteration is sent with a new cover letter (like "[PATCH v2 0/3]" above), itself a reply to the cover letter of the previous iteration (more on that below).

Note
 A single-patch topic is sent with "[PATCH]", "[PATCH v2]", etc. without i/n numbering (in the above thread overview, no single-patch topic appears, though).

The cover letter

In addition to an email per patch, the Git community also expects your patches to come with a cover letter. This is an important component of change submission as it explains to the community from a high level what you’re trying to do, and why, in a way that’s more apparent than just looking at your patches.

The title of your cover letter should be something which succinctly covers the purpose of your entire topic branch. It’s often in the imperative mood, just like our commit message titles. Here is how we’ll title our series:

Add the psuh command ---

The body of the cover letter is used to give additional context to reviewers. Be sure to explain anything your patches don’t make clear on their own, but remember that since the cover letter is not recorded in the commit history, anything that might be useful to future readers of the repository’s history should also be in your commit messages.

Here’s an example body for psuh:

Our internal metrics indicate widespread interest in the command
git-psuh - that is, many users are trying to use it, but finding it is
unavailable, using some unknown workaround instead.

The following handful of patches add the psuh command and implement some
handy features on top of it.

This patchset is part of the MyFirstContribution tutorial and should not
be merged.

At this point the tutorial diverges, in order to demonstrate two different methods of formatting your patchset and getting it reviewed.

The first method to be covered is GitGitGadget, which is useful for those already familiar with GitHub’s common pull request workflow. This method requires a GitHub account.

The second method to be covered is git send-email, which can give slightly more fine-grained control over the emails to be sent. This method requires some setup which can change depending on your system and will not be covered in this tutorial.

Regardless of which method you choose, your engagement with reviewers will be the same; the review process will be covered after the sections on GitGitGadget and git send-email.

Sending Patches via GitGitGadget

One option for sending patches is to follow a typical pull request workflow and send your patches out via GitGitGadget. GitGitGadget is a tool created by Johannes Schindelin to make life as a Git contributor easier for those used to the GitHub PR workflow. It allows contributors to open pull requests against its mirror of the Git project, and does some magic to turn the PR into a set of emails and send them out for you. It also runs the Git continuous integration suite for you. It’s documented at https://gitgitgadget.github.io/.

Forking git/git on GitHub

Before you can send your patch off to be reviewed using GitGitGadget, you will need to fork the Git project and upload your changes. First thing - make sure you have a GitHub account.

Head to the GitHub mirror and look for the Fork button. Place your fork wherever you deem appropriate and create it.

Uploading to Your Own Fork

To upload your branch to your own fork, you’ll need to add the new fork as a remote. You can use git remote -v to show the remotes you have added already. From your new fork’s page on GitHub, you can press "Clone or download" to get the URL; then you need to run the following to add, replacing your own URL and remote name for the examples provided:

$ git remote add remotename git@github.com:remotename/git.git

or to use the HTTPS URL:

$ git remote add remotename https://github.com/remotename/git/.git

Run git remote -v again and you should see the new remote showing up. git fetch remotename (with the real name of your remote replaced) in order to get ready to push.

Next, double-check that you’ve been doing all your development in a new branch by running git branch. If you didn’t, now is a good time to move your new commits to their own branch.

As mentioned briefly at the beginning of this document, we are basing our work on master, so go ahead and update as shown below, or using your preferred workflow.

$ git checkout master
$ git pull -r
$ git rebase master psuh

Finally, you’re ready to push your new topic branch! (Due to our branch and command name choices, be careful when you type the command below.)

$ git push remotename psuh

Now you should be able to go and check out your newly created branch on GitHub.

Sending a PR to GitGitGadget

In order to have your code tested and formatted for review, you need to start by opening a Pull Request against gitgitgadget/git. Head to https://github.com/gitgitgadget/git and open a PR either with the "New pull request" button or the convenient "Compare & pull request" button that may appear with the name of your newly pushed branch.

Review the PR’s title and description, as they’re used by GitGitGadget respectively as the subject and body of the cover letter for your change. Refer to "The cover letter" above for advice on how to title your submission and what content to include in the description.

Note
 For single-patch contributions, your commit message should already be meaningful and explain at a high level the purpose (what is happening and why) of your patch, so you usually do not need any additional context. In that case, remove the PR description that GitHub automatically generates from your commit message (your PR description should be empty). If you do need to supply even more context, you can do so in that space and it will be appended to the email that GitGitGadget will send, between the three-dash line and the diffstat (see Bonus Chapter: One-Patch Changes for how this looks once submitted).

When you’re happy, submit your pull request.

Running CI and Getting Ready to Send

If it’s your first time using GitGitGadget (which is likely, as you’re using this tutorial) then someone will need to give you permission to use the tool. As mentioned in the GitGitGadget documentation, you just need someone who already uses it to comment on your PR with /allow <username>. GitGitGadget will automatically run your PRs through the CI even without the permission given but you will not be able to /submit your changes until someone allows you to use the tool.

Note
 You can typically find someone who can /allow you on GitGitGadget by either examining recent pull requests where someone has been granted /allow (Search: is:pr is:open "/allow"), in which case both the author and the person who granted the /allow can now /allow you, or by inquiring on the #git-devel IRC channel on Libera Chat linking your pull request and asking for someone to /allow you.

If the CI fails, you can update your changes with git rebase -i and push your branch again:

$ git push -f remotename psuh

In fact, you should continue to make changes this way up until the point when your patch is accepted into next.

Sending Your Patches

Now that your CI is passing and someone has granted you permission to use GitGitGadget with the /allow command, sending out for review is as simple as commenting on your PR with /submit.

Updating With Comments

Skip ahead to Responding to Reviews for information on how to reply to review comments you will receive on the mailing list.

Once you have your branch again in the shape you want following all review comments, you can submit again:

$ git push -f remotename psuh

Next, go look at your pull request against GitGitGadget; you should see the CI has been kicked off again. Now while the CI is running is a good time for you to modify your description at the top of the pull request thread; it will be used again as the cover letter. You should use this space to describe what has changed since your previous version, so that your reviewers have some idea of what they’re looking at. When the CI is done running, you can comment once more with /submit - GitGitGadget will automatically add a v2 mark to your changes.

Sending Patches with git send-email

If you don’t want to use GitGitGadget, you can also use Git itself to mail your patches. Some benefits of using Git this way include finer grained control of subject line (for example, being able to use the tag [RFC PATCH] in the subject) and being able to send a “dry run” mail to yourself to ensure it all looks good before going out to the list.

Prerequisite: Setting Up git send-email

Configuration for send-email can vary based on your operating system and email provider, and so will not be covered in this tutorial, beyond stating that in many distributions of Linux, git-send-email is not packaged alongside the typical git install. You may need to install this additional package; there are a number of resources online to help you do so. You will also need to determine the right way to configure it to use your SMTP server; again, as this configuration can change significantly based on your system and email setup, it is out of scope for the context of this tutorial.

Preparing Initial Patchset

Sending emails with Git is a two-part process; before you can prepare the emails themselves, you’ll need to prepare the patches. Luckily, this is pretty simple:

$ git format-patch --cover-letter -o psuh/ --base=auto psuh@{u}..psuh

  1. The --cover-letter option tells format-patch to create a cover letter template for you. You will need to fill in the template before you’re ready to send - but for now, the template will be next to your other patches.

  2. The -o psuh/ option tells format-patch to place the patch files into a directory. This is useful because git send-email can take a directory and send out all the patches from there.

  3. The --base=auto option tells the command to record the "base commit", on which the recipient is expected to apply the patch series. The auto value will cause format-patch to compute the base commit automatically, which is the merge base of tip commit of the remote-tracking branch and the specified revision range.

  4. The psuh@{u}..psuh option tells format-patch to generate patches for the commits you created on the psuh branch since it forked from its upstream (which is origin/master if you followed the example in the "Set up your workspace" section). If you are already on the psuh branch, you can just say @{u}, which means "commits on the current branch since it forked from its upstream", which is the same thing.

The command will make one patch file per commit. After you run, you can go have a look at each of the patches with your favorite text editor and make sure everything looks alright; however, it’s not recommended to make code fixups via the patch file. It’s a better idea to make the change the normal way using git rebase -i or by adding a new commit than by modifying a patch.

Note
 Optionally, you can also use the --rfc flag to prefix your patch subject with “[RFC PATCH]” instead of “[PATCH]”. RFC stands for “request for comments” and indicates that while your code isn’t quite ready for submission, you’d like to begin the code review process. This can also be used when your patch is a proposal, but you aren’t sure whether the community wants to solve the problem with that approach or not - to conduct a sort of design review. You may also see on the list patches marked “WIP” - this means they are incomplete but want reviewers to look at what they have so far. You can add this flag with --subject-prefix=WIP.

Check and make sure that your patches and cover letter template exist in the directory you specified - you’re nearly ready to send out your review!

Preparing Email

Since you invoked format-patch with --cover-letter, you’ve already got a cover letter template ready. Open it up in your favorite editor.

You should see a number of headers present already. Check that your From: header is correct. Then modify your Subject: (see above for how to choose good title for your patch series):

Subject: [PATCH 0/7] Add the 'psuh' command

Make sure you retain the “[PATCH 0/X]” part; that’s what indicates to the Git community that this email is the beginning of a patch series, and many reviewers filter their email for this type of flag.

You’ll need to add some extra parameters when you invoke git send-email to add the cover letter.

Next you’ll have to fill out the body of your cover letter. Again, see above for what content to include.

The template created by git format-patch --cover-letter includes a diffstat. This gives reviewers a summary of what they’re in for when reviewing your topic. The one generated for psuh from the sample implementation looks like this:

 Documentation/git-psuh.adoc | 40 +++++++++++++++++++++
 Makefile                    |  1 +
 builtin.h                   |  1 +
 builtin/psuh.c              | 73 ++++++++++++++++++++++++++++++++++++++
 git.c                       |  1 +
 t/t9999-psuh-tutorial.sh    | 12 +++++++
 6 files changed, 128 insertions(+)
 create mode 100644 Documentation/git-psuh.adoc
 create mode 100644 builtin/psuh.c
 create mode 100755 t/t9999-psuh-tutorial.sh

Finally, the letter will include the version of Git used to generate the patches. You can leave that string alone.

Sending Email

At this point you should have a directory psuh/ which is filled with your patches and a cover letter. Time to mail it out! You can send it like this:

$ git send-email --to=target@example.com psuh/*.patch
Note
 Check git help send-email for some other options which you may find valuable, such as changing the Reply-to address or adding more CC and BCC lines.
Note
 If you’re not sure whom to CC, running contrib/contacts/git-contacts can list potential reviewers. In addition, you can do git send-email --cc-cmd='perl contrib/contacts/git-contacts' feature/*.patch[1] to automatically pass this list of emails to send-email.
Note
 When you are sending a real patch, it will go to git@vger.kernel.org - but please don’t send your patchset from the tutorial to the real mailing list! For now, you can send it to yourself, to make sure you understand how it will look.

After you run the command above, you will be presented with an interactive prompt for each patch that’s about to go out. This gives you one last chance to edit or quit sending something (but again, don’t edit code this way). Once you press y or a at these prompts your emails will be sent! Congratulations!

Awesome, now the community will drop everything and review your changes. (Just kidding - be patient!)

Sending v2

This section will focus on how to send a v2 of your patchset. To learn what should go into v2, skip ahead to Responding to Reviews for information on how to handle comments from reviewers.

We’ll reuse our psuh topic branch for v2. Before we make any changes, we’ll mark the tip of our v1 branch for easy reference:

$ git checkout psuh
$ git branch psuh-v1

Refine your patch series by using git rebase -i to adjust commits based upon reviewer comments. Once the patch series is ready for submission, generate your patches again, but with some new flags:

$ git format-patch -v2 --cover-letter -o psuh/ --range-diff master..psuh-v1 master..

The --range-diff master..psuh-v1 parameter tells format-patch to include a range-diff between psuh-v1 and psuh in the cover letter (see git-range-diff(1)). This helps tell reviewers about the differences between your v1 and v2 patches.

The -v2 parameter tells format-patch to output your patches as version "2". For instance, you may notice that your v2 patches are all named like v2-000n-my-commit-subject.patch. -v2 will also format your patches by prefixing them with "[PATCH v2]" instead of "[PATCH]", and your range-diff will be prefaced with "Range-diff against v1".

After you run this command, format-patch will output the patches to the psuh/ directory, alongside the v1 patches. Using a single directory makes it easy to refer to the old v1 patches while proofreading the v2 patches, but you will need to be careful to send out only the v2 patches. We will use a pattern like psuh/v2-*.patch (not psuh/*.patch, which would match v1 and v2 patches).

Edit your cover letter again. Now is a good time to mention what’s different between your last version and now, if it’s something significant. You do not need the exact same body in your second cover letter; focus on explaining to reviewers the changes you’ve made that may not be as visible.

You will also need to go and find the Message-ID of your previous cover letter. You can either note it when you send the first series, from the output of git send-email, or you can look it up on the mailing list. Find your cover letter in the archives, click on it, then click "permalink" or "raw" to reveal the Message-ID header. It should match:

Message-ID: <foo.12345.author@example.com>

Your Message-ID is <foo.12345.author@example.com>. This example will be used below as well; make sure to replace it with the correct Message-ID for your previous cover letter - that is, if you’re sending v2, use the Message-ID from v1; if you’re sending v3, use the Message-ID from v2.

While you’re looking at the email, you should also note who is CC’d, as it’s common practice in the mailing list to keep all CCs on a thread. You can add these CC lines directly to your cover letter with a line like so in the header (before the Subject line):

CC: author@example.com, Othe R <other@example.com>

Now send the emails again, paying close attention to which messages you pass in to the command:

$ git send-email --to=target@example.com
                 --in-reply-to="<foo.12345.author@example.com>"
                 psuh/v2-*.patch

Bonus Chapter: One-Patch Changes

In some cases, your very small change may consist of only one patch. When that happens, you only need to send one email. Your commit message should already be meaningful and explain at a high level the purpose (what is happening and why) of your patch, but if you need to supply even more context, you can do so below the --- in your patch. Take the example below, which was generated with git format-patch on a single commit, and then edited to add the content between the --- and the diffstat.

From 1345bbb3f7ac74abde040c12e737204689a72723 Mon Sep 17 00:00:00 2001
From: A U Thor <author@example.com>
Date: Thu, 18 Apr 2019 15:11:02 -0700
Subject: [PATCH] README: change the grammar

I think it looks better this way. This part of the commit message will
end up in the commit-log.

Signed-off-by: A U Thor <author@example.com>
---
Let's have a wild discussion about grammar on the mailing list. This
part of my email will never end up in the commit log. Here is where I
can add additional context to the mailing list about my intent, outside
of the context of the commit log. This section was added after `git
format-patch` was run, by editing the patch file in a text editor.

 README.md | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/README.md b/README.md
index 88f126184c..38da593a60 100644
--- a/README.md
+++ b/README.md
@@ -3,7 +3,7 @@
 Git - fast, scalable, distributed revision control system
 =========================================================

-Git is a fast, scalable, distributed revision control system with an
+Git is a fast, scalable, and distributed revision control system with an
 unusually rich command set that provides both high-level operations
 and full access to internals.

--
2.21.0.392.gf8f6787159e-goog

My Patch Got Emailed - Now What?

Please give reviewers enough time to process your initial patch before sending an updated version. That is, resist the temptation to send a new version immediately, because others may have already started reviewing your initial version.

While waiting for review comments, you may find mistakes in your initial patch, or perhaps realize a different and better way to achieve the goal of the patch. In this case you may communicate your findings to other reviewers as follows:

  • If the mistakes you found are minor, send a reply to your patch as if you were a reviewer and mention that you will fix them in an updated version.

  • On the other hand, if you think you want to change the course so drastically that reviews on the initial patch would be a waste of time (for everyone involved), retract the patch immediately with a reply like "I am working on a much better approach, so please ignore this patch and wait for the updated version."

Now, the above is a good practice if you sent your initial patch prematurely without polish. But a better approach of course is to avoid sending your patch prematurely in the first place.

Please be considerate of the time needed by reviewers to examine each new version of your patch. Rather than seeing the initial version right now (followed by several "oops, I like this version better than the previous one" patches over 2 days), reviewers would strongly prefer if a single polished version came 2 days later instead, and that version with fewer mistakes were the only one they would need to review.

Responding to Reviews

After a few days, you will hopefully receive a reply to your patchset with some comments. Woohoo! Now you can get back to work.

It’s good manners to reply to each comment, notifying the reviewer that you have made the change suggested, feel the original is better, or that the comment inspired you to do something a new way which is superior to both the original and the suggested change. This way reviewers don’t need to inspect your v2 to figure out whether you implemented their comment or not.

Reviewers may ask you about what you wrote in the patchset, either in the proposed commit log message or in the changes themselves. You should answer these questions in your response messages, but often the reason why reviewers asked these questions to understand what you meant to write is because your patchset needed clarification to be understood.

Do not be satisfied by just answering their questions in your response and hear them say that they now understand what you wanted to say. Update your patches to clarify the points reviewers had trouble with, and prepare your v2; the words you used to explain your v1 to answer reviewers' questions may be useful thing to use. Your goal is to make your v2 clear enough so that it becomes unnecessary for you to give the same explanation to the next person who reads it.

If you are going to push back on a comment, be polite and explain why you feel your original is better; be prepared that the reviewer may still disagree with you, and the rest of the community may weigh in on one side or the other. As with all code reviews, it’s important to keep an open mind to doing something a different way than you originally planned; other reviewers have a different perspective on the project than you do, and may be thinking of a valid side effect which had not occurred to you. It is always okay to ask for clarification if you aren’t sure why a change was suggested, or what the reviewer is asking you to do.

Make sure your email client has a plaintext email mode and it is turned on; the Git list rejects HTML email. Please also follow the mailing list etiquette outlined in the Maintainer’s Note, which are similar to etiquette rules in most open source communities surrounding bottom-posting and inline replies.

When you’re making changes to your code, it is cleanest - that is, the resulting commits are easiest to look at - if you use git rebase -i (interactive rebase). Take a look at this overview from O’Reilly. The general idea is to modify each commit which requires changes; this way, instead of having a patch A with a mistake, a patch B which was fine and required no upstream reviews in v1, and a patch C which fixes patch A for v2, you can just ship a v2 with a correct patch A and correct patch B. This is changing history, but since it’s local history which you haven’t shared with anyone, that is okay for now! (Later, it may not make sense to do this; take a look at the section below this one for some context.)

After Review Approval

The Git project has four integration branches: seen, next, master, and maint. Your change will be placed into seen fairly early on by the maintainer while it is still in the review process; from there, when it is ready for wider testing, it will be merged into next. Plenty of early testers use next and may report issues. Eventually, changes in next will make it to master, which is typically considered stable. Finally, when a new release is cut, maint is used to base bugfixes onto. As mentioned at the beginning of this document, you can read Documents/SubmittingPatches for some more info about the use of the various integration branches.

Back to now: your code has been lauded by the upstream reviewers. It is perfect. It is ready to be accepted. You don’t need to do anything else; the maintainer will merge your topic branch to next and life is good.

However, if you discover it isn’t so perfect after this point, you may need to take some special steps depending on where you are in the process.

If the maintainer has announced in the "What’s cooking in git.git" email that your topic is marked for next - that is, that they plan to merge it to next but have not yet done so - you should send an email asking the maintainer to wait a little longer: "I’ve sent v4 of my series and you marked it for next, but I need to change this and that - please wait for v5 before you merge it."

If the topic has already been merged to next, rather than modifying your patches with git rebase -i, you should make further changes incrementally - that is, with another commit, based on top of the maintainer’s topic branch as detailed in https://github.com/gitster/git. Your work is still in the same topic but is now incremental, rather than a wholesale rewrite of the topic branch.

The topic branches in the maintainer’s GitHub are mirrored in GitGitGadget, so if you’re sending your reviews out that way, you should be sure to open your PR against the appropriate GitGitGadget/Git branch.

If you’re using git send-email, you can use it the same way as before, but you should generate your diffs from <topic>..<mybranch> and base your work on <topic> instead of master.

1. Scripts under `contrib/` are not part of the core `git` binary and must be called directly. Clone the Git codebase and run `perl contrib/contacts/git-contacts`.
./MyFirstObjectWalk.adoc000066400000000000000000000716021504515377700154320ustar00rootroot00000000000000= My First Object Walk == What's an Object Walk? The object walk is a key concept in Git - this is the process that underpins operations like object transfer and fsck. Beginning from a given commit, the list of objects is found by walking parent relationships between commits (commit X based on commit W) and containment relationships between objects (tree Y is contained within commit X, and blob Z is located within tree Y, giving our working tree for commit X something like `y/z.txt`). A related concept is the revision walk, which is focused on commit objects and their parent relationships and does not delve into other object types. The revision walk is used for operations like `git log`. === Related Reading - `Documentation/user-manual.adoc` under "Hacking Git" contains some coverage of the revision walker in its various incarnations. - `revision.h` - https://eagain.net/articles/git-for-computer-scientists/[Git for Computer Scientists] gives a good overview of the types of objects in Git and what your object walk is really describing. == Setting Up Create a new branch from `master`. ---- git checkout -b revwalk origin/master ---- We'll put our fiddling into a new command. For fun, let's name it `git walken`. Open up a new file `builtin/walken.c` and set up the command handler: ---- /* * "git walken" * * Part of the "My First Object Walk" tutorial. */ #include "builtin.h" #include "trace.h" int cmd_walken(int argc, const char **argv, const char *prefix, struct repository *repo) { trace_printf(_("cmd_walken incoming...\n")); return 0; } ---- NOTE: `trace_printf()`, defined in `trace.h`, differs from `printf()` in that it can be turned on or off at runtime. For the purposes of this tutorial, we will write `walken` as though it is intended for use as a "plumbing" command: that is, a command which is used primarily in scripts, rather than interactively by humans (a "porcelain" command). So we will send our debug output to `trace_printf()` instead. When running, enable trace output by setting the environment variable `GIT_TRACE`. Add usage text and `-h` handling, like all subcommands should consistently do (our test suite will notice and complain if you fail to do so). We'll need to include the `parse-options.h` header. ---- #include "parse-options.h" ... int cmd_walken(int argc, const char **argv, const char *prefix) { const char * const walken_usage[] = { N_("git walken"), NULL, }; struct option options[] = { OPT_END() }; argc = parse_options(argc, argv, prefix, options, walken_usage, 0); ... } ---- Also add the relevant line in `builtin.h` near `cmd_version()`: ---- int cmd_walken(int argc, const char **argv, const char *prefix, struct repository *repo); ---- Include the command in `git.c` in `commands[]` near the entry for `version`, maintaining alphabetical ordering: ---- { "walken", cmd_walken, RUN_SETUP }, ---- Add an entry for the new command in the both the Make and Meson build system, before the entry for `worktree`: - In the `Makefile`: ---- ... BUILTIN_OBJS += builtin/walken.o ... ---- - In the `meson.build` file: ---- builtin_sources = [ ... 'builtin/walken.c', ... ] ---- Build and test out your command, without forgetting to ensure the `DEVELOPER` flag is set, and with `GIT_TRACE` enabled so the debug output can be seen: ---- $ echo DEVELOPER=1 >>config.mak $ make $ GIT_TRACE=1 ./bin-wrappers/git walken ---- NOTE: For a more exhaustive overview of the new command process, take a look at `Documentation/MyFirstContribution.adoc`. NOTE: A reference implementation can be found at https://github.com/nasamuffin/git/tree/revwalk. === `struct rev_cmdline_info` The definition of `struct rev_cmdline_info` can be found in `revision.h`. This struct is contained within the `rev_info` struct and is used to reflect parameters provided by the user over the CLI. `nr` represents the number of `rev_cmdline_entry` present in the array. `alloc` is used by the `ALLOC_GROW` macro. Check `alloc.h` - this variable is used to track the allocated size of the list. Per entry, we find: `item` is the object provided upon which to base the object walk. Items in Git can be blobs, trees, commits, or tags. (See `Documentation/gittutorial-2.adoc`.) `name` is the object ID (OID) of the object - a hex string you may be familiar with from using Git to organize your source in the past. Check the tutorial mentioned above towards the top for a discussion of where the OID can come from. `whence` indicates some information about what to do with the parents of the specified object. We'll explore this flag more later on; take a look at `Documentation/revisions.adoc` to get an idea of what could set the `whence` value. `flags` are used to hint the beginning of the revision walk and are the first block under the `#include`s in `revision.h`. The most likely ones to be set in the `rev_cmdline_info` are `UNINTERESTING` and `BOTTOM`, but these same flags can be used during the walk, as well. === `struct rev_info` This one is quite a bit longer, and many fields are only used during the walk by `revision.c` - not configuration options. Most of the configurable flags in `struct rev_info` have a mirror in `Documentation/rev-list-options.adoc`. It's a good idea to take some time and read through that document. == Basic Commit Walk First, let's see if we can replicate the output of `git log --oneline`. We'll refer back to the implementation frequently to discover norms when performing an object walk of our own. To do so, we'll first find all the commits, in order, which preceded the current commit. We'll extract the name and subject of the commit from each. Ideally, we will also be able to find out which ones are currently at the tip of various branches. === Setting Up Preparing for your object walk has some distinct stages. 1. Perform default setup for this mode, and others which may be invoked. 2. Check configuration files for relevant settings. 3. Set up the `rev_info` struct. 4. Tweak the initialized `rev_info` to suit the current walk. 5. Prepare the `rev_info` for the walk. 6. Iterate over the objects, processing each one. ==== Default Setups Before examining configuration files which may modify command behavior, set up default state for switches or options your command may have. If your command utilizes other Git components, ask them to set up their default states as well. For instance, `git log` takes advantage of `grep` and `diff` functionality, so its `init_log_defaults()` sets its own state (`decoration_style`) and asks `grep` and `diff` to initialize themselves by calling each of their initialization functions. ==== Configuring From `.gitconfig` Next, we should have a look at any relevant configuration settings (i.e., settings readable and settable from `git config`). This is done by providing a callback to `repo_config()`; within that callback, you can also invoke methods from other components you may need that need to intercept these options. Your callback will be invoked once per each configuration value which Git knows about (global, local, worktree, etc.). Similarly to the default values, we don't have anything to do here yet ourselves; however, we should call `git_default_config()` if we aren't calling any other existing config callbacks. Add a new function to `builtin/walken.c`. We'll also need to include the `config.h` header: ---- #include "config.h" ... static int git_walken_config(const char *var, const char *value, const struct config_context *ctx, void *cb) { /* * For now, we don't have any custom configuration, so fall back to * the default config. */ return git_default_config(var, value, ctx, cb); } ---- Make sure to invoke `repo_config()` with it in your `cmd_walken()`: ---- int cmd_walken(int argc, const char **argv, const char *prefix, struct repository *repo) { ... repo_config(repo, git_walken_config, NULL); ... } ---- ==== Setting Up `rev_info` Now that we've gathered external configuration and options, it's time to initialize the `rev_info` object which we will use to perform the walk. This is typically done by calling `repo_init_revisions()` with the repository you intend to target, as well as the `prefix` argument of `cmd_walken` and your `rev_info` struct. Add the `struct rev_info` and the `repo_init_revisions()` call. We'll also need to include the `revision.h` header: ---- #include "revision.h" ... int cmd_walken(int argc, const char **argv, const char *prefix, struct repository *repo) { /* This can go wherever you like in your declarations.*/ struct rev_info rev; ... /* This should go after the repo_config() call. */ repo_init_revisions(repo, &rev, prefix); ... } ---- ==== Tweaking `rev_info` For the Walk We're getting close, but we're still not quite ready to go. Now that `rev` is initialized, we can modify it to fit our needs. This is usually done within a helper for clarity, so let's add one: ---- static void final_rev_info_setup(struct rev_info *rev) { /* * We want to mimic the appearance of `git log --oneline`, so let's * force oneline format. */ get_commit_format("oneline", rev); /* Start our object walk at HEAD. */ add_head_to_pending(rev); } ---- [NOTE] ==== Instead of using the shorthand `add_head_to_pending()`, you could do something like this: ---- struct setup_revision_opt opt; memset(&opt, 0, sizeof(opt)); opt.def = "HEAD"; opt.revarg_opt = REVARG_COMMITTISH; setup_revisions(argc, argv, rev, &opt); ---- Using a `setup_revision_opt` gives you finer control over your walk's starting point. ==== Then let's invoke `final_rev_info_setup()` after the call to `repo_init_revisions()`: ---- int cmd_walken(int argc, const char **argv, const char *prefix, struct repository *repo) { ... final_rev_info_setup(&rev); ... } ---- Later, we may wish to add more arguments to `final_rev_info_setup()`. But for now, this is all we need. ==== Preparing `rev_info` For the Walk Now that `rev` is all initialized and configured, we've got one more setup step before we get rolling. We can do this in a helper, which will both prepare the `rev_info` for the walk, and perform the walk itself. Let's start the helper with the call to `prepare_revision_walk()`, which can return an error without dying on its own: ---- static void walken_commit_walk(struct rev_info *rev) { if (prepare_revision_walk(rev)) die(_("revision walk setup failed")); } ---- NOTE: `die()` prints to `stderr` and exits the program. Since it will print to `stderr` it's likely to be seen by a human, so we will localize it. ==== Performing the Walk! Finally! We are ready to begin the walk itself. Now we can see that `rev_info` can also be used as an iterator; we move to the next item in the walk by using `get_revision()` repeatedly. Add the listed variable declarations at the top and the walk loop below the `prepare_revision_walk()` call within your `walken_commit_walk()`: ---- #include "pretty.h" ... static void walken_commit_walk(struct rev_info *rev) { struct commit *commit; struct strbuf prettybuf = STRBUF_INIT; ... while ((commit = get_revision(rev))) { strbuf_reset(&prettybuf); pp_commit_easy(CMIT_FMT_ONELINE, commit, &prettybuf); puts(prettybuf.buf); } strbuf_release(&prettybuf); } ---- NOTE: `puts()` prints a `char*` to `stdout`. Since this is the part of the command we expect to be machine-parsed, we're sending it directly to stdout. Give it a shot. ---- $ make $ ./bin-wrappers/git walken ---- You should see all of the subject lines of all the commits in your tree's history, in order, ending with the initial commit, "Initial revision of "git", the information manager from hell". Congratulations! You've written your first revision walk. You can play with printing some additional fields from each commit if you're curious; have a look at the functions available in `commit.h`. === Adding a Filter Next, let's try to filter the commits we see based on their author. This is equivalent to running `git log --author=`. We can add a filter by modifying `rev_info.grep_filter`, which is a `struct grep_opt`. First some setup. Add `grep_config()` to `git_walken_config()`: ---- static int git_walken_config(const char *var, const char *value, const struct config_context *ctx, void *cb) { grep_config(var, value, ctx, cb); return git_default_config(var, value, ctx, cb); } ---- Next, we can modify the `grep_filter`. This is done with convenience functions found in `grep.h`. For fun, we're filtering to only commits from folks using a `gmail.com` email address - a not-very-precise guess at who may be working on Git as a hobby. Since we're checking the author, which is a specific line in the header, we'll use the `append_header_grep_pattern()` helper. We can use the `enum grep_header_field` to indicate which part of the commit header we want to search. In `final_rev_info_setup()`, add your filter line: ---- static void final_rev_info_setup(int argc, const char **argv, const char *prefix, struct rev_info *rev) { ... append_header_grep_pattern(&rev->grep_filter, GREP_HEADER_AUTHOR, "gmail"); compile_grep_patterns(&rev->grep_filter); ... } ---- `append_header_grep_pattern()` adds your new "gmail" pattern to `rev_info`, but it won't work unless we compile it with `compile_grep_patterns()`. NOTE: If you are using `setup_revisions()` (for example, if you are passing a `setup_revision_opt` instead of using `add_head_to_pending()`), you don't need to call `compile_grep_patterns()` because `setup_revisions()` calls it for you. NOTE: We could add the same filter via the `append_grep_pattern()` helper if we wanted to, but `append_header_grep_pattern()` adds the `enum grep_context` and `enum grep_pat_token` for us. === Changing the Order There are a few ways that we can change the order of the commits during a revision walk. Firstly, we can use the `enum rev_sort_order` to choose from some typical orderings. `topo_order` is the same as `git log --topo-order`: we avoid showing a parent before all of its children have been shown, and we avoid mixing commits which are in different lines of history. (`git help log`'s section on `--topo-order` has a very nice diagram to illustrate this.) Let's see what happens when we run with `REV_SORT_BY_COMMIT_DATE` as opposed to `REV_SORT_BY_AUTHOR_DATE`. Add the following: ---- static void final_rev_info_setup(int argc, const char **argv, const char *prefix, struct rev_info *rev) { ... rev->topo_order = 1; rev->sort_order = REV_SORT_BY_COMMIT_DATE; ... } ---- Let's output this into a file so we can easily diff it with the walk sorted by author date. ---- $ make $ ./bin-wrappers/git walken > commit-date.txt ---- Then, let's sort by author date and run it again. ---- static void final_rev_info_setup(int argc, const char **argv, const char *prefix, struct rev_info *rev) { ... rev->topo_order = 1; rev->sort_order = REV_SORT_BY_AUTHOR_DATE; ... } ---- ---- $ make $ ./bin-wrappers/git walken > author-date.txt ---- Finally, compare the two. This is a little less helpful without object names or dates, but hopefully we get the idea. ---- $ diff -u commit-date.txt author-date.txt ---- This display indicates that commits can be reordered after they're written, for example with `git rebase`. Let's try one more reordering of commits. `rev_info` exposes a `reverse` flag. Set that flag somewhere inside of `final_rev_info_setup()`: ---- static void final_rev_info_setup(int argc, const char **argv, const char *prefix, struct rev_info *rev) { ... rev->reverse = 1; ... } ---- Run your walk again and note the difference in order. (If you remove the grep pattern, you should see the last commit this call gives you as your current HEAD.) == Basic Object Walk So far we've been walking only commits. But Git has more types of objects than that! Let's see if we can walk _all_ objects, and find out some information about each one. We can base our work on an example. `git pack-objects` prepares all kinds of objects for packing into a bitmap or packfile. The work we are interested in resides in `builtin/pack-objects.c:get_object_list()`; examination of that function shows that the all-object walk is being performed by `traverse_commit_list()` or `traverse_commit_list_filtered()`. Those two functions reside in `list-objects.c`; examining the source shows that, despite the name, these functions traverse all kinds of objects. Let's have a look at the arguments to `traverse_commit_list()`. - `struct rev_info *revs`: This is the `rev_info` used for the walk. If its `filter` member is not `NULL`, then `filter` contains information for how to filter the object list. - `show_commit_fn show_commit`: A callback which will be used to handle each individual commit object. - `show_object_fn show_object`: A callback which will be used to handle each non-commit object (so each blob, tree, or tag). - `void *show_data`: A context buffer which is passed in turn to `show_commit` and `show_object`. In addition, `traverse_commit_list_filtered()` has an additional parameter: - `struct oidset *omitted`: A linked-list of object IDs which the provided filter caused to be omitted. It looks like these methods use callbacks we provide instead of needing us to call it repeatedly ourselves. Cool! Let's add the callbacks first. For the sake of this tutorial, we'll simply keep track of how many of each kind of object we find. At file scope in `builtin/walken.c` add the following tracking variables: ---- static int commit_count; static int tag_count; static int blob_count; static int tree_count; ---- Commits are handled by a different callback than other objects; let's do that one first: ---- static void walken_show_commit(struct commit *cmt, void *buf) { commit_count++; } ---- The `cmt` argument is fairly self-explanatory. But it's worth mentioning that the `buf` argument is actually the context buffer that we can provide to the traversal calls - `show_data`, which we mentioned a moment ago. Since we have the `struct commit` object, we can look at all the same parts that we looked at in our earlier commit-only walk. For the sake of this tutorial, though, we'll just increment the commit counter and move on. The callback for non-commits is a little different, as we'll need to check which kind of object we're dealing with: ---- static void walken_show_object(struct object *obj, const char *str, void *buf) { switch (obj->type) { case OBJ_TREE: tree_count++; break; case OBJ_BLOB: blob_count++; break; case OBJ_TAG: tag_count++; break; case OBJ_COMMIT: BUG("unexpected commit object in walken_show_object\n"); default: BUG("unexpected object type %s in walken_show_object\n", type_name(obj->type)); } } ---- Again, `obj` is fairly self-explanatory, and we can guess that `buf` is the same context pointer that `walken_show_commit()` receives: the `show_data` argument to `traverse_commit_list()` and `traverse_commit_list_filtered()`. Finally, `str` contains the name of the object, which ends up being something like `foo.txt` (blob), `bar/baz` (tree), or `v1.2.3` (tag). To help assure us that we aren't double-counting commits, we'll include some complaining if a commit object is routed through our non-commit callback; we'll also complain if we see an invalid object type. Since those two cases should be unreachable, and would only change in the event of a semantic change to the Git codebase, we complain by using `BUG()` - which is a signal to a developer that the change they made caused unintended consequences, and the rest of the codebase needs to be updated to understand that change. `BUG()` is not intended to be seen by the public, so it is not localized. Our main object walk implementation is substantially different from our commit walk implementation, so let's make a new function to perform the object walk. We can perform setup which is applicable to all objects here, too, to keep separate from setup which is applicable to commit-only walks. We'll start by enabling all types of objects in the `struct rev_info`. We'll also turn on `tree_blobs_in_commit_order`, which means that we will walk a commit's tree and everything it points to immediately after we find each commit, as opposed to waiting for the end and walking through all trees after the commit history has been discovered. With the appropriate settings configured, we are ready to call `prepare_revision_walk()`. ---- static void walken_object_walk(struct rev_info *rev) { rev->tree_objects = 1; rev->blob_objects = 1; rev->tag_objects = 1; rev->tree_blobs_in_commit_order = 1; if (prepare_revision_walk(rev)) die(_("revision walk setup failed")); commit_count = 0; tag_count = 0; blob_count = 0; tree_count = 0; ---- Let's start by calling just the unfiltered walk and reporting our counts. Complete your implementation of `walken_object_walk()`. We'll also need to include the `list-objects.h` header. ---- #include "list-objects.h" ... traverse_commit_list(rev, walken_show_commit, walken_show_object, NULL); printf("commits %d\nblobs %d\ntags %d\ntrees %d\n", commit_count, blob_count, tag_count, tree_count); } ---- NOTE: This output is intended to be machine-parsed. Therefore, we are not sending it to `trace_printf()`, and we are not localizing it - we need scripts to be able to count on the formatting to be exactly the way it is shown here. If we were intending this output to be read by humans, we would need to localize it with `_()`. Finally, we'll ask `cmd_walken()` to use the object walk instead. Discussing command line options is out of scope for this tutorial, so we'll just hardcode a branch we can change at compile time. Where you call `final_rev_info_setup()` and `walken_commit_walk()`, instead branch like so: ---- if (1) { add_head_to_pending(&rev); walken_object_walk(&rev); } else { final_rev_info_setup(argc, argv, prefix, &rev); walken_commit_walk(&rev); } ---- NOTE: For simplicity, we've avoided all the filters and sorts we applied in `final_rev_info_setup()` and simply added `HEAD` to our pending queue. If you want, you can certainly use the filters we added before by moving `final_rev_info_setup()` out of the conditional and removing the call to `add_head_to_pending()`. Now we can try to run our command! It should take noticeably longer than the commit walk, but an examination of the output will give you an idea why. Your output should look similar to this example, but with different counts: ---- Object walk completed. Found 55733 commits, 100274 blobs, 0 tags, and 104210 trees. ---- This makes sense. We have more trees than commits because the Git project has lots of subdirectories which can change, plus at least one tree per commit. We have no tags because we started on a commit (`HEAD`) and while tags can point to commits, commits can't point to tags. NOTE: You will have different counts when you run this yourself! The number of objects grows along with the Git project. === Adding a Filter There are a handful of filters that we can apply to the object walk laid out in `Documentation/rev-list-options.adoc`. These filters are typically useful for operations such as creating packfiles or performing a partial clone. They are defined in `list-objects-filter-options.h`. For the purposes of this tutorial we will use the "tree:1" filter, which causes the walk to omit all trees and blobs which are not directly referenced by commits reachable from the commit in `pending` when the walk begins. (`pending` is the list of objects which need to be traversed during a walk; you can imagine a breadth-first tree traversal to help understand. In our case, that means we omit trees and blobs not directly referenced by `HEAD` or `HEAD`'s history, because we begin the walk with only `HEAD` in the `pending` list.) For now, we are not going to track the omitted objects, so we'll replace those parameters with `NULL`. For the sake of simplicity, we'll add a simple build-time branch to use our filter or not. Preface the line calling `traverse_commit_list()` with the following, which will remind us which kind of walk we've just performed: ---- if (0) { /* Unfiltered: */ trace_printf(_("Unfiltered object walk.\n")); } else { trace_printf( _("Filtered object walk with filterspec 'tree:1'.\n")); parse_list_objects_filter(&rev->filter, "tree:1"); } traverse_commit_list(rev, walken_show_commit, walken_show_object, NULL); ---- The `rev->filter` member is usually built directly from a command line argument, so the module provides an easy way to build one from a string. Even though we aren't taking user input right now, we can still build one with a hardcoded string using `parse_list_objects_filter()`. With the filter spec "tree:1", we are expecting to see _only_ the root tree for each commit; therefore, the tree object count should be less than or equal to the number of commits. (For an example of why that's true: `git commit --revert` points to the same tree object as its grandparent.) === Counting Omitted Objects We also have the capability to enumerate all objects which were omitted by a filter, like with `git log --filter= --filter-print-omitted`. To do this, change `traverse_commit_list()` to `traverse_commit_list_filtered()`, which is able to populate an `omitted` list. Asking for this list of filtered objects may cause performance degradations, however, because in this case, despite filtering objects, the possibly much larger set of all reachable objects must be processed in order to populate that list. First, add the `struct oidset` and related items we will use to iterate it: ---- #include "oidset.h" ... static void walken_object_walk( ... struct oidset omitted; struct oidset_iter oit; struct object_id *oid = NULL; int omitted_count = 0; oidset_init(&omitted, 0); ... ---- Replace the call to `traverse_commit_list()` with `traverse_commit_list_filtered()` and pass a pointer to the `omitted` oidset defined and initialized above: ---- ... traverse_commit_list_filtered(rev, walken_show_commit, walken_show_object, NULL, &omitted); ... ---- Then, after your traversal, the `oidset` traversal is pretty straightforward. Count all the objects within and modify the print statement: ---- /* Count the omitted objects. */ oidset_iter_init(&omitted, &oit); while ((oid = oidset_iter_next(&oit))) omitted_count++; printf("commits %d\nblobs %d\ntags %d\ntrees %d\nomitted %d\n", commit_count, blob_count, tag_count, tree_count, omitted_count); ---- By running your walk with and without the filter, you should find that the total object count in each case is identical. You can also time each invocation of the `walken` subcommand, with and without `omitted` being passed in, to confirm to yourself the runtime impact of tracking all omitted objects. === Changing the Order Finally, let's demonstrate that you can also reorder walks of all objects, not just walks of commits. First, we'll make our handlers chattier - modify `walken_show_commit()` and `walken_show_object()` to print the object as they go: ---- #include "hex.h" ... static void walken_show_commit(struct commit *cmt, void *buf) { trace_printf("commit: %s\n", oid_to_hex(&cmt->object.oid)); commit_count++; } static void walken_show_object(struct object *obj, const char *str, void *buf) { trace_printf("%s: %s\n", type_name(obj->type), oid_to_hex(&obj->oid)); ... } ---- NOTE: Since we will be examining this output directly as humans, we'll use `trace_printf()` here. Additionally, since this change introduces a significant number of printed lines, using `trace_printf()` will allow us to easily silence those lines without having to recompile. (Leave the counter increment logic in place.) With only that change, run again (but save yourself some scrollback): ---- $ GIT_TRACE=1 ./bin-wrappers/git walken 2>&1 | head -n 10 ---- Take a look at the top commit with `git show` and the object ID you printed; it should be the same as the output of `git show HEAD`. Next, let's change a setting on our `struct rev_info` within `walken_object_walk()`. Find where you're changing the other settings on `rev`, such as `rev->tree_objects` and `rev->tree_blobs_in_commit_order`, and add the `reverse` setting at the bottom: ---- ... rev->tree_objects = 1; rev->blob_objects = 1; rev->tag_objects = 1; rev->tree_blobs_in_commit_order = 1; rev->reverse = 1; ... ---- Now, run again, but this time, let's grab the last handful of objects instead of the first handful: ---- $ make $ GIT_TRACE=1 ./bin-wrappers/git walken 2>&1 | tail -n 10 ---- The last commit object given should have the same OID as the one we saw at the top before, and running `git show ` with that OID should give you again the same results as `git show HEAD`. Furthermore, if you run and examine the first ten lines again (with `head` instead of `tail` like we did before applying the `reverse` setting), you should see that now the first commit printed is the initial commit, `e83c5163`. == Wrapping Up Let's review. In this tutorial, we: - Built a commit walk from the ground up - Enabled a grep filter for that commit walk - Changed the sort order of that filtered commit walk - Built an object walk (tags, commits, trees, and blobs) from the ground up - Learned how to add a filter-spec to an object walk - Changed the display order of the filtered object walk ./MyFirstObjectWalk.html000066400000000000000000002205661504515377700154750ustar00rootroot00000000000000 My First Object Walk

What’s an Object Walk?

The object walk is a key concept in Git - this is the process that underpins operations like object transfer and fsck. Beginning from a given commit, the list of objects is found by walking parent relationships between commits (commit X based on commit W) and containment relationships between objects (tree Y is contained within commit X, and blob Z is located within tree Y, giving our working tree for commit X something like y/z.txt).

A related concept is the revision walk, which is focused on commit objects and their parent relationships and does not delve into other object types. The revision walk is used for operations like git log.

Related Reading

  • Documentation/user-manual.adoc under "Hacking Git" contains some coverage of the revision walker in its various incarnations.

  • revision.h

  • Git for Computer Scientists gives a good overview of the types of objects in Git and what your object walk is really describing.

Setting Up

Create a new branch from master.

git checkout -b revwalk origin/master

We’ll put our fiddling into a new command. For fun, let’s name it git walken. Open up a new file builtin/walken.c and set up the command handler:

/*
 * "git walken"
 *
 * Part of the "My First Object Walk" tutorial.
 */

#include "builtin.h"
#include "trace.h"

int cmd_walken(int argc, const char **argv, const char *prefix, struct repository *repo)
{
        trace_printf(_("cmd_walken incoming...\n"));
        return 0;
}
Note
 trace_printf(), defined in trace.h, differs from printf() in that it can be turned on or off at runtime. For the purposes of this tutorial, we will write walken as though it is intended for use as a "plumbing" command: that is, a command which is used primarily in scripts, rather than interactively by humans (a "porcelain" command). So we will send our debug output to trace_printf() instead. When running, enable trace output by setting the environment variable GIT_TRACE.

Add usage text and -h handling, like all subcommands should consistently do (our test suite will notice and complain if you fail to do so). We’ll need to include the parse-options.h header.

#include "parse-options.h"

...

int cmd_walken(int argc, const char **argv, const char *prefix)
{
        const char * const walken_usage[] = {
                N_("git walken"),
                NULL,
        };
        struct option options[] = {
                OPT_END()
        };

        argc = parse_options(argc, argv, prefix, options, walken_usage, 0);

        ...
}

Also add the relevant line in builtin.h near cmd_version():

int cmd_walken(int argc, const char **argv, const char *prefix, struct repository *repo);

Include the command in git.c in commands[] near the entry for version, maintaining alphabetical ordering:

{ "walken", cmd_walken, RUN_SETUP },

Add an entry for the new command in the both the Make and Meson build system, before the entry for worktree:

  • In the Makefile:

...
BUILTIN_OBJS += builtin/walken.o
...

  • In the meson.build file:

builtin_sources = [
   ...
  'builtin/walken.c',
   ...
]

Build and test out your command, without forgetting to ensure the DEVELOPER flag is set, and with GIT_TRACE enabled so the debug output can be seen:

$ echo DEVELOPER=1 >>config.mak
$ make
$ GIT_TRACE=1 ./bin-wrappers/git walken
Note
 For a more exhaustive overview of the new command process, take a look at Documentation/MyFirstContribution.adoc.
Note
 A reference implementation can be found at https://github.com/nasamuffin/git/tree/revwalk.

struct rev_cmdline_info

The definition of struct rev_cmdline_info can be found in revision.h.

This struct is contained within the rev_info struct and is used to reflect parameters provided by the user over the CLI.

nr represents the number of rev_cmdline_entry present in the array.

alloc is used by the ALLOC_GROW macro. Check alloc.h - this variable is used to track the allocated size of the list.

Per entry, we find:

item is the object provided upon which to base the object walk. Items in Git can be blobs, trees, commits, or tags. (See Documentation/gittutorial-2.adoc.)

name is the object ID (OID) of the object - a hex string you may be familiar with from using Git to organize your source in the past. Check the tutorial mentioned above towards the top for a discussion of where the OID can come from.

whence indicates some information about what to do with the parents of the specified object. We’ll explore this flag more later on; take a look at Documentation/revisions.adoc to get an idea of what could set the whence value.

flags are used to hint the beginning of the revision walk and are the first block under the #include`s in `revision.h. The most likely ones to be set in the rev_cmdline_info are UNINTERESTING and BOTTOM, but these same flags can be used during the walk, as well.

struct rev_info

This one is quite a bit longer, and many fields are only used during the walk by revision.c - not configuration options. Most of the configurable flags in struct rev_info have a mirror in Documentation/rev-list-options.adoc. It’s a good idea to take some time and read through that document.

Basic Commit Walk

First, let’s see if we can replicate the output of git log --oneline. We’ll refer back to the implementation frequently to discover norms when performing an object walk of our own.

To do so, we’ll first find all the commits, in order, which preceded the current commit. We’ll extract the name and subject of the commit from each.

Ideally, we will also be able to find out which ones are currently at the tip of various branches.

Setting Up

Preparing for your object walk has some distinct stages.

  1. Perform default setup for this mode, and others which may be invoked.

  2. Check configuration files for relevant settings.

  3. Set up the rev_info struct.

  4. Tweak the initialized rev_info to suit the current walk.

  5. Prepare the rev_info for the walk.

  6. Iterate over the objects, processing each one.

Default Setups

Before examining configuration files which may modify command behavior, set up default state for switches or options your command may have. If your command utilizes other Git components, ask them to set up their default states as well. For instance, git log takes advantage of grep and diff functionality, so its init_log_defaults() sets its own state (decoration_style) and asks grep and diff to initialize themselves by calling each of their initialization functions.

Configuring From .gitconfig

Next, we should have a look at any relevant configuration settings (i.e., settings readable and settable from git config). This is done by providing a callback to repo_config(); within that callback, you can also invoke methods from other components you may need that need to intercept these options. Your callback will be invoked once per each configuration value which Git knows about (global, local, worktree, etc.).

Similarly to the default values, we don’t have anything to do here yet ourselves; however, we should call git_default_config() if we aren’t calling any other existing config callbacks.

Add a new function to builtin/walken.c. We’ll also need to include the config.h header:

#include "config.h"

...

static int git_walken_config(const char *var, const char *value,
                             const struct config_context *ctx, void *cb)
{
        /*
         * For now, we don't have any custom configuration, so fall back to
         * the default config.
         */
        return git_default_config(var, value, ctx, cb);
}

Make sure to invoke repo_config() with it in your cmd_walken():

int cmd_walken(int argc, const char **argv, const char *prefix, struct repository *repo)
{
        ...

        repo_config(repo, git_walken_config, NULL);

        ...
}

Setting Up rev_info

Now that we’ve gathered external configuration and options, it’s time to initialize the rev_info object which we will use to perform the walk. This is typically done by calling repo_init_revisions() with the repository you intend to target, as well as the prefix argument of cmd_walken and your rev_info struct.

Add the struct rev_info and the repo_init_revisions() call. We’ll also need to include the revision.h header:

#include "revision.h"

...

int cmd_walken(int argc, const char **argv, const char *prefix, struct repository *repo)
{
        /* This can go wherever you like in your declarations.*/
        struct rev_info rev;
        ...

        /* This should go after the repo_config() call. */
        repo_init_revisions(repo, &rev, prefix);

        ...
}

Tweaking rev_info For the Walk

We’re getting close, but we’re still not quite ready to go. Now that rev is initialized, we can modify it to fit our needs. This is usually done within a helper for clarity, so let’s add one:

static void final_rev_info_setup(struct rev_info *rev)
{
        /*
         * We want to mimic the appearance of `git log --oneline`, so let's
         * force oneline format.
         */
        get_commit_format("oneline", rev);

        /* Start our object walk at HEAD. */
        add_head_to_pending(rev);
}
Note

Instead of using the shorthand add_head_to_pending(), you could do something like this:

        struct setup_revision_opt opt;

        memset(&opt, 0, sizeof(opt));
        opt.def = "HEAD";
        opt.revarg_opt = REVARG_COMMITTISH;
        setup_revisions(argc, argv, rev, &opt);

Using a setup_revision_opt gives you finer control over your walk’s starting point.

Then let’s invoke final_rev_info_setup() after the call to repo_init_revisions():

int cmd_walken(int argc, const char **argv, const char *prefix, struct repository *repo)
{
        ...

        final_rev_info_setup(&rev);

        ...
}

Later, we may wish to add more arguments to final_rev_info_setup(). But for now, this is all we need.

Preparing rev_info For the Walk

Now that rev is all initialized and configured, we’ve got one more setup step before we get rolling. We can do this in a helper, which will both prepare the rev_info for the walk, and perform the walk itself. Let’s start the helper with the call to prepare_revision_walk(), which can return an error without dying on its own:

static void walken_commit_walk(struct rev_info *rev)
{
        if (prepare_revision_walk(rev))
                die(_("revision walk setup failed"));
}
Note
 die() prints to stderr and exits the program. Since it will print to stderr it’s likely to be seen by a human, so we will localize it.

Performing the Walk!

Finally! We are ready to begin the walk itself. Now we can see that rev_info can also be used as an iterator; we move to the next item in the walk by using get_revision() repeatedly. Add the listed variable declarations at the top and the walk loop below the prepare_revision_walk() call within your walken_commit_walk():

#include "pretty.h"

...

static void walken_commit_walk(struct rev_info *rev)
{
        struct commit *commit;
        struct strbuf prettybuf = STRBUF_INIT;

        ...

        while ((commit = get_revision(rev))) {
                strbuf_reset(&prettybuf);
                pp_commit_easy(CMIT_FMT_ONELINE, commit, &prettybuf);
                puts(prettybuf.buf);
        }
        strbuf_release(&prettybuf);
}
Note
 puts() prints a char* to stdout. Since this is the part of the command we expect to be machine-parsed, we’re sending it directly to stdout.

Give it a shot.

$ make
$ ./bin-wrappers/git walken

You should see all of the subject lines of all the commits in your tree’s history, in order, ending with the initial commit, "Initial revision of "git", the information manager from hell". Congratulations! You’ve written your first revision walk. You can play with printing some additional fields from each commit if you’re curious; have a look at the functions available in commit.h.

Adding a Filter

Next, let’s try to filter the commits we see based on their author. This is equivalent to running git log --author=<pattern>. We can add a filter by modifying rev_info.grep_filter, which is a struct grep_opt.

First some setup. Add grep_config() to git_walken_config():

static int git_walken_config(const char *var, const char *value,
                             const struct config_context *ctx, void *cb)
{
        grep_config(var, value, ctx, cb);
        return git_default_config(var, value, ctx, cb);
}

Next, we can modify the grep_filter. This is done with convenience functions found in grep.h. For fun, we’re filtering to only commits from folks using a gmail.com email address - a not-very-precise guess at who may be working on Git as a hobby. Since we’re checking the author, which is a specific line in the header, we’ll use the append_header_grep_pattern() helper. We can use the enum grep_header_field to indicate which part of the commit header we want to search.

In final_rev_info_setup(), add your filter line:

static void final_rev_info_setup(int argc, const char **argv,
                const char *prefix, struct rev_info *rev)
{
        ...

        append_header_grep_pattern(&rev->grep_filter, GREP_HEADER_AUTHOR,
                "gmail");
        compile_grep_patterns(&rev->grep_filter);

        ...
}

append_header_grep_pattern() adds your new "gmail" pattern to rev_info, but it won’t work unless we compile it with compile_grep_patterns().

Note
 If you are using setup_revisions() (for example, if you are passing a setup_revision_opt instead of using add_head_to_pending()), you don’t need to call compile_grep_patterns() because setup_revisions() calls it for you.
Note
 We could add the same filter via the append_grep_pattern() helper if we wanted to, but append_header_grep_pattern() adds the enum grep_context and enum grep_pat_token for us.

Changing the Order

There are a few ways that we can change the order of the commits during a revision walk. Firstly, we can use the enum rev_sort_order to choose from some typical orderings.

topo_order is the same as git log --topo-order: we avoid showing a parent before all of its children have been shown, and we avoid mixing commits which are in different lines of history. (git help log's section on --topo-order has a very nice diagram to illustrate this.)

Let’s see what happens when we run with REV_SORT_BY_COMMIT_DATE as opposed to REV_SORT_BY_AUTHOR_DATE. Add the following:

static void final_rev_info_setup(int argc, const char **argv,
                const char *prefix, struct rev_info *rev)
{
        ...

        rev->topo_order = 1;
        rev->sort_order = REV_SORT_BY_COMMIT_DATE;

        ...
}

Let’s output this into a file so we can easily diff it with the walk sorted by author date.

$ make
$ ./bin-wrappers/git walken > commit-date.txt

Then, let’s sort by author date and run it again.

static void final_rev_info_setup(int argc, const char **argv,
                const char *prefix, struct rev_info *rev)
{
        ...

        rev->topo_order = 1;
        rev->sort_order = REV_SORT_BY_AUTHOR_DATE;

        ...
}
$ make
$ ./bin-wrappers/git walken > author-date.txt

Finally, compare the two. This is a little less helpful without object names or dates, but hopefully we get the idea.

$ diff -u commit-date.txt author-date.txt

This display indicates that commits can be reordered after they’re written, for example with git rebase.

Let’s try one more reordering of commits. rev_info exposes a reverse flag. Set that flag somewhere inside of final_rev_info_setup():

static void final_rev_info_setup(int argc, const char **argv, const char *prefix,
                struct rev_info *rev)
{
        ...

        rev->reverse = 1;

        ...
}

Run your walk again and note the difference in order. (If you remove the grep pattern, you should see the last commit this call gives you as your current HEAD.)

Basic Object Walk

So far we’ve been walking only commits. But Git has more types of objects than that! Let’s see if we can walk all objects, and find out some information about each one.

We can base our work on an example. git pack-objects prepares all kinds of objects for packing into a bitmap or packfile. The work we are interested in resides in builtin/pack-objects.c:get_object_list(); examination of that function shows that the all-object walk is being performed by traverse_commit_list() or traverse_commit_list_filtered(). Those two functions reside in list-objects.c; examining the source shows that, despite the name, these functions traverse all kinds of objects. Let’s have a look at the arguments to traverse_commit_list().

  • struct rev_info *revs: This is the rev_info used for the walk. If its filter member is not NULL, then filter contains information for how to filter the object list.

  • show_commit_fn show_commit: A callback which will be used to handle each individual commit object.

  • show_object_fn show_object: A callback which will be used to handle each non-commit object (so each blob, tree, or tag).

  • void *show_data: A context buffer which is passed in turn to show_commit and show_object.

In addition, traverse_commit_list_filtered() has an additional parameter:

  • struct oidset *omitted: A linked-list of object IDs which the provided filter caused to be omitted.

It looks like these methods use callbacks we provide instead of needing us to call it repeatedly ourselves. Cool! Let’s add the callbacks first.

For the sake of this tutorial, we’ll simply keep track of how many of each kind of object we find. At file scope in builtin/walken.c add the following tracking variables:

static int commit_count;
static int tag_count;
static int blob_count;
static int tree_count;

Commits are handled by a different callback than other objects; let’s do that one first:

static void walken_show_commit(struct commit *cmt, void *buf)
{
        commit_count++;
}

The cmt argument is fairly self-explanatory. But it’s worth mentioning that the buf argument is actually the context buffer that we can provide to the traversal calls - show_data, which we mentioned a moment ago.

Since we have the struct commit object, we can look at all the same parts that we looked at in our earlier commit-only walk. For the sake of this tutorial, though, we’ll just increment the commit counter and move on.

The callback for non-commits is a little different, as we’ll need to check which kind of object we’re dealing with:

static void walken_show_object(struct object *obj, const char *str, void *buf)
{
        switch (obj->type) {
        case OBJ_TREE:
                tree_count++;
                break;
        case OBJ_BLOB:
                blob_count++;
                break;
        case OBJ_TAG:
                tag_count++;
                break;
        case OBJ_COMMIT:
                BUG("unexpected commit object in walken_show_object\n");
        default:
                BUG("unexpected object type %s in walken_show_object\n",
                        type_name(obj->type));
        }
}

Again, obj is fairly self-explanatory, and we can guess that buf is the same context pointer that walken_show_commit() receives: the show_data argument to traverse_commit_list() and traverse_commit_list_filtered(). Finally, str contains the name of the object, which ends up being something like foo.txt (blob), bar/baz (tree), or v1.2.3 (tag).

To help assure us that we aren’t double-counting commits, we’ll include some complaining if a commit object is routed through our non-commit callback; we’ll also complain if we see an invalid object type. Since those two cases should be unreachable, and would only change in the event of a semantic change to the Git codebase, we complain by using BUG() - which is a signal to a developer that the change they made caused unintended consequences, and the rest of the codebase needs to be updated to understand that change. BUG() is not intended to be seen by the public, so it is not localized.

Our main object walk implementation is substantially different from our commit walk implementation, so let’s make a new function to perform the object walk. We can perform setup which is applicable to all objects here, too, to keep separate from setup which is applicable to commit-only walks.

We’ll start by enabling all types of objects in the struct rev_info. We’ll also turn on tree_blobs_in_commit_order, which means that we will walk a commit’s tree and everything it points to immediately after we find each commit, as opposed to waiting for the end and walking through all trees after the commit history has been discovered. With the appropriate settings configured, we are ready to call prepare_revision_walk().

static void walken_object_walk(struct rev_info *rev)
{
        rev->tree_objects = 1;
        rev->blob_objects = 1;
        rev->tag_objects = 1;
        rev->tree_blobs_in_commit_order = 1;

        if (prepare_revision_walk(rev))
                die(_("revision walk setup failed"));

        commit_count = 0;
        tag_count = 0;
        blob_count = 0;
        tree_count = 0;

Let’s start by calling just the unfiltered walk and reporting our counts. Complete your implementation of walken_object_walk(). We’ll also need to include the list-objects.h header.

#include "list-objects.h"

...

        traverse_commit_list(rev, walken_show_commit, walken_show_object, NULL);

        printf("commits %d\nblobs %d\ntags %d\ntrees %d\n", commit_count,
                blob_count, tag_count, tree_count);
}
Note
 This output is intended to be machine-parsed. Therefore, we are not sending it to trace_printf(), and we are not localizing it - we need scripts to be able to count on the formatting to be exactly the way it is shown here. If we were intending this output to be read by humans, we would need to localize it with _().

Finally, we’ll ask cmd_walken() to use the object walk instead. Discussing command line options is out of scope for this tutorial, so we’ll just hardcode a branch we can change at compile time. Where you call final_rev_info_setup() and walken_commit_walk(), instead branch like so:

        if (1) {
                add_head_to_pending(&rev);
                walken_object_walk(&rev);
        } else {
                final_rev_info_setup(argc, argv, prefix, &rev);
                walken_commit_walk(&rev);
        }
Note
 For simplicity, we’ve avoided all the filters and sorts we applied in final_rev_info_setup() and simply added HEAD to our pending queue. If you want, you can certainly use the filters we added before by moving final_rev_info_setup() out of the conditional and removing the call to add_head_to_pending().

Now we can try to run our command! It should take noticeably longer than the commit walk, but an examination of the output will give you an idea why. Your output should look similar to this example, but with different counts:

Object walk completed. Found 55733 commits, 100274 blobs, 0 tags, and 104210 trees.

This makes sense. We have more trees than commits because the Git project has lots of subdirectories which can change, plus at least one tree per commit. We have no tags because we started on a commit (HEAD) and while tags can point to commits, commits can’t point to tags.

Note
 You will have different counts when you run this yourself! The number of objects grows along with the Git project.

Adding a Filter

There are a handful of filters that we can apply to the object walk laid out in Documentation/rev-list-options.adoc. These filters are typically useful for operations such as creating packfiles or performing a partial clone. They are defined in list-objects-filter-options.h. For the purposes of this tutorial we will use the "tree:1" filter, which causes the walk to omit all trees and blobs which are not directly referenced by commits reachable from the commit in pending when the walk begins. (pending is the list of objects which need to be traversed during a walk; you can imagine a breadth-first tree traversal to help understand. In our case, that means we omit trees and blobs not directly referenced by HEAD or HEAD's history, because we begin the walk with only HEAD in the pending list.)

For now, we are not going to track the omitted objects, so we’ll replace those parameters with NULL. For the sake of simplicity, we’ll add a simple build-time branch to use our filter or not. Preface the line calling traverse_commit_list() with the following, which will remind us which kind of walk we’ve just performed:

        if (0) {
                /* Unfiltered: */
                trace_printf(_("Unfiltered object walk.\n"));
        } else {
                trace_printf(
                        _("Filtered object walk with filterspec 'tree:1'.\n"));

                parse_list_objects_filter(&rev->filter, "tree:1");
        }
        traverse_commit_list(rev, walken_show_commit,
                             walken_show_object, NULL);

The rev->filter member is usually built directly from a command line argument, so the module provides an easy way to build one from a string. Even though we aren’t taking user input right now, we can still build one with a hardcoded string using parse_list_objects_filter().

With the filter spec "tree:1", we are expecting to see only the root tree for each commit; therefore, the tree object count should be less than or equal to the number of commits. (For an example of why that’s true: git commit --revert points to the same tree object as its grandparent.)

Counting Omitted Objects

We also have the capability to enumerate all objects which were omitted by a filter, like with git log --filter=<spec> --filter-print-omitted. To do this, change traverse_commit_list() to traverse_commit_list_filtered(), which is able to populate an omitted list. Asking for this list of filtered objects may cause performance degradations, however, because in this case, despite filtering objects, the possibly much larger set of all reachable objects must be processed in order to populate that list.

First, add the struct oidset and related items we will use to iterate it:

#include "oidset.h"

...

static void walken_object_walk(
        ...

        struct oidset omitted;
        struct oidset_iter oit;
        struct object_id *oid = NULL;
        int omitted_count = 0;
        oidset_init(&omitted, 0);

        ...

Replace the call to traverse_commit_list() with traverse_commit_list_filtered() and pass a pointer to the omitted oidset defined and initialized above:

        ...

                traverse_commit_list_filtered(rev,
                        walken_show_commit, walken_show_object, NULL, &omitted);

        ...

Then, after your traversal, the oidset traversal is pretty straightforward. Count all the objects within and modify the print statement:

        /* Count the omitted objects. */
        oidset_iter_init(&omitted, &oit);

        while ((oid = oidset_iter_next(&oit)))
                omitted_count++;

        printf("commits %d\nblobs %d\ntags %d\ntrees %d\nomitted %d\n",
                commit_count, blob_count, tag_count, tree_count, omitted_count);

By running your walk with and without the filter, you should find that the total object count in each case is identical. You can also time each invocation of the walken subcommand, with and without omitted being passed in, to confirm to yourself the runtime impact of tracking all omitted objects.

Changing the Order

Finally, let’s demonstrate that you can also reorder walks of all objects, not just walks of commits. First, we’ll make our handlers chattier - modify walken_show_commit() and walken_show_object() to print the object as they go:

#include "hex.h"

...

static void walken_show_commit(struct commit *cmt, void *buf)
{
        trace_printf("commit: %s\n", oid_to_hex(&cmt->object.oid));
        commit_count++;
}

static void walken_show_object(struct object *obj, const char *str, void *buf)
{
        trace_printf("%s: %s\n", type_name(obj->type), oid_to_hex(&obj->oid));

        ...
}
Note
 Since we will be examining this output directly as humans, we’ll use trace_printf() here. Additionally, since this change introduces a significant number of printed lines, using trace_printf() will allow us to easily silence those lines without having to recompile.

(Leave the counter increment logic in place.)

With only that change, run again (but save yourself some scrollback):

$ GIT_TRACE=1 ./bin-wrappers/git walken 2>&1 | head -n 10

Take a look at the top commit with git show and the object ID you printed; it should be the same as the output of git show HEAD.

Next, let’s change a setting on our struct rev_info within walken_object_walk(). Find where you’re changing the other settings on rev, such as rev->tree_objects and rev->tree_blobs_in_commit_order, and add the reverse setting at the bottom:

        ...

        rev->tree_objects = 1;
        rev->blob_objects = 1;
        rev->tag_objects = 1;
        rev->tree_blobs_in_commit_order = 1;
        rev->reverse = 1;

        ...

Now, run again, but this time, let’s grab the last handful of objects instead of the first handful:

$ make
$ GIT_TRACE=1 ./bin-wrappers/git walken 2>&1 | tail -n 10

The last commit object given should have the same OID as the one we saw at the top before, and running git show <oid> with that OID should give you again the same results as git show HEAD. Furthermore, if you run and examine the first ten lines again (with head instead of tail like we did before applying the reverse setting), you should see that now the first commit printed is the initial commit, e83c5163.

Wrapping Up

Let’s review. In this tutorial, we:

  • Built a commit walk from the ground up

  • Enabled a grep filter for that commit walk

  • Changed the sort order of that filtered commit walk

  • Built an object walk (tags, commits, trees, and blobs) from the ground up

  • Learned how to add a filter-spec to an object walk

  • Changed the display order of the filtered object walk

./RelNotes/000077500000000000000000000000001504515377700127645ustar00rootroot00000000000000./RelNotes/1.5.0.1.adoc000066400000000000000000000023471504515377700144220ustar00rootroot00000000000000GIT v1.5.0.1 Release Notes ========================== Fixes since v1.5.0 ------------------ * Documentation updates - Clarifications and corrections to 1.5.0 release notes. - The main documentation did not link to git-remote documentation. - Clarified introductory text of git-rebase documentation. - Converted remaining mentions of update-index on Porcelain documents to git-add/git-rm. - Some i18n.* configuration variables were incorrectly described as core.*; fixed. * Bugfixes - git-add and git-update-index on a filesystem on which executable bits are unreliable incorrectly reused st_mode bits even when the path changed between symlink and regular file. - git-daemon marks the listening sockets with FD_CLOEXEC so that it won't be leaked into the children. - segfault from git-blame when the mandatory pathname parameter was missing was fixed; usage() message is given instead. - git-rev-list did not read $GIT_DIR/config file, which means that did not honor i18n.logoutputencoding correctly. * Tweaks - sliding mmap() inefficiently mmaped the same region of a packfile with an access pattern that used objects in the reverse order. This has been made more efficient. ./RelNotes/1.5.0.2.adoc000066400000000000000000000044041504515377700144170ustar00rootroot00000000000000GIT v1.5.0.2 Release Notes ========================== Fixes since v1.5.0.1 -------------------- * Bugfixes - Automated merge conflict handling when changes to symbolic links conflicted were completely broken. The merge-resolve strategy created a regular file with conflict markers in it in place of the symbolic link. The default strategy, merge-recursive was even more broken. It removed the path that was pointed at by the symbolic link. Both of these problems have been fixed. - 'git diff maint master next' did not correctly give combined diff across three trees. - 'git fast-import' portability fix for Solaris. - 'git show-ref --verify' without arguments did not error out but segfaulted. - 'git diff :tracked-file `pwd`/an-untracked-file' gave an extra slashes after a/ and b/. - 'git format-patch' produced too long filenames if the commit message had too long line at the beginning. - Running 'make all' and then without changing anything running 'make install' still rebuilt some files. This was inconvenient when building as yourself and then installing as root (especially problematic when the source directory is on NFS and root is mapped to nobody). - 'git-rerere' failed to deal with two unconflicted paths that sorted next to each other. - 'git-rerere' attempted to open(2) a symlink and failed if there was a conflict. Since a conflicting change to a symlink would not benefit from rerere anyway, the command now ignores conflicting changes to symlinks. - 'git-repack' did not like to pass more than 64 arguments internally to underlying 'rev-list' logic, which made it impossible to repack after accumulating many (small) packs in the repository. - 'git-diff' to review the combined diff during a conflicted merge were not reading the working tree version correctly when changes to a symbolic link conflicted. It should have read the data using readlink(2) but read from the regular file the symbolic link pointed at. - 'git-remote' did not like period in a remote's name. * Documentation updates - added and clarified core.bare, core.legacyheaders configurations. - updated "git-clone --depth" documentation. * Assorted git-gui fixes. ./RelNotes/1.5.0.3.adoc000066400000000000000000000030511504515377700144150ustar00rootroot00000000000000GIT v1.5.0.3 Release Notes ========================== Fixes since v1.5.0.2 -------------------- * Bugfixes - 'git.el' honors the commit coding system from the configuration. - 'blameview' in contrib/ correctly digs deeper when a line is clicked. - 'http-push' correctly makes sure the remote side has leading path. Earlier it started in the middle of the path, and incorrectly. - 'git-merge' did not exit with non-zero status when the working tree was dirty and cannot fast forward. It does now. - 'cvsexportcommit' does not lose yet-to-be-used message file. - int-vs-size_t typefix when running combined diff on files over 2GB long. - 'git apply --whitespace=strip' should not touch unmodified lines. - 'git-mailinfo' choke when a logical header line was too long. - 'git show A..B' did not error out. Negative ref ("not A" in this example) does not make sense for the purpose of the command, so now it errors out. - 'git fmt-merge-msg --file' without file parameter did not correctly error out. - 'git archimport' barfed upon encountering a commit without summary. - 'git index-pack' did not protect itself from getting a short read out of pread(2). - 'git http-push' had a few buffer overruns. - Build dependency fixes to rebuild fetch.o when other headers change. * Documentation updates - user-manual updates. - Options to 'git remote add' were described insufficiently. - Configuration format.suffix was not documented. - Other formatting and spelling fixes. ./RelNotes/1.5.0.4.adoc000066400000000000000000000007161504515377700144230ustar00rootroot00000000000000GIT v1.5.0.4 Release Notes ========================== Fixes since v1.5.0.3 -------------------- * Bugfixes - git.el does not add duplicate sign-off lines. - git-commit shows the full stat of the resulting commit, not just about the files in the current directory, when run from a subdirectory. - "git-checkout -m '@{8 hours ago}'" had a funny failure from eval; fixed. - git-gui updates. * Documentation updates * User manual updates ./RelNotes/1.5.0.5.adoc000066400000000000000000000011411504515377700144150ustar00rootroot00000000000000GIT v1.5.0.5 Release Notes ========================== Fixes since v1.5.0.3 -------------------- * Bugfixes - git-merge (hence git-pull) did not refuse fast-forwarding when the working tree had local changes that would have conflicted with it. - git.el does not add duplicate sign-off lines. - git-commit shows the full stat of the resulting commit, not just about the files in the current directory, when run from a subdirectory. - "git-checkout -m '@{8 hours ago}'" had a funny failure from eval; fixed. - git-gui updates. * Documentation updates * User manual updates ./RelNotes/1.5.0.6.adoc000066400000000000000000000007441504515377700144260ustar00rootroot00000000000000GIT v1.5.0.6 Release Notes ========================== Fixes since v1.5.0.5 -------------------- * Bugfixes - a handful small fixes to gitweb. - build procedure for user-manual is fixed not to require locally installed stylesheets. - "git commit $paths" on paths whose earlier contents were already updated in the index were failing out. * Documentation - user-manual has better cross references. - gitweb installation/deployment procedure is now documented. ./RelNotes/1.5.0.7.adoc000066400000000000000000000007421504515377700144250ustar00rootroot00000000000000GIT v1.5.0.7 Release Notes ========================== Fixes since v1.5.0.6 -------------------- * Bugfixes - git-upload-pack failed to close unused pipe ends, resulting in many zombies to hang around. - git-rerere was recording the contents of earlier hunks duplicated in later hunks. This prevented resolving the same conflict when performing the same merge the other way around. * Documentation - a few documentation fixes from Debian package maintainer. ./RelNotes/1.5.0.adoc000066400000000000000000000443161504515377700142650ustar00rootroot00000000000000GIT v1.5.0 Release Notes ======================== Old news -------- This section is for people who are upgrading from ancient versions of git. Although all of the changes in this section happened before the current v1.4.4 release, they are summarized here in the v1.5.0 release notes for people who skipped earlier versions. As of git v1.5.0 there are some optional features that changes the repository to allow data to be stored and transferred more efficiently. These features are not enabled by default, as they will make the repository unusable with older versions of git. Specifically, the available options are: - There is a configuration variable core.legacyheaders that changes the format of loose objects so that they are more efficient to pack and to send out of the repository over git native protocol, since v1.4.2. However, loose objects written in the new format cannot be read by git older than that version; people fetching from your repository using older clients over dumb transports (e.g. http) using older versions of git will also be affected. To let git use the new loose object format, you have to set core.legacyheaders to false. - Since v1.4.3, configuration repack.usedeltabaseoffset allows packfile to be created in more space efficient format, which cannot be read by git older than that version. To let git use the new format for packfiles, you have to set repack.usedeltabaseoffset to true. The above two new features are not enabled by default and you have to explicitly ask for them, because they make repositories unreadable by older versions of git, and in v1.5.0 we still do not enable them by default for the same reason. We will change this default probably 1 year after 1.4.2's release, when it is reasonable to expect everybody to have new enough version of git. - 'git pack-refs' appeared in v1.4.4; this command allows tags to be accessed much more efficiently than the traditional 'one-file-per-tag' format. Older git-native clients can still fetch from a repository that packed and pruned refs (the server side needs to run the up-to-date version of git), but older dumb transports cannot. Packing of refs is done by an explicit user action, either by use of "git pack-refs --prune" command or by use of "git gc" command. - 'git -p' to paginate anything -- many commands do pagination by default on a tty. Introduced between v1.4.1 and v1.4.2; this may surprise old timers. - 'git archive' superseded 'git tar-tree' in v1.4.3; - 'git cvsserver' was new invention in v1.3.0; - 'git repo-config', 'git grep', 'git rebase' and 'gitk' were seriously enhanced during v1.4.0 timeperiod. - 'gitweb' became part of git.git during v1.4.0 timeperiod and seriously modified since then. - reflog is an v1.4.0 invention. This allows you to name a revision that a branch used to be at (e.g. "git diff master@{yesterday} master" allows you to see changes since yesterday's tip of the branch). Updates in v1.5.0 since v1.4.4 series ------------------------------------- * Index manipulation - git-add is to add contents to the index (aka "staging area" for the next commit), whether the file the contents happen to be is an existing one or a newly created one. - git-add without any argument does not add everything anymore. Use 'git-add .' instead. Also you can add otherwise ignored files with an -f option. - git-add tries to be more friendly to users by offering an interactive mode ("git-add -i"). - git-commit used to refuse to commit if was different between HEAD and the index (i.e. update-index was used on it earlier). This check was removed. - git-rm is much saner and safer. It is used to remove paths from both the index file and the working tree, and makes sure you are not losing any local modification before doing so. - git-reset ... can be used to revert index entries for selected paths. - git-update-index is much less visible. Many suggestions to use the command in git output and documentation have now been replaced by simpler commands such as "git add" or "git rm". * Repository layout and objects transfer - The data for origin repository is stored in the configuration file $GIT_DIR/config, not in $GIT_DIR/remotes/, for newly created clones. The latter is still supported and there is no need to convert your existing repository if you are already comfortable with your workflow with the layout. - git-clone always uses what is known as "separate remote" layout for a newly created repository with a working tree. A repository with the separate remote layout starts with only one default branch, 'master', to be used for your own development. Unlike the traditional layout that copied all the upstream branches into your branch namespace (while renaming their 'master' to your 'origin'), the new layout puts upstream branches into local "remote-tracking branches" with their own namespace. These can be referenced with names such as "origin/$upstream_branch_name" and are stored in .git/refs/remotes rather than .git/refs/heads where normal branches are stored. This layout keeps your own branch namespace less cluttered, avoids name collision with your upstream, makes it possible to automatically track new branches created at the remote after you clone from it, and makes it easier to interact with more than one remote repository (you can use "git remote" to add other repositories to track). There might be some surprises: * 'git branch' does not show the remote tracking branches. It only lists your own branches. Use '-r' option to view the tracking branches. * If you are forking off of a branch obtained from the upstream, you would have done something like 'git branch my-next next', because traditional layout dropped the tracking branch 'next' into your own branch namespace. With the separate remote layout, you say 'git branch next origin/next', which allows you to use the matching name 'next' for your own branch. It also allows you to track a remote other than 'origin' (i.e. where you initially cloned from) and fork off of a branch from there the same way (e.g. "git branch mingw j6t/master"). Repositories initialized with the traditional layout continue to work. - New branches that appear on the origin side after a clone is made are also tracked automatically. This is done with an wildcard refspec "refs/heads/*:refs/remotes/origin/*", which older git does not understand, so if you clone with 1.5.0, you would need to downgrade remote.*.fetch in the configuration file to specify each branch you are interested in individually if you plan to fetch into the repository with older versions of git (but why would you?). - Similarly, wildcard refspec "refs/heads/*:refs/remotes/me/*" can be given to "git-push" command to update the tracking branches that is used to track the repository you are pushing from on the remote side. - git-branch and git-show-branch know remote tracking branches (use the command line switch "-r" to list only tracked branches). - git-push can now be used to delete a remote branch or a tag. This requires the updated git on the remote side (use "git push :refs/heads/" to delete "branch"). - git-push more aggressively keeps the transferred objects packed. Earlier we recommended to monitor amount of loose objects and repack regularly, but you should repack when you accumulated too many small packs this way as well. Updated git-count-objects helps you with this. - git-fetch also more aggressively keeps the transferred objects packed. This behavior of git-push and git-fetch can be tweaked with a single configuration transfer.unpacklimit (but usually there should not be any need for a user to tweak it). - A new command, git-remote, can help you manage your remote tracking branch definitions. - You may need to specify explicit paths for upload-pack and/or receive-pack due to your ssh daemon configuration on the other end. This can now be done via remote.*.uploadpack and remote.*.receivepack configuration. * Bare repositories - Certain commands change their behavior in a bare repository (i.e. a repository without associated working tree). We use a fairly conservative heuristic (if $GIT_DIR is ".git", or ends with "/.git", the repository is not bare) to decide if a repository is bare, but "core.bare" configuration variable can be used to override the heuristic when it misidentifies your repository. - git-fetch used to complain updating the current branch but this is now allowed for a bare repository. So is the use of 'git-branch -f' to update the current branch. - Porcelain-ish commands that require a working tree refuses to work in a bare repository. * Reflog - Reflog records the history from the view point of the local repository. In other words, regardless of the real history, the reflog shows the history as seen by one particular repository (this enables you to ask "what was the current revision in _this_ repository, yesterday at 1pm?"). This facility is enabled by default for repositories with working trees, and can be accessed with the "branch@{time}" and "branch@{Nth}" notation. - "git show-branch" learned showing the reflog data with the new -g option. "git log" has -g option to view reflog entries in a more verbose manner. - git-branch knows how to rename branches and moves existing reflog data from the old branch to the new one. - In addition to the reflog support in v1.4.4 series, HEAD reference maintains its own log. "HEAD@{5.minutes.ago}" means the commit you were at 5 minutes ago, which takes branch switching into account. If you want to know where the tip of your current branch was at 5 minutes ago, you need to explicitly say its name (e.g. "master@{5.minutes.ago}") or omit the refname altogether i.e. "@{5.minutes.ago}". - The commits referred to by reflog entries are now protected against pruning. The new command "git reflog expire" can be used to truncate older reflog entries and entries that refer to commits that have been pruned away previously with older versions of git. Existing repositories that have been using reflog may get complaints from fsck-objects and may not be able to run git-repack, if you had run git-prune from older git; please run "git reflog expire --stale-fix --all" first to remove reflog entries that refer to commits that are no longer in the repository when that happens. * Cruft removal - We used to say "old commits are retrievable using reflog and 'master@{yesterday}' syntax as long as you haven't run git-prune". We no longer have to say the latter half of the above sentence, as git-prune does not remove things reachable from reflog entries. - There is a toplevel garbage collector script, 'git-gc', that runs periodic cleanup functions, including 'git-repack -a -d', 'git-reflog expire', 'git-pack-refs --prune', and 'git-rerere gc'. - The output from fsck ("fsck-objects" is called just "fsck" now, but the old name continues to work) was needlessly alarming in that it warned missing objects that are reachable only from dangling objects. This has been corrected and the output is much more useful. * Detached HEAD - You can use 'git-checkout' to check out an arbitrary revision or a tag as well, instead of named branches. This will dissociate your HEAD from the branch you are currently on. A typical use of this feature is to "look around". E.g. $ git checkout v2.6.16 ... compile, test, etc. $ git checkout v2.6.17 ... compile, test, etc. - After detaching your HEAD, you can go back to an existing branch with usual "git checkout $branch". Also you can start a new branch using "git checkout -b $newbranch" to start a new branch at that commit. - You can even pull from other repositories, make merges and commits while your HEAD is detached. Also you can use "git reset" to jump to arbitrary commit, while still keeping your HEAD detached. Remember that a detached state is volatile, i.e. it will be forgotten as soon as you move away from it with the checkout or reset command, unless a branch is created from it as mentioned above. It is also possible to rescue a lost detached state from the HEAD reflog. * Packed refs - Repositories with hundreds of tags have been paying large overhead, both in storage and in runtime, due to the traditional one-ref-per-file format. A new command, git-pack-refs, can be used to "pack" them in more efficient representation (you can let git-gc do this for you). - Clones and fetches over dumb transports are now aware of packed refs and can download from repositories that use them. * Configuration - configuration related to color setting are consolidated under color.* namespace (older diff.color.*, status.color.* are still supported). - 'git-repo-config' command is accessible as 'git-config' now. * Updated features - git-describe uses better criteria to pick a base ref. It used to pick the one with the newest timestamp, but now it picks the one that is topologically the closest (that is, among ancestors of commit C, the ref T that has the shortest output from "git-rev-list T..C" is chosen). - git-describe gives the number of commits since the base ref between the refname and the hash suffix. E.g. the commit one before v2.6.20-rc6 in the kernel repository is: v2.6.20-rc5-306-ga21b069 which tells you that its object name begins with a21b069, v2.6.20-rc5 is an ancestor of it (meaning, the commit contains everything -rc5 has), and there are 306 commits since v2.6.20-rc5. - git-describe with --abbrev=0 can be used to show only the name of the base ref. - git-blame learned a new option, --incremental, that tells it to output the blames as they are assigned. A sample script to use it is also included as contrib/blameview. - git-blame starts annotating from the working tree by default. * Less external dependency - We no longer require the "merge" program from the RCS suite. All 3-way file-level merges are now done internally. - The original implementation of git-merge-recursive which was in Python has been removed; we have a C implementation of it now. - git-shortlog is no longer a Perl script. It no longer requires output piped from git-log; it can accept revision parameters directly on the command line. * I18n - We have always encouraged the commit message to be encoded in UTF-8, but the users are allowed to use legacy encoding as appropriate for their projects. This will continue to be the case. However, a non UTF-8 commit encoding _must_ be explicitly set with i18n.commitencoding in the repository where a commit is made; otherwise git-commit-tree will complain if the log message does not look like a valid UTF-8 string. - The value of i18n.commitencoding in the originating repository is recorded in the commit object on the "encoding" header, if it is not UTF-8. git-log and friends notice this, and re-encodes the message to the log output encoding when displaying, if they are different. The log output encoding is determined by "git log --encoding=", i18n.logoutputencoding configuration, or i18n.commitencoding configuration, in the decreasing order of preference, and defaults to UTF-8. - Tools for e-mailed patch application now default to -u behavior; i.e. it always re-codes from the e-mailed encoding to the encoding specified with i18n.commitencoding. This unfortunately forces projects that have happily been using a legacy encoding without setting i18n.commitencoding to set the configuration, but taken with other improvement, please excuse us for this very minor one-time inconvenience. * e-mailed patches - See the above I18n section. - git-format-patch now enables --binary without being asked. git-am does _not_ default to it, as sending binary patch via e-mail is unusual and is harder to review than textual patches and it is prudent to require the person who is applying the patch to explicitly ask for it. - The default suffix for git-format-patch output is now ".patch", not ".txt". This can be changed with --suffix=.txt option, or setting the config variable "format.suffix" to ".txt". * Foreign SCM interfaces - git-svn now requires the Perl SVN:: libraries, the command-line backend was too slow and limited. - the 'commit' subcommand of git-svn has been renamed to 'set-tree', and 'dcommit' is the recommended replacement for day-to-day work. - git fast-import backend. * User support - Quite a lot of documentation updates. - Bash completion scripts have been updated heavily. - Better error messages for often used Porcelainish commands. - Git GUI. This is a simple Tk based graphical interface for common Git operations. * Sliding mmap - We used to assume that we can mmap the whole packfile while in use, but with a large project this consumes huge virtual memory space and truly huge ones would not fit in the userland address space on 32-bit platforms. We now mmap huge packfile in pieces to avoid this problem. * Shallow clones - There is a partial support for 'shallow' repositories that keeps only recent history. A 'shallow clone' is created by specifying how deep that truncated history should be (e.g. "git clone --depth 5 git://some.where/repo.git"). Currently a shallow repository has number of limitations: - Cloning and fetching _from_ a shallow clone are not supported (nor tested -- so they might work by accident but they are not expected to). - Pushing from nor into a shallow clone are not expected to work. - Merging inside a shallow repository would work as long as a merge base is found in the recent history, but otherwise it will be like merging unrelated histories and may result in huge conflicts. but this would be more than adequate for people who want to look at near the tip of a big project with a deep history and send patches in e-mail format. ./RelNotes/1.5.1.1.adoc000066400000000000000000000041641504515377700144220ustar00rootroot00000000000000GIT v1.5.1.1 Release Notes ========================== Fixes since v1.5.1 ------------------ * Documentation updates - The --left-right option of rev-list and friends is documented. - The documentation for cvsimport has been majorly improved. - "git-show-ref --exclude-existing" was documented. * Bugfixes - The implementation of -p option in "git cvsexportcommit" had the meaning of -C (context reduction) option wrong, and loosened the context requirements when it was told to be strict. - "git cvsserver" did not behave like the real cvsserver when client side removed a file from the working tree without doing anything else on the path. In such a case, it should restore it from the checked out revision. - "git fsck" issued an alarming error message on detached HEAD. It is not an error since at least 1.5.0. - "git send-email" produced of References header of unbounded length; fixed this with line-folding. - "git archive" to download from remote site should not require you to be in a git repository, but it incorrectly did. - "git apply" ignored -p for "diff --git" formatted patches. - "git rerere" recorded a conflict that had one side empty (the other side adds) incorrectly; this made merging in the other direction fail to use previously recorded resolution. - t4200 test was broken where "wc -l" pads its output with spaces. - "git branch -m old new" to rename branch did not work without a configuration file in ".git/config". - The sample hook for notification e-mail was misnamed. - gitweb did not show type-changing patch correctly in the blobdiff view. - git-svn did not error out with incorrect command line options. - git-svn fell into an infinite loop when insanely long commit message was found. - git-svn dcommit and rebase was confused by patches that were merged from another branch that is managed by git-svn. - git-svn used to get confused when globbing remote branch/tag spec (e.g. "branches = proj/branches/*:refs/remotes/origin/*") is used and there was a plain file that matched the glob. ./RelNotes/1.5.1.2.adoc000066400000000000000000000031631504515377700144210ustar00rootroot00000000000000GIT v1.5.1.2 Release Notes ========================== Fixes since v1.5.1.1 -------------------- * Bugfixes - "git clone" over http from a repository that has lost the loose refs by running "git pack-refs" were broken (a code to deal with this was added to "git fetch" in v1.5.0, but it was missing from "git clone"). - "git diff a/ b/" incorrectly fell in "diff between two filesystem objects" codepath, when the user most likely wanted to limit the extent of output to two tracked directories. - git-quiltimport had the same bug as we fixed for git-applymbox in v1.5.1.1 -- it gave an alarming "did not have any patch" message (but did not actually fail and was harmless). - various git-svn fixes. - Sample update hook incorrectly always refused requests to delete branches through push. - git-blame on a very long working tree path had buffer overrun problem. - git-apply did not like to be fed two patches in a row that created and then modified the same file. - git-svn was confused when a non-project was stored directly under trunk/, branches/ and tags/. - git-svn wants the Error.pm module that was at least as new as what we ship as part of git; install ours in our private installation location if the one on the system is older. - An earlier update to command line integer parameter parser was botched and made 'update-index --cacheinfo' completely useless. * Documentation updates - Various documentation updates from J. Bruce Fields, Frank Lichtenheld, Alex Riesen and others. Andrew Ruder started a war on undocumented options. ./RelNotes/1.5.1.3.adoc000066400000000000000000000030361504515377700144210ustar00rootroot00000000000000GIT v1.5.1.3 Release Notes ========================== Fixes since v1.5.1.2 -------------------- * Bugfixes - git-add tried to optimize by finding common leading directories across its arguments but botched, causing very confused behaviour. - unofficial rpm.spec file shipped with git was letting ETC_GITCONFIG set to /usr/etc/gitconfig. Tweak the official Makefile to make it harder for distro people to make the same mistake, by setting the variable to /etc/gitconfig if prefix is set to /usr. - git-svn inconsistently stripped away username from the URL only when svnsync_props was in use. - git-svn got confused when handling symlinks on Mac OS. - git-send-email was not quoting recipient names that have period '.' in them. Also it did not allow overriding envelope sender, which made it impossible to send patches to certain subscriber-only lists. - built-in write_tree() routine had a sequence that renamed a file that is still open, which some systems did not like. - when memory is very tight, sliding mmap code to read packfiles incorrectly closed the fd that was still being used to read the pack. - import-tars contributed front-end for fastimport was passing wrong directory modes without checking. - git-fastimport trusted its input too much and allowed to create corrupt tree objects with entries without a name. - git-fetch needlessly barfed when too long reflog action description was given by the caller. Also contains various documentation updates. ./RelNotes/1.5.1.4.adoc000066400000000000000000000015131504515377700144200ustar00rootroot00000000000000GIT v1.5.1.4 Release Notes ========================== Fixes since v1.5.1.3 -------------------- * Bugfixes - "git-http-fetch" did not work around a bug in libcurl earlier than 7.16 (curl_multi_remove_handle() was broken). - "git cvsserver" handles a file that was once removed and then added again correctly. - import-tars script (in contrib/) handles GNU tar archives that contain pathnames longer than 100 bytes (long-link extension) correctly. - xdelta test program did not build correctly. - gitweb sometimes tried incorrectly to apply function to decode utf8 twice, resulting in corrupt output. - "git blame -C" mishandled text at the end of a group of lines. - "git log/rev-list --boundary" did not produce output correctly without --left-right option. - Many documentation updates. ./RelNotes/1.5.1.5.adoc000066400000000000000000000026261504515377700144270ustar00rootroot00000000000000GIT v1.5.1.5 Release Notes ========================== Fixes since v1.5.1.4 -------------------- * Bugfixes - git-send-email did not understand aliases file for mutt, which allows leading whitespaces. - git-format-patch emitted Content-Type and Content-Transfer-Encoding headers for non ASCII contents, but failed to add MIME-Version. - git-name-rev had a buffer overrun with a deep history. - contributed script import-tars did not get the directory in tar archives interpreted correctly. - git-svn was reported to segfault for many people on list and #git; hopefully this has been fixed. - "git-svn clone" does not try to minimize the URL (i.e. connect to higher level hierarchy) by default, as this can prevent clone to fail if only part of the repository (e.g. 'trunk') is open to public. - "git checkout branch^0" did not detach the head when you are already on 'branch'; backported the fix from the 'master'. - "git-config section.var" did not correctly work when existing configuration file had both [section] and [section "name"] next to each other. - "git clone ../other-directory" was fooled if the current directory $PWD points at is a symbolic link. - (build) tree_entry_extract() function was both static inline and extern, which caused trouble compiling with Forte12 compilers on Sun. - Many many documentation fixes and updates. ./RelNotes/1.5.1.6.adoc000066400000000000000000000027571504515377700144350ustar00rootroot00000000000000GIT v1.5.1.6 Release Notes ========================== Fixes since v1.5.1.4 -------------------- * Bugfixes - git-send-email did not understand aliases file for mutt, which allows leading whitespaces. - git-format-patch emitted Content-Type and Content-Transfer-Encoding headers for non ASCII contents, but failed to add MIME-Version. - git-name-rev had a buffer overrun with a deep history. - contributed script import-tars did not get the directory in tar archives interpreted correctly. - git-svn was reported to segfault for many people on list and #git; hopefully this has been fixed. - git-svn also had a bug to crash svnserve by sending a bad sequence of requests. - "git-svn clone" does not try to minimize the URL (i.e. connect to higher level hierarchy) by default, as this can prevent clone to fail if only part of the repository (e.g. 'trunk') is open to public. - "git checkout branch^0" did not detach the head when you are already on 'branch'; backported the fix from the 'master'. - "git-config section.var" did not correctly work when existing configuration file had both [section] and [section "name"] next to each other. - "git clone ../other-directory" was fooled if the current directory $PWD points at is a symbolic link. - (build) tree_entry_extract() function was both static inline and extern, which caused trouble compiling with Forte12 compilers on Sun. - Many many documentation fixes and updates. ./RelNotes/1.5.1.adoc000066400000000000000000000317501504515377700142640ustar00rootroot00000000000000GIT v1.5.1 Release Notes ======================== Updates since v1.5.0 -------------------- * Deprecated commands and options. - git-diff-stages and git-resolve have been removed. * New commands and options. - "git log" and friends take --reverse, which instructs them to give their output in the order opposite from their usual. They typically output from new to old, but with this option their output would read from old to new. "git shortlog" usually lists older commits first, but with this option, they are shown from new to old. - "git log --pretty=format:" to allow more flexible custom log output. - "git diff" learned --ignore-space-at-eol. This is a weaker form of --ignore-space-change. - "git diff --no-index pathA pathB" can be used as diff replacement with git specific enhancements. - "git diff --no-index" can read from '-' (standard input). - "git diff" also learned --exit-code to exit with non-zero status when it found differences. In the future we might want to make this the default but that would be a rather big backward incompatible change; it will stay as an option for now. - "git diff --quiet" is --exit-code with output turned off, meant for scripted use to quickly determine if there is any tree-level difference. - Textual patch generation with "git diff" without -w/-b option has been significantly optimized. "git blame" got faster because of the same change. - "git log" and "git rev-list" has been optimized significantly when they are used with pathspecs. - "git branch --track" can be used to set up configuration variables to help it easier to base your work on branches you track from a remote site. - "git format-patch --attach" now emits attachments. Use --inline to get an inlined multipart/mixed. - "git name-rev" learned --refs=, to limit the tags used for naming the given revisions only to the ones matching the given pattern. - "git remote update" is to run "git fetch" for defined remotes to update tracking branches. - "git cvsimport" can now take '-d' to talk with a CVS repository different from what are recorded in CVS/Root (overriding it with environment CVSROOT does not work). - "git bundle" can help sneaker-netting your changes between repositories. - "git mergetool" can help 3-way file-level conflict resolution with your favorite graphical merge tools. - A new configuration "core.symlinks" can be used to disable symlinks on filesystems that do not support them; they are checked out as regular files instead. - You can name a commit object with its first line of the message. The syntax to use is ':/message text'. E.g. $ git show ":/object name: introduce ':/' notation" means the same thing as: $ git show 28a4d940443806412effa246ecc7768a21553ec7 - "git bisect" learned a new command "run" that takes a script to run after each revision is checked out to determine if it is good or bad, to automate the bisection process. - "git log" family learned a new traversal option --first-parent, which does what the name suggests. * Updated behavior of existing commands. - "git-merge-recursive" used to barf when there are more than one common ancestors for the merge, and merging them had a rename/rename conflict. This has been fixed. - "git fsck" does not barf on corrupt loose objects. - "git rm" does not remove newly added files without -f. - "git archimport" allows remapping when coming up with git branch names from arch names. - git-svn got almost a rewrite. - core.autocrlf configuration, when set to 'true', makes git to convert CRLF at the end of lines in text files to LF when reading from the filesystem, and convert in reverse when writing to the filesystem. The variable can be set to 'input', in which case the conversion happens only while reading from the filesystem but files are written out with LF at the end of lines. Currently, which paths to consider 'text' (i.e. be subjected to the autocrlf mechanism) is decided purely based on the contents, but the plan is to allow users to explicitly override this heuristic based on paths. - The behavior of 'git-apply', when run in a subdirectory, without --index nor --cached were inconsistent with that of the command with these options. This was fixed to match the behavior with --index. A patch that is meant to be applied with -p1 from the toplevel of the project tree can be applied with any custom -p option. A patch that is not relative to the toplevel needs to be applied with -p option with or without --index (or --cached). - "git diff" outputs a trailing HT when pathnames have embedded SP on +++/--- header lines, in order to help "GNU patch" to parse its output. "git apply" was already updated to accept this modified output format since ce74618d (Sep 22, 2006). - "git cvsserver" runs hooks/update and honors its exit status. - "git cvsserver" can be told to send everything with -kb. - "git diff --check" also honors the --color output option. - "git name-rev" used to stress the fact that a ref is a tag too much, by saying something like "v1.2.3^0~22". It now says "v1.2.3~22" in such a case (it still says "v1.2.3^0" if it does not talk about an ancestor of the commit that is tagged, which makes sense). - "git rev-list --boundary" now shows boundary markers for the commits omitted by --max-age and --max-count condition. - The configuration mechanism now reads $(prefix)/etc/gitconfig. - "git apply --verbose" shows what preimage lines were wanted when it couldn't find them. - "git status" in a read-only repository got a bit saner. - "git fetch" (hence "git clone" and "git pull") are less noisy when the output does not go to tty. - "git fetch" between repositories with many refs were slow even when there are not many changes that needed transferring. This has been sped up by partially rewriting the heaviest parts in C. - "git mailinfo" which splits an e-mail into a patch and the meta-information was rewritten, thanks to Don Zickus. It handles nested multipart better. The command was broken for a brief period on 'master' branch since 1.5.0 but the breakage is fixed now. - send-email learned configurable bcc and chain-reply-to. - "git remote show $remote" also talks about branches that would be pushed if you run "git push remote". - Using objects from packs is now seriously optimized by clever use of a cache. This should be most noticeable in git-log family of commands that involve reading many tree objects. In addition, traversing revisions while filtering changes with pathspecs is made faster by terminating the comparison between the trees as early as possible. * Hooks - The part to send out notification e-mails was removed from the sample update hook, as it was not an appropriate place to do so. The proper place to do this is the new post-receive hook. An example hook has been added to contrib/hooks/. * Others - git-revert, git-gc and git-cherry-pick are now built-ins. Fixes since v1.5.0 ------------------ These are all in v1.5.0.x series. * Documentation updates - Clarifications and corrections to 1.5.0 release notes. - The main documentation did not link to git-remote documentation. - Clarified introductory text of git-rebase documentation. - Converted remaining mentions of update-index on Porcelain documents to git-add/git-rm. - Some i18n.* configuration variables were incorrectly described as core.*; fixed. - added and clarified core.bare, core.legacyheaders configurations. - updated "git-clone --depth" documentation. - user-manual updates. - Options to 'git remote add' were described insufficiently. - Configuration format.suffix was not documented. - Other formatting and spelling fixes. - user-manual has better cross references. - gitweb installation/deployment procedure is now documented. * Bugfixes - git-upload-pack closes unused pipe ends; earlier this caused many zombies to hang around. - git-rerere was recording the contents of earlier hunks duplicated in later hunks. This prevented resolving the same conflict when performing the same merge the other way around. - git-add and git-update-index on a filesystem on which executable bits are unreliable incorrectly reused st_mode bits even when the path changed between symlink and regular file. - git-daemon marks the listening sockets with FD_CLOEXEC so that it won't be leaked into the children. - segfault from git-blame when the mandatory pathname parameter was missing was fixed; usage() message is given instead. - git-rev-list did not read $GIT_DIR/config file, which means that did not honor i18n.logoutputencoding correctly. - Automated merge conflict handling when changes to symbolic links conflicted were completely broken. The merge-resolve strategy created a regular file with conflict markers in it in place of the symbolic link. The default strategy, merge-recursive was even more broken. It removed the path that was pointed at by the symbolic link. Both of these problems have been fixed. - 'git diff maint master next' did not correctly give combined diff across three trees. - 'git fast-import' portability fix for Solaris. - 'git show-ref --verify' without arguments did not error out but segfaulted. - 'git diff :tracked-file `pwd`/an-untracked-file' gave an extra slashes after a/ and b/. - 'git format-patch' produced too long filenames if the commit message had too long line at the beginning. - Running 'make all' and then without changing anything running 'make install' still rebuilt some files. This was inconvenient when building as yourself and then installing as root (especially problematic when the source directory is on NFS and root is mapped to nobody). - 'git-rerere' failed to deal with two unconflicted paths that sorted next to each other. - 'git-rerere' attempted to open(2) a symlink and failed if there was a conflict. Since a conflicting change to a symlink would not benefit from rerere anyway, the command now ignores conflicting changes to symlinks. - 'git-repack' did not like to pass more than 64 arguments internally to underlying 'rev-list' logic, which made it impossible to repack after accumulating many (small) packs in the repository. - 'git-diff' to review the combined diff during a conflicted merge were not reading the working tree version correctly when changes to a symbolic link conflicted. It should have read the data using readlink(2) but read from the regular file the symbolic link pointed at. - 'git-remote' did not like period in a remote's name. - 'git.el' honors the commit coding system from the configuration. - 'blameview' in contrib/ correctly digs deeper when a line is clicked. - 'http-push' correctly makes sure the remote side has leading path. Earlier it started in the middle of the path, and incorrectly. - 'git-merge' did not exit with non-zero status when the working tree was dirty and cannot fast forward. It does now. - 'cvsexportcommit' does not lose yet-to-be-used message file. - int-vs-size_t typefix when running combined diff on files over 2GB long. - 'git apply --whitespace=strip' should not touch unmodified lines. - 'git-mailinfo' choke when a logical header line was too long. - 'git show A..B' did not error out. Negative ref ("not A" in this example) does not make sense for the purpose of the command, so now it errors out. - 'git fmt-merge-msg --file' without file parameter did not correctly error out. - 'git archimport' barfed upon encountering a commit without summary. - 'git index-pack' did not protect itself from getting a short read out of pread(2). - 'git http-push' had a few buffer overruns. - Build dependency fixes to rebuild fetch.o when other headers change. - git.el does not add duplicate sign-off lines. - git-commit shows the full stat of the resulting commit, not just about the files in the current directory, when run from a subdirectory. - "git-checkout -m '@{8 hours ago}'" had a funny failure from eval; fixed. - git-merge (hence git-pull) did not refuse fast-forwarding when the working tree had local changes that would have conflicted with it. - a handful small fixes to gitweb. - build procedure for user-manual is fixed not to require locally installed stylesheets. - "git commit $paths" on paths whose earlier contents were already updated in the index were failing out. * Tweaks - sliding mmap() inefficiently mmaped the same region of a packfile with an access pattern that used objects in the reverse order. This has been made more efficient. ./RelNotes/1.5.2.1.adoc000066400000000000000000000026611504515377700144230ustar00rootroot00000000000000GIT v1.5.2.1 Release Notes ========================== Fixes since v1.5.2 ------------------ * Bugfixes - Temporary files that are used when invoking external diff programs did not tolerate a long TMPDIR. - git-daemon did not notice when it could not write into its pid file. - git-status did not honor core.excludesFile configuration like git-add did. - git-annotate did not work from a subdirectory while git-blame did. - git-cvsserver should have disabled access to a repository with "gitcvs.pserver.enabled = false" set even when "gitcvs.enabled = true" was set at the same time. It didn't. - git-cvsimport did not work correctly in a repository with its branch heads were packed with pack-refs. - ident unexpansion to squash "$Id: xxx $" that is in the repository copy removed incorrect number of bytes. - git-svn misbehaved when the subversion repository did not provide MD5 checksums for files. - git rebase (and git am) misbehaved on commits that have '\n' (literally backslash and en, not a linefeed) in the title. - code to decode base85 used in binary patches had one error return codepath wrong. - RFC2047 Q encoding output by git-format-patch used '_' for a space, which is not understood by some programs. It uses =20 which is safer. - git-fastimport --import-marks was broken; fixed. - A lot of documentation updates, clarifications and fixes. ./RelNotes/1.5.2.2.adoc000066400000000000000000000040461504515377700144230ustar00rootroot00000000000000GIT v1.5.2.2 Release Notes ========================== Fixes since v1.5.2.1 -------------------- * Usability fix - git-gui is shipped with its updated blame interface. It is rumored that the older one was not just unusable but was active health hazard, but this one is actually pretty. Please see for yourself. * Bugfixes - "git checkout fubar" was utterly confused when there is a branch fubar and a tag fubar at the same time. It correctly checks out the branch fubar now. - "git clone /path/foo" to clone a local /path/foo.git repository left an incorrect configuration. - "git send-email" correctly unquotes RFC 2047 quoted names in the patch-email before using their values. - We did not accept number of seconds since epoch older than year 2000 as a valid timestamp. We now interpret positive integers more than 8 digits as such, which allows us to express timestamps more recent than March 1973. - git-cvsimport did not work when you have GIT_DIR to point your repository at a nonstandard location. - Some systems (notably, Solaris) lack hstrerror() to make h_errno human readable; prepare a replacement implementation. - .gitignore file listed git-core.spec but what we generate is git.spec, and nobody noticed for a long time. - "git-merge-recursive" does not try to run file level merge on binary files. - "git-branch --track" did not create tracking configuration correctly when the branch name had slash in it. - The email address of the user specified with user.email configuration was overridden by EMAIL environment variable. - The tree parser did not warn about tree entries with nonsense file modes, and assumed they must be blobs. - "git log -z" without any other request to generate diff still invoked the diff machinery, wasting cycles. * Documentation - Many updates to fix stale or missing documentation. - Although our documentation was primarily meant to be formatted with AsciiDoc7, formatting with AsciiDoc8 is supported better. ./RelNotes/1.5.2.3.adoc000066400000000000000000000014341504515377700144220ustar00rootroot00000000000000GIT v1.5.2.3 Release Notes ========================== Fixes since v1.5.2.2 -------------------- * Bugfixes - Version 2 pack index format was introduced in version 1.5.2 to support pack files that has offset that cannot be represented in 32-bit. The runtime code to validate such an index mishandled such an index for an empty pack. - Commit walkers (most notably, fetch over http protocol) tried to traverse commit objects contained in trees (aka subproject); they shouldn't. - A build option NO_R_TO_GCC_LINKER was not explained in Makefile comment correctly. * Documentation Fixes and Updates - git-config --regexp was not documented properly. - git-repack -a was not documented properly. - git-remote -n was not documented properly. ./RelNotes/1.5.2.4.adoc000066400000000000000000000014421504515377700144220ustar00rootroot00000000000000GIT v1.5.2.4 Release Notes ========================== Fixes since v1.5.2.3 -------------------- * Bugfixes - "git-gui" bugfixes, including a handful fixes to run it better on Cygwin/MSYS. - "git checkout" failed to switch back and forth between branches, one of which has "frotz -> xyzzy" symlink and file "xyzzy/filfre", while the other one has a file "frotz/filfre". - "git prune" used to segfault upon seeing a commit that is referred to by a tree object (aka "subproject"). - "git diff --name-status --no-index" mishandled an added file. - "git apply --reverse --whitespace=warn" still complained about whitespaces that a forward application would have introduced. * Documentation Fixes and Updates - A handful documentation updates. ./RelNotes/1.5.2.5.adoc000066400000000000000000000016401504515377700144230ustar00rootroot00000000000000GIT v1.5.2.5 Release Notes ========================== Fixes since v1.5.2.4 -------------------- * Bugfixes - "git add -u" had a serious data corruption problem in one special case (when the changes to a subdirectory's files consist only deletion of files). - "git add -u " did not work from a subdirectory. - "git apply" left an empty directory after all its files are renamed away. - "git $anycmd foo/bar", when there is a file 'foo' in the working tree, complained that "git $anycmd foo/bar --" form should be used to disambiguate between revs and files, which was completely bogus. - "git checkout-index" and other commands that checks out files to the work tree tried unlink(2) on directories, which is a sane thing to do on sane systems, but not on Solaris when you are root. * Documentation Fixes and Updates - A handful documentation fixes. ./RelNotes/1.5.2.adoc000066400000000000000000000162731504515377700142700ustar00rootroot00000000000000GIT v1.5.2 Release Notes ======================== Updates since v1.5.1 -------------------- * Plumbing level superproject support. You can include a subdirectory that has an independent git repository in your index and tree objects of your project ("superproject"). This plumbing (i.e. "core") level superproject support explicitly excludes recursive behaviour. The "subproject" entries in the index and trees of a superproject are incompatible with older versions of git. Experimenting with the plumbing level support is encouraged, but be warned that unless everybody in your project updates to this release or later, using this feature would make your project inaccessible by people with older versions of git. * Plumbing level gitattributes support. The gitattributes mechanism allows you to add 'attributes' to paths in your project, and affect the way certain git operations work. Currently you can influence if a path is considered a binary or text (the former would be treated by 'git diff' not to produce textual output; the latter can go through the line endings conversion process in repositories with core.autocrlf set), expand and unexpand '$Id$' keyword with blob object name, specify a custom 3-way merge driver, and specify a custom diff driver. You can also apply arbitrary filter to contents on check-in/check-out codepath but this feature is an extremely sharp-edged razor and needs to be handled with caution (do not use it unless you understand the earlier mailing list discussion on keyword expansion). These conversions apply when checking files in or out, and exporting via git-archive. * The packfile format now optionally supports 64-bit index. This release supports the "version 2" format of the .idx file. This is automatically enabled when a huge packfile needs more than 32-bit to express offsets of objects in the pack. * Comes with an updated git-gui 0.7.1 * Updated gitweb: - can show combined diff for merges; - uses font size of user's preference, not hardcoded in pixels; - can now 'grep'; * New commands and options. - "git bisect start" can optionally take a single bad commit and zero or more good commits on the command line. - "git shortlog" can optionally be told to wrap its output. - "subtree" merge strategy allows another project to be merged in as your subdirectory. - "git format-patch" learned a new --subject-prefix= option, to override the built-in "[PATCH]". - "git add -u" is a quick way to do the first stage of "git commit -a" (i.e. update the index to match the working tree); it obviously does not make a commit. - "git clean" honors a new configuration, "clean.requireforce". When set to true, this makes "git clean" a no-op, preventing you from losing files by typing "git clean" when you meant to say "make clean". You can still say "git clean -f" to override this. - "git log" family of commands learned --date={local,relative,default} option. --date=relative is synonym to the --relative-date. --date=local gives the timestamp in local timezone. * Updated behavior of existing commands. - When $GIT_COMMITTER_EMAIL or $GIT_AUTHOR_EMAIL is not set but $EMAIL is set, the latter is used as a substitute. - "git diff --stat" shows size of preimage and postimage blobs for binary contents. Earlier it only said "Bin". - "git lost-found" shows stuff that are unreachable except from reflogs. - "git checkout branch^0" now detaches HEAD at the tip commit on the named branch, instead of just switching to the branch (use "git checkout branch" to switch to the branch, as before). - "git bisect next" can be used after giving only a bad commit without giving a good one (this starts bisection half-way to the root commit). We used to refuse to operate without a good and a bad commit. - "git push", when pushing into more than one repository, does not stop at the first error. - "git archive" does not insist you to give --format parameter anymore; it defaults to "tar". - "git cvsserver" can use backends other than sqlite. - "gitview" (in contrib/ section) learned to better support "git-annotate". - "git diff $commit1:$path2 $commit2:$path2" can now report mode changes between the two blobs. - Local "git fetch" from a repository whose object store is one of the alternates (e.g. fetching from the origin in a repository created with "git clone -l -s") avoids downloading objects unnecessarily. - "git blame" uses .mailmap to canonicalize the author name just like "git shortlog" does. - "git pack-objects" pays attention to pack.depth configuration variable. - "git cherry-pick" and "git revert" does not use .msg file in the working tree to prepare commit message; instead it uses $GIT_DIR/MERGE_MSG as other commands do. * Builds - git-p4import has never been installed; now there is an installation option to do so. - gitk and git-gui can be configured out. - Generated documentation pages automatically get version information from GIT_VERSION. - Parallel build with "make -j" descending into subdirectory was fixed. * Performance Tweaks - Optimized "git-rev-list --bisect" (hence "git-bisect"). - Optimized "git-add $path" in a large directory, most of whose contents are ignored. - Optimized "git-diff-tree" for reduced memory footprint. - The recursive merge strategy updated a worktree file that was changed identically in two branches, when one of them renamed it. We do not do that when there is no rename, so match that behaviour. This avoids excessive rebuilds. - The default pack depth has been increased to 50, as the recent addition of delta_base_cache makes deeper delta chains much less expensive to access. Depending on the project, it was reported that this reduces the resulting pack file by 10% or so. Fixes since v1.5.1 ------------------ All of the fixes in v1.5.1 maintenance series are included in this release, unless otherwise noted. * Bugfixes - Switching branches with "git checkout" refused to work when a path changes from a file to a directory between the current branch and the new branch, in order not to lose possible local changes in the directory that is being turned into a file with the switch. We now allow such a branch switch after making sure that there is no locally modified file nor un-ignored file in the directory. This has not been backported to 1.5.1.x series, as it is rather an intrusive change. - Merging branches that have a file in one and a directory in another at the same path used to get quite confused. We handle such a case a bit more carefully, even though that is still left as a conflict for the user to sort out. This will not be backported to 1.5.1.x series, as it is rather an intrusive change. - git-fetch had trouble with a remote with insanely large number of refs. - "git clean -d -X" now does not remove non-excluded directories. - rebasing (without -m) a series that changes a symlink to a directory in the middle of a path confused git-apply greatly and refused to operate. ./RelNotes/1.5.3.1.adoc000066400000000000000000000005061504515377700144200ustar00rootroot00000000000000GIT v1.5.3.1 Release Notes ========================== Fixes since v1.5.3 ------------------ This is solely to fix the generated RPM's dependencies. We used to have git-p4 package but we do not anymore. As suggested on the mailing list, this release makes git-core "Obsolete" git-p4, so that yum update would not complain. ./RelNotes/1.5.3.2.adoc000066400000000000000000000036131504515377700144230ustar00rootroot00000000000000GIT v1.5.3.2 Release Notes ========================== Fixes since v1.5.3.1 -------------------- * git-push sent thin packs by default, which was not good for the public distribution server (no point in saving transfer while pushing; no point in making the resulting pack less optimum). * git-svn sometimes terminated with "Malformed network data" when talking over svn:// protocol. * git-send-email re-issued the same message-id about 10% of the time if you fired off 30 messages within a single second. * git-stash was not terminating the log message of commits it internally creates with LF. * git-apply failed to check the size of the patch hunk when its beginning part matched the remainder of the preimage exactly, even though the preimage recorded in the hunk was much larger (therefore the patch should not have applied), leading to a segfault. * "git rm foo && git commit foo" complained that 'foo' needs to be added first, instead of committing the removal, which was a nonsense. * git grep -c said "/dev/null: 0". * git-add -u failed to recognize a blob whose type changed between the index and the work tree. * The limit to rename detection has been tightened a lot to reduce performance problems with a huge change. * cvsimport and svnimport barfed when the input tried to move a tag. * "git apply -pN" did not chop the right number of directories. * "git svnimport" did not like SVN tags with funny characters in them. * git-gui 0.8.3, with assorted fixes, including: - font-chooser on X11 was unusable with large number of fonts; - a diff that contained a deleted symlink made it barf; - an untracked symbolic link to a directory made it fart; - a file with % in its name made it vomit; Documentation updates --------------------- User manual has been somewhat restructured. I think the new organization is much easier to read. ./RelNotes/1.5.3.3.adoc000066400000000000000000000016001504515377700144160ustar00rootroot00000000000000GIT v1.5.3.3 Release Notes ========================== Fixes since v1.5.3.2 -------------------- * git-quiltimport did not like it when a patch described in the series file does not exist. * p4 importer missed executable bit in some cases. * The default shell on some FreeBSD did not execute the argument parsing code correctly and made git unusable. * git-svn incorrectly spawned pager even when the user explicitly asked not to. * sample post-receive hook overquoted the envelope sender value. * git-am got confused when the patch contained a change that is only about type and not contents. * git-mergetool did not show our and their version of the conflicted file when started from a subdirectory of the project. * git-mergetool did not pass correct options when invoking diff3. * git-log sometimes invoked underlying "diff" machinery unnecessarily. ./RelNotes/1.5.3.4.adoc000066400000000000000000000022701504515377700144230ustar00rootroot00000000000000GIT v1.5.3.4 Release Notes ========================== Fixes since v1.5.3.3 -------------------- * Change to "git-ls-files" in v1.5.3.3 that was introduced to support partial commit of removal better had a segfaulting bug, which was diagnosed and fixed by Keith and Carl. * Performance improvements for rename detection has been backported from the 'master' branch. * "git-for-each-ref --format='%(numparent)'" was not working correctly at all, and --format='%(parent)' was not working for merge commits. * Sample "post-receive-hook" incorrectly sent out push notification e-mails marked as "From: " the committer of the commit that happened to be at the tip of the branch that was pushed, not from the person who pushed. * "git-remote" did not exit non-zero status upon error. * "git-add -i" did not respond very well to EOF from tty nor bogus input. * "git-rebase -i" squash subcommand incorrectly made the author of later commit the author of resulting commit, instead of taking from the first one in the squashed series. * "git-stash apply --index" was not documented. * autoconfiguration learned that "ar" command is found as "gas" on some systems. ./RelNotes/1.5.3.5.adoc000066400000000000000000000064601504515377700144310ustar00rootroot00000000000000GIT v1.5.3.5 Release Notes ========================== Fixes since v1.5.3.4 -------------------- * Comes with git-gui 0.8.4. * "git-config" silently ignored options after --list; now it will error out with a usage message. * "git-config --file" failed if the argument used a relative path as it changed directories before opening the file. * "git-config --file" now displays a proper error message if it cannot read the file specified on the command line. * "git-config", "git-diff", "git-apply" failed if run from a subdirectory with relative GIT_DIR and GIT_WORK_TREE set. * "git-blame" crashed if run during a merge conflict. * "git-add -i" did not handle single line hunks correctly. * "git-rebase -i" and "git-stash apply" failed if external diff drivers were used for one or more files in a commit. They now avoid calling the external diff drivers. * "git-log --follow" did not work unless diff generation (e.g. -p) was also requested. * "git-log --follow -B" did not work at all. Fixed. * "git-log -M -B" did not correctly handle cases of very large files being renamed and replaced by very small files in the same commit. * "git-log" printed extra newlines between commits when a diff was generated internally (e.g. -S or --follow) but not displayed. * "git-push" error message is more helpful when pushing to a repository with no matching refs and none specified. * "git-push" now respects + (force push) on wildcard refspecs, matching the behavior of git-fetch. * "git-filter-branch" now updates the working directory when it has finished filtering the current branch. * "git-instaweb" no longer fails on Mac OS X. * "git-cvsexportcommit" didn't always create new parent directories before trying to create new child directories. Fixed. * "git-fetch" printed a scary (but bogus) error message while fetching a tag that pointed to a tree or blob. The error did not impact correctness, only user perception. The bogus error is no longer printed. * "git-ls-files --ignored" did not properly descend into non-ignored directories that themselves contained ignored files if d_type was not supported by the filesystem. This bug impacted systems such as AFS. Fixed. * Git segfaulted when reading an invalid .gitattributes file. Fixed. * post-receive-email example hook was fixed for non-fast-forward updates. * Documentation updates for supported (but previously undocumented) options of "git-archive" and "git-reflog". * "make clean" no longer deletes the configure script that ships with the git tarball, making multiple architecture builds easier. * "git-remote show origin" spewed a warning message from Perl when no remote is defined for the current branch via branch..remote configuration settings. * Building with NO_PERL_MAKEMAKER excessively rebuilt contents of perl/ subdirectory by rewriting perl.mak. * http.sslVerify configuration settings were not used in scripted Porcelains. * "git-add" leaked a bit of memory while scanning for files to add. * A few workarounds to squelch false warnings from recent gcc have been added. * "git-send-pack $remote frotz" segfaulted when there is nothing named 'frotz' on the local end. * "git-rebase --interactive" did not handle its "--strategy" option properly. ./RelNotes/1.5.3.6.adoc000066400000000000000000000026301504515377700144250ustar00rootroot00000000000000GIT v1.5.3.6 Release Notes ========================== Fixes since v1.5.3.5 -------------------- * git-cvsexportcommit handles root commits better. * git-svn dcommit used to clobber when sending a series of patches. * git-svn dcommit failed after attempting to rebase when started with a dirty index; now it stops upfront. * git-grep sometimes refused to work when your index was unmerged. * "git-grep -A1 -B2" acted as if it was told to run "git -A1 -B21". * git-hash-object did not honor configuration variables, such as core.compression. * git-index-pack choked on a huge pack on 32-bit machines, even when large file offsets are supported. * atom feeds from git-web said "10" for the month of November. * a memory leak in commit walker was plugged. * When git-send-email inserted the original author's From: address in body, it did not mark the message with Content-type: as needed. * git-revert and git-cherry-pick incorrectly refused to start when the work tree was dirty. * git-clean did not honor core.excludesfile configuration. * git-add mishandled ".gitignore" files when applying them to subdirectories. * While importing a too branchy history, git-fastimport did not honor delta depth limit properly. * Support for zlib implementations that lack ZLIB_VERNUM and definition of deflateBound() has been added. * Quite a lot of documentation clarifications. ./RelNotes/1.5.3.7.adoc000066400000000000000000000030601504515377700144240ustar00rootroot00000000000000GIT v1.5.3.7 Release Notes ========================== Fixes since v1.5.3.6 -------------------- * git-send-email added 8-bit contents to the payload without marking it as 8-bit in a CTE header. * "git-bundle create a.bndl HEAD" dereferenced the symref and did not record the ref as 'HEAD'; this prevented a bundle from being used as a normal source of git-clone. * The code to reject nonsense command line of the form "git-commit -a paths..." and "git-commit --interactive paths..." were broken. * Adding a signature that is not ASCII-only to an original commit that is ASCII-only would make the result non-ASCII. "git-format-patch -s" did not mark such a message correctly with MIME encoding header. * git-add sometimes did not mark the resulting index entry stat-clean. This affected only cases when adding the contents with the same length as the previously staged contents, and the previous staging made the index entry "racily clean". * git-commit did not honor GIT_INDEX_FILE the user had in the environment. * When checking out a revision, git-checkout did not report where the updated HEAD is if you happened to have a file called HEAD in the work tree. * "git-rev-list --objects" mishandled a tree that points at a submodule. * "git cvsimport" was not ready for packed refs that "git gc" can produce and gave incorrect results. * Many scripted Porcelains were confused when you happened to have a file called "HEAD" in your work tree. Also it contains updates to the user manual and documentation. ./RelNotes/1.5.3.8.adoc000066400000000000000000000014171504515377700144310ustar00rootroot00000000000000GIT v1.5.3.8 Release Notes ========================== Fixes since v1.5.3.7 -------------------- * Some documentation used "email.com" as an example domain. * git-svn fix to handle funky branch and project names going over http/https correctly. * git-svn fix to tone down a needlessly alarming warning message. * git-clone did not correctly report errors while fetching over http. * git-send-email added redundant Message-Id: header to the outgoing e-mail when the patch text already had one. * a read-beyond-end-of-buffer bug in configuration file updater was fixed. * git-grep used to show the same hit repeatedly for unmerged paths. * After amending the patch title in "git-am -i", the command did not report the patch it applied with the updated title. ./RelNotes/1.5.3.adoc000066400000000000000000000333161504515377700142660ustar00rootroot00000000000000GIT v1.5.3 Release Notes ======================== Updates since v1.5.2 -------------------- * The commit walkers other than http are officially deprecated, but still supported for now. * The submodule support has Porcelain layer. Note that the current submodule support is minimal and this is deliberately so. A design decision we made is that operations at the supermodule level do not recurse into submodules by default. The expectation is that later we would add a mechanism to tell git which submodules the user is interested in, and this information might be used to determine the recursive behaviour of certain commands (e.g. "git checkout" and "git diff"), but currently we haven't agreed on what that mechanism should look like. Therefore, if you use submodules, you would probably need "git submodule update" on the submodules you care about after running a "git checkout" at the supermodule level. * There are a handful pack-objects changes to help you cope better with repositories with pathologically large blobs in them. * For people who need to import from Perforce, a front-end for fast-import is in contrib/fast-import/. * Comes with git-gui 0.8.2. * Comes with updated gitk. * New commands and options. - "git log --date=" can use more formats: iso8601, rfc2822. - The hunk header output from "git diff" family can be customized with the attributes mechanism. See gitattributes(5) for details. - "git stash" allows you to quickly save away your work in progress and replay it later on an updated state. - "git rebase" learned an "interactive" mode that let you pick and reorder which commits to rebuild. - "git fsck" can save its findings in $GIT_DIR/lost-found, without a separate invocation of "git lost-found" command. The blobs stored by lost-found are stored in plain format to allow you to grep in them. - $GIT_WORK_TREE environment variable can be used together with $GIT_DIR to work in a subdirectory of a working tree that is not located at "$GIT_DIR/..". - Giving "--file=" option to "git config" is the same as running the command with GIT_CONFIG= environment. - "git log" learned a new option "--follow", to follow renaming history of a single file. - "git filter-branch" lets you rewrite the revision history of specified branches. You can specify a number of filters to modify the commits, files and trees. - "git cvsserver" learned new options (--base-path, --export-all, --strict-paths) inspired by "git daemon". - "git daemon --base-path-relaxed" can help migrating a repository URL that did not use to use --base-path to use --base-path. - "git commit" can use "-t templatefile" option and commit.template configuration variable to prime the commit message given to you in the editor. - "git submodule" command helps you manage the projects from the superproject that contain them. - In addition to core.compression configuration option, core.loosecompression and pack.compression options can independently tweak zlib compression levels used for loose and packed objects. - "git ls-tree -l" shows size of blobs pointed at by the tree entries, similar to "/bin/ls -l". - "git rev-list" learned --regexp-ignore-case and --extended-regexp options to tweak its matching logic used for --grep filtering. - "git describe --contains" is a handier way to call more obscure command "git name-rev --tags". - "git gc --aggressive" tells the command to spend more cycles to optimize the repository harder. - "git repack" learned a "window-memory" limit which dynamically reduces the window size to stay within the specified memory usage. - "git repack" can be told to split resulting packs to avoid exceeding limit specified with "--max-pack-size". - "git fsck" gained --verbose option. This is really really verbose but it might help you identify exact commit that is corrupt in your repository. - "git format-patch" learned --numbered-files option. This may be useful for MH users. - "git format-patch" learned format.subjectprefix configuration variable, which serves the same purpose as "--subject-prefix" option. - "git tag -n -l" shows tag annotations while listing tags. - "git cvsimport" can optionally use the separate-remote layout. - "git blame" can be told to see through commits that change whitespaces and indentation levels with "-w" option. - "git send-email" can be told not to thread the messages when sending out more than one patches. - "git send-email" can also be told how to find whom to cc the message to for each message via --cc-cmd. - "git config" learned NUL terminated output format via -z to help scripts. - "git add" learned "--refresh ..." option to selectively refresh the cached stat information. - "git init -q" makes the command quieter. - "git -p command" now has a cousin of opposite sex, "git --no-pager command". * Updated behavior of existing commands. - "gitweb" can offer multiple snapshot formats. ***NOTE*** Unfortunately, this changes the format of the $feature{snapshot}{default} entry in the per-site configuration file 'gitweb_config.perl'. It used to be a three-element tuple that describe a single format; with the new configuration item format, you only have to say the name of the format ('tgz', 'tbz2' or 'zip'). Please update the your configuration file accordingly. - "git clone" uses -l (hardlink files under .git) by default when cloning locally. - URL used for "git clone" and friends can specify nonstandard SSH port by using ssh://host:port/path/to/repo syntax. - "git bundle create" can now create a bundle without negative refs, i.e. "everything since the beginning up to certain points". - "git diff" (but not the plumbing level "git diff-tree") now recursively descends into trees by default. - "git diff" does not show differences that come only from stat-dirtiness in the form of "diff --git" header anymore. It runs "update-index --refresh" silently as needed. - "git tag -l" used to match tags by globbing its parameter as if it has wildcard '*' on both ends, which made "git tag -l gui" to match tag 'gitgui-0.7.0'; this was very annoying. You now have to add asterisk on the sides you want to wildcard yourself. - The editor to use with many interactive commands can be overridden with GIT_EDITOR environment variable, or if it does not exist, with core.editor configuration variable. As before, if you have neither, environment variables VISUAL and EDITOR are consulted in this order, and then finally we fall back on "vi". - "git rm --cached" does not complain when removing a newly added file from the index anymore. - Options to "git log" to affect how --grep/--author options look for given strings now have shorter abbreviations. -i is for ignore case, and -E is for extended regexp. - "git log" learned --log-size to show the number of bytes in the log message part of the output to help qgit. - "git log --name-status" does not require you to give "-r" anymore. As a general rule, Porcelain commands should recurse when showing diff. - "git format-patch --root A" can be used to format everything since the beginning up to A. This was supported with "git format-patch --root A A" for a long time, but was not properly documented. - "git svn dcommit" retains local merge information. - "git svnimport" allows an empty string to be specified as the trunk/ directory. This is necessary to suck data from a SVN repository that doe not have trunk/ branches/ and tags/ organization at all. - "git config" to set values also honors type flags like --bool and --int. - core.quotepath configuration can be used to make textual git output to emit most of the characters in the path literally. - "git mergetool" chooses its backend more wisely, taking notice of its environment such as use of X, Gnome/KDE, etc. - "gitweb" shows merge commits a lot nicer than before. The default view uses more compact --cc format, while the UI allows to choose normal diff with any parent. - snapshot files "gitweb" creates from a repository at $path/$project/.git are more useful. We use $project part in the filename, which we used to discard. - "git cvsimport" creates lightweight tags; there is no interesting information we can record in an annotated tag, and the handcrafted ones the old code created was not properly formed anyway. - "git push" pretends that you immediately fetched back from the remote by updating corresponding remote tracking branches if you have any. - The diffstat given after a merge (or a pull) honors the color.diff configuration. - "git commit --amend" is now compatible with various message source options such as -m/-C/-c/-F. - "git apply --whitespace=strip" removes blank lines added at the end of the file. - "git fetch" over git native protocols with "-v" option shows connection status, and the IP address of the other end, to help diagnosing problems. - We used to have core.legacyheaders configuration, when set to false, allowed git to write loose objects in a format that mimics the format used by objects stored in packs. It turns out that this was not so useful. Although we will continue to read objects written in that format, we do not honor that configuration anymore and create loose objects in the legacy/traditional format. - "--find-copies-harder" option to diff family can now be spelled as "-C -C" for brevity. - "git mailsplit" (hence "git am") can read from Maildir formatted mailboxes. - "git cvsserver" does not barf upon seeing "cvs login" request. - "pack-objects" honors "delta" attribute set in .gitattributes. It does not attempt to deltify blobs that come from paths with delta attribute set to false. - "new-workdir" script (in contrib) can now be used with a bare repository. - "git mergetool" learned to use gvimdiff. - "gitview" (in contrib) has a better blame interface. - "git log" and friends did not handle a commit log message that is larger than 16kB; they do now. - "--pretty=oneline" output format for "git log" and friends deals with "malformed" commit log messages that have more than one lines in the first paragraph better. We used to show the first line, cutting the title at mid-sentence; we concatenate them into a single line and treat the result as "oneline". - "git p4import" has been demoted to contrib status. For a superior option, checkout the "git p4" front end to "git fast-import" (also in contrib). The man page and p4 rpm have been removed as well. - "git mailinfo" (hence "am") now tries to see if the message is in utf-8 first, instead of assuming iso-8859-1, if incoming e-mail does not say what encoding it is in. * Builds - old-style function definitions (most notably, a function without parameter defined with "func()", not "func(void)") have been eradicated. - "git tag" and "git verify-tag" have been rewritten in C. * Performance Tweaks - "git pack-objects" avoids re-deltification cost by caching small enough delta results it creates while looking for the best delta candidates. - "git pack-objects" learned a new heuristic to prefer delta that is shallower in depth over the smallest delta possible. This improves both overall packfile access performance and packfile density. - diff-delta code that is used for packing has been improved to work better on big files. - when there are more than one pack files in the repository, the runtime used to try finding an object always from the newest packfile; it now tries the same packfile as we found the object requested the last time, which exploits the locality of references. - verifying pack contents done by "git fsck --full" got boost by carefully choosing the order to verify objects in them. - "git read-tree -m" to read into an already populated index has been optimized vastly. The effect of this can be seen when switching branches that have differences in only a handful paths. - "git add paths..." and "git commit paths..." has also been heavily optimized. Fixes since v1.5.2 ------------------ All of the fixes in v1.5.2 maintenance series are included in this release, unless otherwise noted. * Bugfixes - "gitweb" had trouble handling non UTF-8 text with older Encode.pm Perl module. - "git svn" misparsed the data from the commits in the repository when the user had "color.diff = true" in the configuration. This has been fixed. - There was a case where "git svn dcommit" clobbered changes made on the SVN side while committing multiple changes. - "git-write-tree" had a bad interaction with racy-git avoidance and gitattributes mechanisms. - "git --bare command" overrode existing GIT_DIR setting and always made it treat the current working directory as GIT_DIR. - "git ls-files --error-unmatch" does not complain if you give the same path pattern twice by mistake. - "git init" autodetected core.filemode but not core.symlinks, which made a new directory created automatically by "git clone" cumbersome to use on filesystems that require these configurations to be set. - "git log" family of commands behaved differently when run as "git log" (no pathspec) and as "git log --" (again, no pathspec). This inconsistency was introduced somewhere in v1.3.0 series but now has been corrected. - "git rebase -m" incorrectly displayed commits that were skipped. ./RelNotes/1.5.4.1.adoc000066400000000000000000000010141504515377700144140ustar00rootroot00000000000000GIT v1.5.4.1 Release Notes ========================== Fixes since v1.5.4 ------------------ * "git-commit -C $tag" used to work but rewrite in C done in 1.5.4 broke it. * An entry in the .gitattributes file that names a pattern in a subdirectory of the directory it is in did not match correctly (e.g. pattern "b/*.c" in "a/.gitattributes" should match "a/b/foo.c" but it didn't). * Customized color specification was parsed incorrectly when numeric color values are used. This was fixed in 1.5.4.1. ./RelNotes/1.5.4.2.adoc000066400000000000000000000027721504515377700144310ustar00rootroot00000000000000GIT v1.5.4.2 Release Notes ========================== Fixes since v1.5.4 ------------------ * The configuration parser was not prepared to see string valued variables misspelled as boolean and segfaulted. * Temporary files left behind due to interrupted object transfers were not cleaned up with "git prune". * "git config --unset" was confused when the unset variables were spelled with continuation lines in the config file. * The merge message detection in "git cvsimport" did not catch a message that began with "Merge...". * "git status" suggests "git rm --cached" for unstaging the earlier "git add" before the initial commit. * "git status" output was incorrect during a partial commit. * "git bisect" refused to start when the HEAD was detached. * "git bisect" allowed a wildcard character in the commit message expanded while writing its log file. * Manual pages were not formatted correctly with docbook xsl 1.72; added a workaround. * "git-commit -C $tag" used to work but rewrite in C done in 1.5.4 broke it. This was fixed in 1.5.4.1. * An entry in the .gitattributes file that names a pattern in a subdirectory of the directory it is in did not match correctly (e.g. pattern "b/*.c" in "a/.gitattributes" should match "a/b/foo.c" but it didn't). This was fixed in 1.5.4.1. * Customized color specification was parsed incorrectly when numeric color values are used. This was fixed in 1.5.4.1. * http transport misbehaved when linked with curl-gnutls. ./RelNotes/1.5.4.3.adoc000066400000000000000000000017451504515377700144310ustar00rootroot00000000000000GIT v1.5.4.3 Release Notes ========================== Fixes since v1.5.4.2 -------------------- * RPM spec used to pull in everything with 'git'. This has been changed so that 'git' package contains just the core parts, and we now supply 'git-all' metapackage to slurp in everything. This should match end user's expectation better. * When some refs failed to update, git-push reported "failure" which was unclear if some other refs were updated or all of them failed atomically (the answer is the former). Reworded the message to clarify this. * "git clone" from a repository whose HEAD was misconfigured did not set up the remote properly. Now it tries to do better. * Updated git-push documentation to clarify what "matching" means, in order to reduce user confusion. * Updated git-add documentation to clarify "add -u" operates in the current subdirectory you are in, just like other commands. * git-gui updates to work on OSX and Windows better. ./RelNotes/1.5.4.4.adoc000066400000000000000000000050711504515377700144260ustar00rootroot00000000000000GIT v1.5.4.4 Release Notes ========================== Fixes since v1.5.4.3 -------------------- * Building and installing with an overtight umask such as 077 made installed templates unreadable by others, while the rest of the install are done in a way that is friendly to umask 022. * "git cvsexportcommit -w $cvsdir" misbehaved when GIT_DIR is set to a relative directory. * "git http-push" had an invalid memory access that could lead it to segfault. * When "git rebase -i" gave control back to the user for a commit that is marked to be edited, it just said "modify it with commit --amend", without saying what to do to continue after modifying it. Give an explicit instruction to run "rebase --continue" to be more helpful. * "git send-email" in 1.5.4.3 issued a bogus empty In-Reply-To: header. * "git bisect" showed mysterious "won't bisect on seeked tree" error message. This was leftover from Cogito days to prevent "bisect" starting from a cg-seeked state. We still keep the Cogito safety, but running "git bisect start" when another bisect was in effect will clean up and start over. * "git push" with an explicit PATH to receive-pack did not quite work if receive-pack was not on usual PATH. We earlier fixed the same issue with "git fetch" and upload-pack, but somehow forgot to do so in the other direction. * git-gui's info dialog was not displayed correctly when the user tries to commit nothing (i.e. without staging anything). * "git revert" did not properly fail when attempting to run with a dirty index. * "git merge --no-commit --no-ff " incorrectly made commits. * "git merge --squash --no-ff ", which is a nonsense combination of options, was not rejected. * "git ls-remote" and "git remote show" against an empty repository failed, instead of just giving an empty result (regression). * "git fast-import" did not handle a renamed path whose name needs to be quoted, due to a bug in unquote_c_style() function. * "git cvsexportcommit" was confused when multiple files with the same basename needed to be pushed out in the same commit. * "git daemon" did not send early errors to syslog. * "git log --merge" did not work well with --left-right option. * "git svn" prompted for client cert password every time it accessed the server. * The reset command in "git fast-import" data stream was documented to end with an optional LF, but it actually required one. * "git svn dcommit/rebase" did not honor --rewrite-root option. Also included are a handful documentation updates. ./RelNotes/1.5.4.5.adoc000066400000000000000000000043011504515377700144220ustar00rootroot00000000000000GIT v1.5.4.5 Release Notes ========================== Fixes since v1.5.4.4 -------------------- * "git fetch there" when the URL information came from the Cogito style branches/there file did not update refs/heads/there (regression in 1.5.4). * Bogus refspec configuration such as "remote.there.fetch = =" were not detected as errors (regression in 1.5.4). * You couldn't specify a custom editor whose path contains a whitespace via GIT_EDITOR (and core.editor). * The subdirectory filter to "git filter-branch" mishandled a history where the subdirectory becomes empty and then later becomes non-empty. * "git shortlog" gave an empty line if the original commit message was malformed (e.g. a botched import from foreign SCM). Now it finds the first non-empty line and uses it for better information. * When the user fails to give a revision parameter to "git svn", an error from the Perl interpreter was issued because the script lacked proper error checking. * After "git rebase" stopped due to conflicts, if the user played with "git reset" and friends, "git rebase --abort" failed to go back to the correct commit. * Additional work trees prepared with git-new-workdir (in contrib/) did not share git-svn metadata directory .git/svn with the original. * "git-merge-recursive" did not mark addition of the same path with different filemodes correctly as a conflict. * "gitweb" gave malformed URL when pathinfo stype paths are in use. * "-n" stands for "--no-tags" again for "git fetch". * "git format-patch" did not detect the need to add 8-bit MIME header when the user used format.header configuration. * "rev~" revision specifier used to mean "rev", which was inconsistent with how "rev^" worked. Now "rev~" is the same as "rev~1" (hence it also is the same as "rev^1"), and "rev~0" is the same as "rev^0" (i.e. it has to be a commit). * "git quiltimport" did not grok empty lines, lines in "file -pNNN" format to specify the prefix levels and lines with trailing comments. * "git rebase -m" triggered pre-commit verification, which made "rebase --continue" impossible. As usual, it also comes with many documentation fixes and clarifications. ./RelNotes/1.5.4.6.adoc000066400000000000000000000026401504515377700144270ustar00rootroot00000000000000GIT v1.5.4.6 Release Notes ========================== I personally do not think there is any reason anybody should want to run v1.5.4.X series these days, because 'master' version is always more stable than any tagged released version of git. This is primarily to futureproof "git-shell" to accept requests without a dash between "git" and subcommand name (e.g. "git upload-pack") which the newer client will start to make sometime in the future. Fixes since v1.5.4.5 -------------------- * Command line option "-n" to "git-repack" was not correctly parsed. * Error messages from "git-apply" when the patchfile cannot be opened have been improved. * Error messages from "git-bisect" when given nonsense revisions have been improved. * reflog syntax that uses time e.g. "HEAD@{10 seconds ago}:path" did not stop parsing at the closing "}". * "git rev-parse --symbolic-full-name ^master^2" printed solitary "^", but it should print nothing. * "git apply" did not enforce "match at the beginning" correctly. * a path specification "a/b" in .gitattributes file should not match "sub/a/b", but it did. * "git log --date-order --topo-order" did not override the earlier date-order with topo-order as expected. * "git fast-export" did not export octopus merges correctly. * "git archive --prefix=$path/" mishandled gitattributes. As usual, it also comes with many documentation fixes and clarifications. ./RelNotes/1.5.4.7.adoc000066400000000000000000000005261504515377700144310ustar00rootroot00000000000000GIT v1.5.4.7 Release Notes ========================== Fixes since 1.5.4.7 ------------------- * Removed support for an obsolete gitweb request URI, whose implementation ran "git diff" Porcelain, instead of using plumbing, which would have run an external diff command specified in the repository configuration as the gitweb user. ./RelNotes/1.5.4.adoc000066400000000000000000000341301504515377700142620ustar00rootroot00000000000000GIT v1.5.4 Release Notes ======================== Removal ------- * "git svnimport" was removed in favor of "git svn". It is still there in the source tree (contrib/examples) but unsupported. * As git-commit and git-status have been rewritten, "git runstatus" helper script lost all its users and has been removed. Temporarily disabled -------------------- * "git http-push" is known not to work well with cURL library older than 7.16, and we had reports of repository corruption. It is disabled on such platforms for now. Unfortunately, 1.5.3.8 shares the same issue. In other words, this does not mean you will be fine if you stick to an older git release. For now, please do not use http-push from older git with cURL older than 7.16 if you value your data. A proper fix will hopefully materialize in later versions. Deprecation notices ------------------- * From v1.6.0, git will by default install dashed form of commands (e.g. "git-commit") outside of users' normal $PATH, and will install only selected commands ("git" itself, and "gitk") in $PATH. This implies: - Using dashed forms of git commands (e.g. "git-commit") from the command line has been informally deprecated since early 2006, but now it officially is, and will be removed in the future. Use dash-less forms (e.g. "git commit") instead. - Using dashed forms from your scripts, without first prepending the return value from "git --exec-path" to the scripts' PATH, has been informally deprecated since early 2006, but now it officially is. - Use of dashed forms with "PATH=$(git --exec-path):$PATH; export PATH" early in your script is not deprecated with this change. Users are strongly encouraged to adjust their habits and scripts now to prepare for this change. * The post-receive hook was introduced in March 2007 to supersede the post-update hook, primarily to overcome the command line length limitation of the latter. Use of post-update hook will be deprecated in future versions of git, starting from v1.6.0. * "git lost-found" was deprecated in favor of "git fsck"'s --lost-found option, and will be removed in the future. * "git peek-remote" is deprecated, as "git ls-remote" was written in C and works for all transports; "git peek-remote" will be removed in the future. * "git repo-config" which was an old name for "git config" command has been supported without being advertised for a long time. The next feature release will remove it. * From v1.6.0, the repack.usedeltabaseoffset config option will default to true, which will give denser packfiles (i.e. more efficient storage). The downside is that git older than version 1.4.4 will not be able to directly use a repository packed using this setting. * From v1.6.0, the pack.indexversion config option will default to 2, which is slightly more efficient, and makes repacking more immune to data corruptions. Git older than version 1.5.2 may revert to version 1 of the pack index with a manual "git index-pack" to be able to directly access corresponding pack files. Updates since v1.5.3 -------------------- * Comes with much improved gitk, with i18n. * Comes with git-gui 0.9.2 with i18n. * gitk is now merged as a subdirectory of git.git project, in preparation for its i18n. * progress displays from many commands are a lot nicer to the eye. Transfer commands show throughput data. * many commands that pay attention to per-directory .gitignore now do so lazily, which makes the usual case go much faster. * Output processing for '--pretty=format:' has been optimized. * Rename detection of diff family while detecting exact matches has been greatly optimized. * Rename detection of diff family tries to make more natural looking pairing. Earlier, if multiple identical rename sources were found in the preimage, the source used was picked pretty much at random. * Value "true" for color.diff and color.status configuration used to mean "always" (even when the output is not going to a terminal). This has been corrected to mean the same thing as "auto". * "git diff" Porcelain now respects diff.external configuration, which is another way to specify GIT_EXTERNAL_DIFF. * "git diff" can be told to use different prefixes other than "a/" and "b/" e.g. "git diff --src-prefix=l/ --dst-prefix=k/". * "git diff" sometimes did not quote paths with funny characters properly. * "git log" (and any revision traversal commands) misbehaved when --diff-filter is given but was not asked to actually produce diff. * HTTP proxy can be specified per remote repository using remote.*.httpproxy configuration, or global http.proxy configuration variable. * Various Perforce importer updates. * Example update and post-receive hooks have been improved. * Any command that wants to take a commit object name can now use ":/string" syntax to name a commit. * "git reset" is now built-in and its output can be squelched with -q. * "git reset --hard" does not make any sense in a bare repository, but did not error out; fixed. * "git send-email" can optionally talk over ssmtp and use SMTP-AUTH. * "git rebase" learned --whitespace option. * In "git rebase", when you decide not to replay a particular change after the command stopped with a conflict, you can say "git rebase --skip" without first running "git reset --hard", as the command now runs it for you. * "git rebase --interactive" mode can now work on detached HEAD. * Other minor to serious bugs in "git rebase -i" have been fixed. * "git rebase" now detaches head during its operation, so after a successful "git rebase" operation, the reflog entry branch@{1} for the current branch points at the commit before the rebase was started. * "git rebase -i" also triggers rerere to help your repeated merges. * "git merge" can call the "post-merge" hook. * "git pack-objects" can optionally run deltification with multiple threads. * "git archive" can optionally substitute keywords in files marked with export-subst attribute. * "git cherry-pick" made a misguided attempt to repeat the original command line in the generated log message, when told to cherry-pick a commit by naming a tag that points at it. It does not anymore. * "git for-each-ref" learned %(xxxdate:) syntax to show the various date fields in different formats. * "git gc --auto" is a low-impact way to automatically run a variant of "git repack" that does not lose unreferenced objects (read: safer than the usual one) after the user accumulates too many loose objects. * "git clean" has been rewritten in C. * You need to explicitly set clean.requireForce to "false" to allow "git clean" without -f to do any damage (lack of the configuration variable used to mean "do not require -f option to lose untracked files", but we now use the safer default). * The kinds of whitespace errors "git diff" and "git apply" notice (and fix) can be controlled via 'core.whitespace' configuration variable and 'whitespace' attribute in .gitattributes file. * "git push" learned --dry-run option to show what would happen if a push is run. * "git push" does not update a tracking ref on the local side when the remote refused to update the corresponding ref. * "git push" learned --mirror option. This is to push the local refs one-to-one to the remote, and deletes refs from the remote that do not exist anymore in the repository on the pushing side. * "git push" can remove a corrupt ref at the remote site with the usual ":ref" refspec. * "git remote" knows --mirror mode. This is to set up configuration to push into a remote repository to store local branch heads to the same branch on the remote side, and remove branch heads locally removed from local repository at the same time. Suitable for pushing into a back-up repository. * "git remote" learned "rm" subcommand. * "git cvsserver" can be run via "git shell". Also, "cvs" is recognized as a synonym for "git cvsserver", so that CVS users can be switched to git just by changing their login shell. * "git cvsserver" acts more like receive-pack by running post-receive and post-update hooks. * "git am" and "git rebase" are far less verbose. * "git pull" learned to pass --[no-]ff option to underlying "git merge". * "git pull --rebase" is a different way to integrate what you fetched into your current branch. * "git fast-export" produces data-stream that can be fed to fast-import to reproduce the history recorded in a git repository. * "git add -i" takes pathspecs to limit the set of files to work on. * "git add -p" is a short-hand to go directly to the selective patch subcommand in the interactive command loop and to exit when done. * "git add -i" UI has been colorized. The interactive prompt and menu can be colored by setting color.interactive configuration. The diff output (including the hunk picker) are colored with color.diff configuration. * "git commit --allow-empty" allows you to create a single-parent commit that records the same tree as its parent, overriding the usual safety valve. * "git commit --amend" can amend a merge that does not change the tree from its first parent. * "git commit" used to unconditionally strip comment lines that began with '#' and removed excess blank lines. This behavior has been made configurable. * "git commit" has been rewritten in C. * "git stash random-text" does not create a new stash anymore. It was a UI mistake. Use "git stash save random-text", or "git stash" (without extra args) for that. * "git stash clear extra-text" does not clear the whole stash anymore. It is tempting to expect "git stash clear stash@{2}" to drop only a single named stash entry, and it is rude to discard everything when that is asked (but not provided). * "git prune --expire