| Message ID | a0cf68f8ba2adefae4fceeab0d438d05e355e695.1663959324.git.gitgitgadget@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Add the Git Change command | expand |
On Fri, Sep 23, 2022 at 11:56 AM Stefan Xenos via GitGitGadget <gitgitgadget@gmail.com> wrote: > > From: Stefan Xenos <sxenos@google.com> > > This document describes what a change graph for > git would look like, the behavior of the evolve command, > and the changes planned for other commands. > > It was originally proposed in 2018, see > https://public-inbox.org/git/20181115005546.212538-1-sxenos@google.com/ > > Signed-off-by: Stefan Xenos <sxenos@google.com> > Signed-off-by: Chris Poucet <poucet@google.com> > --- > Documentation/technical/evolve.txt | 1051 ++++++++++++++++++++++++++++ > 1 file changed, 1051 insertions(+) > create mode 100644 Documentation/technical/evolve.txt > > diff --git a/Documentation/technical/evolve.txt b/Documentation/technical/evolve.txt > new file mode 100644 > index 00000000000..68ee2457e52 > --- /dev/null > +++ b/Documentation/technical/evolve.txt > @@ -0,0 +1,1051 @@ > +Evolve > +====== > + > +Objective > +========= > +Create an "evolve" command to help users craft a high quality commit history. > +Users can improve commits one at a time and in any order, then run git evolve to > +rewrite their recent history to ensure everything is up-to-date. We track > +amendments to a commit over time in a change graph. Users can share their > +progress with others by exchanging their change graphs using the standard push, > +fetch, and format-patch commands. > + > +Status > +====== > +This proposal has not been implemented yet. > + > +Background > +========== > +Imagine you have three sequential changes up for review and you receive feedback > +that requires editing all three changes. We'll define the word "change" > +formally later, but for the moment let's say that a change is a work-in-progress > +whose final version will be submitted as a commit in the future. > + > +While you're editing one change, more feedback arrives on one of the others. > +What do you do? > + > +The evolve command is a convenient way to work with chains of commits that are > +under review. Whenever you rebase or amend a commit, the repository remembers > +that the old commit is obsolete and has been replaced by the new one. Then, at > +some point in the future, you can run "git evolve" and the correct sequence of > +rebases will occur in the correct order such that no commit has an obsolete > +parent. > + > +Part of making the "evolve" command work involves tracking the edits to a commit > +over time, which is why we need an change graph. However, the change > +graph will also bring other benefits: > + > +- Users can view the history of a change directly (the sequence of amends and > + rebases it has undergone, orthogonal to the history of the branch it is on). > +- It will be possible to quickly locate and list all the changes the user > + currently has in progress. > +- It can be used as part of other high-level commands that combine or split > + changes. > +- It can be used to decorate commits (in git log, gitk, etc) that are either > + obsolete or are the tip of a work in progress. > +- By pushing and pulling the change graph, users can collaborate more > + easily on changes-in-progress. This is better than pushing and pulling the > + commits themselves since the change graph can be used to locate a more > + specific merge base, allowing for better merges between different versions of > + the same change. > +- It could be used to correctly rebase local changes and other local branches > + after running git-filter-branch. > +- It can replace the change-id footer used by gerrit. > + > +Goals > +----- > +Legend: Goals marked with P0 are required. Goals marked with Pn should be > +attempted unless they interfere with goals marked with Pn-1. > + > +P0. All commands that modify commits (such as the normal commit --amend or > + rebase command) should mark the old commit as being obsolete and replaced by > + the new one. No additional commands should be required to keep the > + change graph up-to-date. > +P0. Any commit that may be involved in a future evolve command should not be > + garbage collected. Specifically: > + - Commits that obsolete another should not be garbage collected until > + user-specified conditions have occurred and the change has expired from > + the reflog. User specified conditions for removing changes include: > + - The user explicitly deleted the change. > + - The change was merged into a specific branch. > + - Commits that have been obsoleted by another should not be garbage > + collected if any of their replacements are still being retained. > +P0. A commit can be obsoleted by more than one replacement (called divergence). > +P0. Users must be able to resolve divergence (convergence). > +P1. Users should be able to share chains of obsolete changes in order to > + collaborate on WIP changes. > +P2. Such sharing should be at the user’s option. That is, it should be possible > + to directly share a change without also sharing the file states or commit > + comments from the obsolete changes that led up to it, and the choice not to > + share those commits should not require changing any commit hashes. > +P2. It should be possible to discard part or all of the change graph > + without discarding the commits themselves that are already present in > + branches and the reflog. > +P2. Provide sufficient information to replace gerrit's Change-Id footers. > + > +Similar technologies > +-------------------- > +There are some other technologies that address the same end-user problem. > + > +Rebase -i can be used to solve the same problem, but users can't easily switch > +tasks midway through an interactive rebase or have more than one interactive > +rebase going on at the same time. It can't handle the case where you have > +multiple changes sharing the same parent when that parent needs to be rebased > +and won't let you collaborate with others on resolving a complicated interactive > +rebase. You can think of rebase -i as a top-down approach and the evolve command > +as the bottom-up approach to the same problem. I'll mention some other tools in this space too: revup amend (https://github.com/Skydio/revup/blob/main/docs/amend.md) (I'm the author) allows insertion of cached changes into any commit in the current history, and then reapplies the rest of history on top of those changes. It uses a "git apply --cached" engine under the hood so doesn't touch the working directory (although it will soon use the new git merge-tree). When paired with "revup upload" which creates and pushes multiple branches in the background for you, its possible to work on a "graph" of changes on a single branch linearly, then have the true graph structure created at upload time. git-revise (https://github.com/mystor/git-revise) does some very similar things except it uses "git merge-file" combined with manually merging the resulting trees. git branchstack (https://github.com/krobelus/git-branchstack) can also create branches in the background for you with the same mechanism. These tools don't store any external state, but as such also don't provide any specific collaboration mechanism for individual changes, so I'm interested in the "evolve" approach as well. > + > +Several patch queue managers have been built on top of git (such as topgit, > +stgit, and quilt). They address the same user need. However they also rely on > +state managed outside git that needs to be kept in sync. Such state can be > +easily damaged when running a git native command that is unaware of the patch > +queue. They also typically require an explicit initialization step to be done by > +the user which creates workflow problems. > + > +Mercurial implements a very similar feature in its EvolveExtension. The behavior > +of the evolve command itself is very similar, but the storage format for the > +change graph differs. In the case of mercurial, each change set can have one or > +more obsolescence markers that point to other changesets that they replace. This > +is similar to the "Commit Headers" approach considered in the other options > +appendix. The approach proposed here stores obsolescence information in a > +separate metacommit graph, which makes exchanging of obsolescence information > +optional. > + > +Mercurial's default behavior makes it easy to find and switch between > +non-obsolete changesets that aren't currently on any branch. We introduce the > +notion of a new ref namespace that enables a similar workflow via a different > +mechanism. Mercurial has the notion of changeset phases which isn't present > +in git and creates new ways for a changeset to diverge. Git doesn't need > +to deal with these issues, but it has to deal with the problems of picking an > +upstream branch as a target for rebases and protecting obsolescence information > +from GC. We also introduce some additional transformations (see > +obsolescence-over-cherry-pick, below) that aren't present in the mercurial > +implementation. > + > +Semi-related work > +----------------- > +There are other technologies that address different problems but have some > +similarities with this proposal. > + > +Replacements (refs/replace) are superficially similar to obsolescences in that > +they describe that one commit should be replaced by another. However, they > +differ in both how they are created and how they are intended to be used. > +Obsolescences are created automatically by the commands a user runs, and they > +describe the user’s intent to perform a future rebase. Obsolete commits still > +appear in branches, logs, etc like normal commits (possibly with an extra > +decoration that marks them as obsolete). Replacements are typically created > +explicitly by the user, they are meant to be kept around for a long time, and > +they describe a replacement to be applied at read-time rather than as the input > +to a future operation. When a replaced commit is queried, it is typically hidden > +and swapped out with its replacement as though the replacement has already > +occurred. > + > +Git-imerge is a project to help make complicated merges easier, particularly > +when merging or rebasing long chains of patches. It is not an alternative to > +the change graph, but its algorithm of applying smaller incremental merges > +could be used as part of the evolve algorithm in the future. > + > +Overview > +======== > +We introduce the notion of “meta-commits” which describe how one commit was > +created from other commits. A branch of meta-commits is known as a change. > +Changes are created and updated automatically whenever a user runs a command > +that creates a commit. They are used for locating obsolete commits, providing a > +list of a user’s unsubmitted work in progress, and providing a stable name for > +each unsubmitted change. > + > +Users can exchange edit histories by pushing and fetching changes. > + > +New commands will be introduced for manipulating changes and resolving > +divergence between them. Existing commands that create commits will be updated > +to modify the meta-commit graph and create changes where necessary. > + > +Example usage > +------------- > +# First create three dependent changes > +$ echo foo>bar.txt && git add . > +$ git commit -m "This is a test" > +created change metas/this_is_a_test > +$ echo foo2>bar2.txt && git add . > +$ git commit -m "This is also a test" > +created change metas/this_is_also_a_test > +$ echo foo3>bar3.txt && git add . > +$ git commit -m "More testing" > +created change metas/more_testing > + > +# List all our changes in progress > +$ git change list > +metas/this_is_a_test > +metas/this_is_also_a_test > +* metas/more_testing > +metas/some_change_already_merged_upstream > + > +# Now modify the earliest change, using its stable name > +$ git reset --hard metas/this_is_a_test > +$ echo morefoo>>bar.txt && git add . && git commit --amend --no-edit > + > +# Use git-evolve to fix up any dependent changes > +$ git evolve > +rebasing metas/this_is_also_a_test onto metas/this_is_a_test > +rebasing metas/more_testing onto metas/this_is_also_a_test > +Done > + > +# Use git-obslog to view the history of the this_is_a_test change > +$ git log --obslog > +93f110 metas/this_is_a_test@{0} commit (amend): This is a test > +930219 metas/this_is_a_test@{1} commit: This is a test > + > +# Now create an unrelated change > +$ git reset --hard origin/master > +$ echo newchange>unrelated.txt && git add . > +$ git commit -m "Unrelated change" > +created change metas/unrelated_change > + > +# Fetch the latest code from origin/master and use git-evolve > +# to rebase all dependent changes. > +$ git fetch origin master > +$ git evolve origin/master > +deleting metas/some_change_already_merged_upstream > +rebasing metas/this_is_a_test onto origin/master > +rebasing metas/this_is_also_a_test onto metas/this_is_a_test > +rebasing metas/more_testing onto metas/this_is_also_a_test > +rebasing metas/unrelated_change onto origin/master > +Conflict detected! Resolve it and then use git evolve --continue to resume. > + > +# Sort out the conflict > +$ git mergetool > +$ git evolve origin/master > +Done > + > +# Share the full history of edits for the this_is_a_test change > +# with a review server > +$ git push origin metas/this_is_a_test:refs/for/master > +# Share the lastest commit for “Unrelated change”, without history > +$ git push origin HEAD:refs/for/master > + > +Detailed design > +=============== > +Obsolescence information is stored as a graph of meta-commits. A meta-commit is > +a specially-formatted merge commit that describes how one commit was created > +from others. > + > +Meta-commits look like this: > + > +$ git cat-file -p <example_meta_commit> > +tree 4b825dc642cb6eb9a060e54bf8d69288fbee4904 > +parent aa7ce55545bf2c14bef48db91af1a74e2347539a > +parent d64309ee51d0af12723b6cb027fc9f195b15a5e9 > +parent 7e1bbcd3a0fa854a7a9eac9bf1eea6465de98136 > +author Stefan Xenos <sxenos@gmail.com> 1540841596 -0700 > +committer Stefan Xenos <sxenos@gmail.com> 1540841596 -0700 > +parent-type c r o > + > +This says “commit aa7ce555 makes commit d64309ee obsolete. It was created by > +cherry-picking commit 7e1bbcd3”. > + > +The tree for meta-commits is always the empty tree, but future versions of git > +may attach other trees here. For forward-compatibility fsck should ignore such > +trees if found on future repository versions. This will allow future versions of > +git to add metadata to the meta-commit tree without breaking forwards > +compatibility. > + > +The commit comment for a meta-commit is an auto-generated user-readable string > +describing the command that produced the meta commit. These strings are shown > +to the user when they view the obslog. > + > +Parent-type > +----------- > +The “parent-type” field in the commit header identifies a commit as a > +meta-commit and indicates the meaning for each of its parents. It is never > +present for normal commits. It contains a space-deliminated list of enum values > +whose order matches the order of the parents. Possible parent types are: > + > +- c: (content) the content parent identifies the commit that this meta-commit is > + describing. > +- r: (replaced) indicates that this parent is made obsolete by the content > + parent. > +- o: (origin) indicates that the content parent was generated by cherry-picking > + this parent. > +- a: (abandoned) used in place of a content parent for abandoned changes. Points > + to the final content commit for the change at the time it was abandoned. > + > +There must be exactly one content or abandoned parent for each meta-commit and > +it is always the first parent. The content commit will always be a normal commit > +and not a meta-commit. However, future versions of git may create meta-commits > +for other meta-commits and the fsck tool must be aware of this for forwards > +compatibility. > + > +A meta-commit can have zero or more replaced parents. An amend operation creates > +a single replaced parent. A merge used to resolve divergence (see divergence, > +below) will create multiple replaced parents. A meta-commit may have no > +replaced parents if it describes a cherry-pick or squash merge that copies one > +or more commits but does not replace them. > + > +A meta-commit can have zero or more origin parents. A cherry-pick creates a > +single origin parent. Certain types of squash merge will create multiple origin > +parents. Origin parents don't directly cause their origin to become obsolete, > +but are used when computing blame or locating a merge base. The section > +on obsolescence over cherry-picks describes how the evolve command uses > +origin parents. > + > +A replaced parent or origin parent may be either a normal commit (indicating > +the oldest-known version of a change) or another meta-commit (for a change that > +has already been modified one or more times). > + > +The parent-type field needs to go after the committer field since git's rules > +for forwards-compatibility require that new fields to be at the end of the > +header. Putting a new field in the middle of the header would break fsck. > + > +The presence of an abandoned parent indicates that the change should be pruned > +by the evolve command, and removed from the repository's history. Any follow-up > +changes should rebased onto the parent of the pruned commit. The abandoned > +parent points to the version of the change that should be restored if the user > +attempts to restore the change. > + > +Changes > +------- > +A branch of meta-commits describes how a commit was produced and what previous > +commits it is based on. It is also an identifier for a thing the user is > +currently working on. We refer to such a meta-branch as a change. > + > +Local changes are stored in the new refs/metas namespace. Remote changes are > +stored in the refs/remote/<remotename>/metas namespace. > + > +The list of changes in refs/metas is more than just a mechanism for the evolve > +command to locate obsolete commits. It is also a convenient list of all of a > +user’s work in progress and their current state - a list of things they’re > +likely to want to come back to. > + > +Strictly speaking, it is the presence of the branch in the refs/metas namespace > +that marks a branch as being a change, not the fact that it points to a > +metacommit. Metacommits are only created when a commit is amended or rebased, so > +in the case where a change points to a commit that has never been modified, the > +change points to that initial commit rather than a metacommit. > + > +Changes are also stored in the refs/hiddenmetas namespace. Hiddenmetas holds > +metadata for historical changes that are not currently in progress by the user. > +Commands like filter-branch and other bulk import commands create metadata in > +this namespace. > + > +Note that the changes in hiddenmetas get special treatment in several ways: > + > +- They are not cleaned up automatically once merged, since it is expected that > + they refer to historical changes. > +- User commands that modify changes don't append to these changes as they would > + to a change in refs/metas. > +- They are not displayed when the user lists their local changes. > + > +Obsolescence > +------------ > +A commit is considered obsolete if it is reachable from the “replaces” edges > +anywhere in the history of a change and it isn’t the head of that change. > +Commits may be the content for 0 or more meta-commits. If the same commit > +appears in multiple changes, it is not obsolete if it is the head of any of > +those changes. > + > +Note that there is an exception to this rule. The metas namespace takes > +precedence over the hiddenmetas namespace for the purpose of obsolescence. That > +is, if a change appears in a replaces edge of a change in the metas namespace, > +it is obsolete even if it also appears as the head of a change in the > +hiddenmetas namespace. > + > +This special case prevents the hiddenmetas namespace from creating divergence > +with the user's work in progress, and allows the user to resolve historical > +divergence by creating new changes in the metas namespace. > + > +Divergence > +---------- > +From the user’s perspective, two changes are divergent if they both ask for > +different replacements to the same commit. More precisely, a target commit is > +considered divergent if there is more than one commit at the head of a change in > +refs/metas that leads to the target commit via an unbroken chain of “replaces” > +parents. > + > +Much like a merge conflict, divergence is a situation that requires user > +intervention to resolve. The evolve command will stop when it encounters > +divergence and prompt the user to resolve the problem. Users can solve the > +problem in several ways: > + > +- Discard one of the changes (by deleting its change branch). > +- Merge the two changes (producing a single change branch). > +- Copy one of the changes (keep both commits, but one of them gets a new > + metacommit appended to its history that is connected to its predecessor via an > + origin edge rather than a replaces edge. That new change no longer obsoletes > + the original.) > + > +Obsolescence across cherry-picks > +-------------------------------- > +By default the evolve command will treat cherry-picks and squash merges as being > +completely separate from the original. Further amendments to the original commit > +will have no effect on the cherry-picked copy. However, this behavior may not be > +desirable in all circumstances. > + > +The evolve command may at some point support an option to look for cases where > +the source of a cherry-pick or squash merge has itself been amended, and > +automatically apply that same change to the cherry-picked copy. In such cases, > +it would traverse origin edges rather than ignoring them, and would treat a > +commit with origin edges as being obsolete if any of its origins were obsolete. > + > +Garbage collection > +------------------ > +For GC purposes, meta-commits are normal commits. Just as a commit causes its > +parents and tree to be retained, a meta-commit also causes its parents to be > +retained. > + > +Change creation > +--------------- > +Changes are created automatically whenever the user runs a command like “commit” > +that has the semantics of creating a new change. They also move forward > +automatically even if they’re not checked out. For example, whenever the user > +runs a command like “commit --amend” that modifies a commit, all branches in > +refs/metas that pointed to the old commit move forward to point to its > +replacement instead. This also happens when the user is working from a detached > +head. > + > +This does not mean that every commit has a corresponding change. By default, > +changes only exist for recent locally-created commits. Users may explicitly pull > +changes from other users or keep their changes around for a long time, but > +either behavior requires a user to opt-in. Code review systems like gerrit may > +also choose to keep changes around forever. > + > +Note that the changes in refs/metas serve a dual function as both a way to > +identify obsolete changes and as a way for the user to keep track of their work > +in progress. If we were only concerned with identifying obsolete changes, it > +would be sufficient to create the change branch lazily the first time a commit > +is obsoleted. Addressing the second use - of refs/metas as a mechanism for > +keeping track of work in progress - is the reason for eagerly creating the > +change on first commit. > + > +Change naming > +------------- > +When a change is first created, the only requirement for its name is that it > +must be unique. Good names would also serve as useful mnemonics and be easy to > +type. For example, a short word from the commit message containing no numbers or > +special characters and that shows up with low frequency in other commit messages > +would make a good choice. > + > +Different users may prefer different heuristics for their change names. For this > +reason a new hook will be introduced to compute change names. Git will invoke > +the hook for all newly-created changes and will append a numeric suffix if the > +name isn’t unique. The default heuristics are not specified by this proposal and > +may change during implementation. > + > +Change deletion > +--------------- > +Changes are normally only interesting to a user while a commit is still in > +development and under review. Once the commit has submitted wherever it is > +going, its change can be discarded. > + > +The normal way of deleting changes makes this easy to do - changes are deleted > +by the evolve command when it detects that the change is present in an upstream > +branch. It does this in two ways: if the latest commit in a change either shows > +up in the branch history or the change becomes empty after a rebase, it is > +considered merged and the change is discarded. In this context, an “upstream > +branch” is any branch passed in as the upstream argument of the evolve command. > + > +In case this sometimes deletes a useful change, such automatic deletions are > +recorded in the reflog allowing them to be easily recovered. > + > +Sharing changes > +--------------- > +Change histories are shared by pushing or fetching meta-commits and change > +branches. This provides users with a lot of control of what to share and > +repository implementations with control over what to retain. > + > +Users that only want to share the content of a commit can do so by pushing the > +commit itself as they currently would. Users that want to share an edit history > +for the commit can push its change, which would point to a meta-commit rather > +than the commit itself if there is any history to share. Note that multiple > +changes can refer to the same commits, so it’s possible to construct and push a > +different history for the same commit in order to remove sensitive or irrelevant > +intermediate states. > + > +Imagine the user is working on a change “mychange” that is currently the latest > +commit on master. They have two ways to share it: > + > +# User shares just a commit without its history > +> git push origin master > + > +# User shares the full history of the commit to a review system > +> git push origin metas/mychange:refs/for/master > + > +# User fetches a collaborator’s modifications to their change > +> git fetch remotename metas/mychange > +# Which updates the ref remote/remotename/metas/mychange > + > +This will cause more intermediate states to be shared with the server than would > +have been shared previously. A review system like gerrit would need to keep > +track of which states had been explicitly pushed versus other intermediate > +states in order to de-emphasize (or hide) the extra intermediate states from the > +user interface. > + > +Merge-base > +---------- > +Merge-base will be changed to search the meta-commit graph for common ancestors > +as well as the commit graph, and will generally prefer results from the > +meta-commit graph over the commit graph. Merge-base will consider meta-commits > +from all changes, and will traverse both origin and obsolete edges. > + > +The reason for this is that - when merging two versions of the same commit > +together - an earlier version of that same commit will usually be much more > +similar than their common parent. This should make the workflow of collaborating > +on unsubmitted patches as convenient as the workflow for collaborating in a > +topic branch by eliminating repeated merges. > + > +Configuration > +------------- > +The core.enableChanges configuration variable enables the creation and update > +of change branches. This is enabled by default. > + > +User interface > +-------------- > +All git porcelain commands that create commits are classified as having one of > +four behaviors: modify, create, copy, or import. These behaviors are discussed > +in more detail below. > + > +Modify commands > +--------------- > +Modification commands (commit --amend, rebase) will mark the old commit as > +obsolete by creating a new meta-commit that references the old one as a > +replaced parent. In the event that multiple changes point to the same commit, > +this is done independently for every such change. > + > +More specifically, modifications work like this: > + > +1. Locate all existing changes for which the old commit is the content for the > + head of the change branch. If no such branch exists, create one that points > + to the old commit. Changes that include this commit in their history but not > + at their head are explicitly not included. > +2. For every such change, create a new meta-commit that references the new > + commit as its content and references the old head of the change as a > + replaced parent. > +3. Move the change branch forward to point to the new meta-commit. > + > +Copy commands > +------------- > +Copy commands (cherry-pick, merge --squash) create a new meta-commit that > +references the old commits as origin parents. Besides the fact that the new > +parents are tagged differently, copy commands work the same way as modify > +commands. > + > +Create commands > +--------------- > +Creation commands (commit, merge) create a new commit and a new change that > +points to that commit. The do not create any meta-commits. > + > +Import commands > +--------------- > +Import commands (fetch, pull) do not create any new meta-commits or changes > +unless that is specifically what they are importing. For example, the fetch > +command would update remote/origin/metas/change35 and fetch all referenced > +meta-commits if asked to do so directly, but it wouldn’t create any changes or > +meta-commits for commits discovered on the master branch when running “git fetch > +origin master”. > + > +Other commands > +-------------- > +Some commands don’t fit cleanly into one of the above categories. > + > +Semantically, filter-branch should be treated as a modify command, but doing so > +is likely to create a lot of irrelevant clutter in the changes namespace and the > +large number of extra change refs may introduce performance problems. We > +recommend treating filter-branch as an import command initially, but making it > +behave more like a modify command in future follow-up work. One possible > +solution may be to treat commits that are part of existing changes as being > +modified but to avoid creating changes for other rewritten changes. Another > +solution may be to record the modifications as changes in the hiddenmetas > +namespace. > + > +Once the evolve command can handle obsolescence across cherry-picks, such > +cherry-picks will result in a hybrid move-and-copy operation. It will create > +cherry-picks that replace other cherry-picks, which will have both origin edges > +(pointing to the new source commit being picked) and replacement edges (pointing > +to the previous cherry-pick being replaced). > + > +Evolve > +------ > +The evolve command performs the correct sequence of rebases such that no change > +has an obsolete parent. The syntax looks like this: > + > +git evolve [upstream…] > + > +It takes an optional list of upstream branches. All changes whose parent shows > +up in the history of one of the upstream branches will be rebased onto the > +upstream branch before resolving obsolete parents. > + > +Any change whose latest state is found in an upstream branch (or that ends up > +empty after rebase) will be deleted. This is the normal mechanism for deleting > +changes. Changes are created automatically on the first commit, and are deleted > +automatically when evolve determines that they’ve been merged upstream. > + > +Orphan commits are commits with obsolete parents. The evolve command then > +repeatedly rebases orphan commits with non-orphan parents until there are either > +no orphan commits left, or a merge conflict is discovered. It will also > +terminate if it detects a divergent parent or a cycle that can't be resolved > +using any of the enabled transformations. > + > +When evolve discovers divergence, it will first check if it can resolve the > +divergence automatically using one of its enabled transformations. Supported > +transformations are: > + > +- Check if the user has already merged the divergent changes in a follow-up > + change. That is, look for an existing merge in a follow-up change where all > + the parents are divergent versions of the same change. Squash that merge with > + its parents and use the result as the resolution for the divergence. > + > +- Attempt to auto-merge all the divergent changes (disabled by default). > + > +Each of the transformations can be enabled or disabled by command line options. > + > +Cycles can occur when two changes reference one another as parents. This can > +happen when both changes use an obsolete version of the other change as their > +parent. Although there are never cycles in the commit graph, users can create > +cycles in the change graph by rebasing changes onto obsolete commits. The evolve > +command has a transformation that will detect and break cycles by arbitrarily > +picking one of the changes to go first. If this generates a merge conflict, > +it tries each of the other changes in sequence to see if any ordering merges > +cleanly. If no possible ordering merges cleanly, it picks one and terminates > +to let the user resolve the merge conflict. > + > +If the working tree is dirty, evolve will attempt to stash the user's changes > +before applying the evolve and then reapply those changes afterward, in much > +the same way as rebase --autostash does. > + > +Checkout > +-------- > +Running checkout on a change by name has the same effect as checking out a > +detached head pointing to the latest commit on that change-branch. There is no > +need to ever have HEAD point to a change since changes always move forward when > +necessary, no matter what branch the user has checked out > + > +Meta-commits themselves cannot be checked out by their hash. > + > +Reset > +----- > +Resetting a branch to a change by name is the same as resetting to the content > +(or abandoned) commit at that change’s head. > + > +Commit > +------ > +Commit --amend gets modify semantics and will move existing changes forward. The > +normal form of commit gets create semantics and will create a new change. > + > +$ touch foo && git add . && git commit -m "foo" && git tag A > +$ touch bar && git add . && git commit -m "bar" && git tag B > +$ touch baz && git add . && git commit -m "baz" && git tag C > + > +This produces the following commits: > +A(tree=[foo]) > +B(tree=[foo, bar], parent=A) > +C(tree=[foo, bar, baz], parent=B) > + > +...along with three changes: > +metas/foo = A > +metas/bar = B > +metas/baz = C > + > +Running commit --amend does the following: > +$ git checkout B > +$ touch zoom && git add . && git commit --amend -m "baz and zoom" > +$ git tag D > + > +Commits: > +A(tree=[foo]) > +B(tree=[foo, bar], parent=A) > +C(tree=[foo, bar, baz], parent=B) > +D(tree=[foo, bar, zoom], parent=A) > +Dmeta(content=D, obsolete=B) > + > +Changes: > +metas/foo = A > +metas/bar = Dmeta > +metas/baz = C > + > +Merge > +----- > +Merge gets create, modify, or copy semantics based on what is being merged and > +the options being used. > + > +The --squash version of merge gets copy semantics (it produces a new change that > +is marked as a copy of all the original changes that were squashed into it). > + > +The “modify” version of merge replaces both of the original commits with the > +resulting merge commit. This is one of the standard mechanisms for resolving > +divergence. The parents of the merge commit are the parents of the two commits > +being merged. The resulting commit will not be a merge commit if both of the > +original commits had the same parent or if one was the parent of the other. > + > +The “create” version of merge creates a new change pointing to a merge commit > +that has both original commits as parents. The result is what merge produces now > +- a new merge commit. However, this version of merge doesn’t directly resolve > +divergence. > + > +To select between these two behaviors, merge gets new “--amend” and “--noamend” > +options which select between the “create” and “modify” behaviors respectively, > +with noamend being the default. > + > +For example, imagine we created two divergent changes like this: > + > +$ touch foo && git add . && git commit -m "foo" && git tag A > +$ touch bar && git add . && git commit -m "bar" && git tag B > +$ touch baz && git add . && git commit --amend -m "bar and baz" > +$ git tag C > +$ git checkout B > +$ touch bam && git add . && git commit --amend -m "bar and bam" > +$ git tag D > + > +At this point the commit graph looks like this: > + > +A(tree=[foo]) > +B(tree=[bar], parent=A) > +C(tree=[bar, baz], parent=A) > +D(tree=[bar, bam], parent=A) > +Cmeta(content=C, obsoletes=B) > +Dmeta(content=D, obsoletes=B) > + > +There would be three active changes with heads pointing as follows: > + > +metas/changeA=A > +metas/changeB=Cmeta > +metas/changeB2=Dmeta > + > +ChangeB and changeB2 are divergent at this point. Lets consider what happens if > +perform each type of merge between changeB and changeB2. > + > +Merge example: Amend merge > +One way to resolve divergent changes is to use an amend merge. Recall that HEAD > +is currently pointing to D at this point. > + > +$ git merge --amend metas/changeB > + > +Here we’ve asked for an amend merge since we’re trying to resolve divergence > +between two versions of the same change. There are no conflicts so we end up > +with this: > + > +E(tree=[bar, baz, bam], parent=A) > +Emeta(content=E, obsoletes=[Cmeta, Dmeta]) > + > +With the following branches: > + > +metas/changeA=A > +metas/changeB=Emeta > +metas/changeB2=Emeta > + > +Notice that the result of the “amend merge” is a replacement for C and D rather > +than a new commit with C and D as parents (as a normal merge would have > +produced). The parents of the amend merge are the parents of C and D which - in > +this case - is just A, so the result is not a merge commit. Also notice that > +changeB and changeB2 are now aliases for the same change. > + > +Merge example: Noamend merge > +Consider what would have happened if we’d used a noamend merge instead. Recall > +that HEAD was at D and our branches looked like this: > + > +metas/changeA=A > +metas/changeB=Cmeta > +metas/changeB2=Dmeta > + > +$ git merge --noamend metas/changeB > + > +That would produce the sort of merge we’d normally expect today: > + > +F(tree=[bar, baz, bam], parent=[C, D]) > + > +And our changes would look like this: > +metas/changeA=A > +metas/changeB=Cmeta > +metas/changeB2=Dmeta > +metas/changeF=F > + > +In this case, changeB and changeB2 are still divergent and we’ve created a new > +change for our merge commit. However, this is just a temporary state. The next > +time we run the “evolve” command, it will discover the divergence but also > +discover the merge commit F that resolves it. Evolve will suggest converting F > +into an amend merge in order to resolve the divergence and will display the > +command for doing so. > + > +Rebase > +------ > +In general the rebase command is treated as a modify command. When a change is > +rebased, the new commit replaces the original. > + > +Rebase --abort is special. Its intent is to restore git to the state it had > +prior to running rebase. It should move back any changes to point to the refs > +they had prior to running rebase and delete any new changes that were created as > +part of the rebase. To achieve this, rebase will save the state of all changes > +in refs/metas prior to running rebase and will restore the entire namespace > +after rebase completes (deleting any newly-created changes). Newly-created > +metacommits are left in place, but will have no effect until garbage collected > +since metacommits are only used if they are reachable from refs/metas. > + > +Change > +------ > +The “change” command can be used to list, rename, reset or delete change. It has > +a number of subcommands. > + > +The "list" subcommand lists local changes. If given the -r argument, it lists > +remote changes. > + > +The "rename" subcommand renames a change, given its old and new name. If the old > +name is omitted and there is exactly one change pointing to the current HEAD, > +that change is renamed. If there are no changes pointing to the current HEAD, > +one is created with the given name. > + > +The "forget" subcommand deletes a change by deleting its ref from the metas/ > +namespace. This is the normal way to delete extra aliases for a change if the > +change has more than one name. By default, this will refuse to delete the last > +alias for a change if there are any other changes that reference this change as > +a parent. > + > +The "update" subcommand adds a new state to a change. It uses the default > +algorithm for assigning change names. If the content commit is omitted, HEAD is > +used. If given the optional --force argument, it will overwrite any existing > +change of the same name. This latter form of "update" can be used to effectively > +reset changes. > + > +The "update" command can accept any number of --origin and --replace arguments. > +If any are present, the resulting change branch will point to a metacommit > +containing the given origin and replacement edges. > + > +The "abandon" command deletes a change using obsolescence markers. It marks the > +change as being obsolete and having been replaced by its parent. If given no > +arguments, it applies to the current commit. Running evolve will cause any > +abandoned changes to be removed from the branch. Any child changes will be > +reparented on top of the parent of the abandoned change. If the current change > +is abandoned, HEAD will move to point to its parent. > + > +The "restore" command restores a previously-abandoned change. > + > +The "prune" command deletes all obsolete changes and all changes that are > +present in the given branch. Note that such changes can be recovered from the > +reflog. > + > +Combined with the GC protection that is offered, this is intended to facilitate > +a workflow that relies on changes instead of branches. Users could choose to > +work with no local branches and use changes instead - both for mailing list and > +gerrit workflows. > + > +Log > +--- > +When a commit is shown in git log that is part of a change, it is decorated with > +extra change information. If it is the head of a change, the name of the change > +is shown next to the list of branches. If it is obsolete, it is decorated with > +the text “obsolete, <n> commits behind <changename>”. > + > +Log gets a new --obslog argument indicating that the obsolescence graph should > +be followed instead of the commit graph. This also changes the default > +formatting options to make them more appropriate for viewing different > +iterations of the same commit. > + > +Pull > +---- > + > +Pull gets an --evolve argument that will automatically attempt to run "evolve" > +on any affected branches after pulling. > + > +We also introduce an "evolve" enum value for the branch.<name>.rebase config > +value. When set, the evolve behavior will happen automatically for that branch > +after every pull even if the --evolve argument is not used. > + > +Next > +---- > + > +The "next" command will reset HEAD to a non-obsolete commit that refers to this > +change as its parent. If there is more than one such change, the user will be > +prompted. If given the --evolve argument, the next commit will be evolved if > +necessary first. > + > +The "next" command can be thought of as the opposite of > +"git reset --hard HEAD^" in that it navigates to a child commit rather than a > +parent. > + > +Prev > +---- > + > +The "prev" command will reset HEAD to the latest version of the parent change. > +If the parent change isn't obsolete, this is equivalent to > +"git reset --hard HEAD^". If the parent commit is obsolete, it resets to the > +latest replacement for the parent commit. > + > +Other options considered > +======================== > +We considered several other options for storing the obsolescence graph. This > +section describes the other options and why they were rejected. > + > +Commit header > +------------- > +Add an “obsoletes” field to the commit header that points backwards from a > +commit to the previous commits it obsoletes. > + > +Pros: > +- Very simple > +- Easy to traverse from a commit to the previous commits it obsoletes. > +Cons: > +- Adds a cost to the storage format, even for commits where the change history > + is uninteresting. > +- Unconditionally prevents the change history from being garbage collected. > +- Always causes the change history to be shared when pushing or pulling changes. > + > +Git notes > +--------- > +Instead of storing obsolescence information in metacommits, the metacommit > +content could go in a new notes namespace - say refs/notes/metacommit. Each note > +would contain the list of obsolete and origin parents. An automerger could > +be supplied to make it easy to merge the metacommit notes from different remotes. > + > +Pros: > +- Easy to locate all commits obsoleted by a given commit (since there would only > + be one metacommit for any given commit). > +Cons: > +- Wrong GC behavior (obsolete commits wouldn’t automatically be retained by GC) > + unless we introduced a special case for these kinds of notes. > +- No way to selectively share or pull the metacommits for one specific change. > + It would be all-or-nothing, which would be expensive. This could be addressed > + by changes to the protocol, but this would be invasive. > +- Requires custom auto-merging behavior on fetch. > + > +Tags > +---- > +Put the content of the metacommit in a message attached to tag on the > +replacement commit. This is very similar to the git notes approach and has the > +same pros and cons. > + > +Simple forward references > +------------------------- > +Record an edge from an obsolete commit to its replacement in this form: > + > +refs/obsoletes/<A> > + > +pointing to commit <B> as an indication that B is the replacement for the > +obsolete commit A. > + > +Pros: > +- Protects <B> from being garbage collected. > +- Fast lookup for the evolve operation, without additional search structures > + (“what is the replacement for <A>?” is very fast). > + > +Cons: > +- Can’t represent divergence (which is a P0 requirement). > +- Creates lots of refs (which can be inefficient) > +- Doesn’t provide a way to fetch only refs for a specific change. > +- The obslog command requires a search of all refs. > + > +Complex forward references > +-------------------------- > +Record an edge from an obsolete commit to its replacement in this form: > + > +refs/obsoletes/<change_id>/obs<A>_<B> > + > +Pointing to commit <B> as an indication that B is the replacement for obsolete > +commit A. > + > +Pros: > +- Permits sharing and fetching refs for only a specific change. > +- Supports divergence > +- Protects <B> from being garbage collected. > + > +Cons: > +- Creates lots of refs, which is inefficient. > +- Doesn’t provide a good lookup structure for lookups in either direction. > + > +Backward references > +------------------- > +Record an edge from a replacement commit to the obsolete one in this form: > + > +refs/obsolescences/<B> > + > +Cons: > +- Doesn’t provide a way to resolve divergence (which is a P0 requirement). > +- Doesn’t protect <B> from being garbage collected (which could be fixed by > + combining this with a refs/metas namespace, as in the metacommit variant). > + > +Obsolescences file > +------------------ > +Create a custom file (or files) in .git recording obsolescences. > + > +Pros: > +- Can store exactly the information we want with exactly the performance we want > + for all operations. For example, there could be a disk-based hashtable > + permitting constant time lookups in either direction. > + > +Cons: > +- Handling GC, pushing, and pulling would all require custom solutions. GC > + issues could be addressed with a repository format extension. > + > +Squash points > +------------- > +We treat changes like topic branches, and use special squash points to mark > +places in the commit graph that separate changes. > + > +We create and update change branches in refs/metas at the same time we > +would have in the metacommit proposal. However, rather than pointing to a > +metacommit branch they point to normal commits and are treated as “squash > +points” - markers for sequences of commits intended to be squashed together on > +submission. > + > +Amends and rebases work differently than they do now. Rather than actually > +containing the desired state of a commit, they contain a delta from the previous > +version along with a squash point indicating that the preceding changes are > +intended to be squashed on submission. Specifically, amends would become new > +changes and rebases would become merge commits with the old commit and new > +parent as parents. > + > +When the changes are finally submitted, the squashes are executed, producing the > +final version of the commit. > + > +In addition to the squash points, git would maintain a set of “nosquash” tags > +for commits that were used as ancestors of a change that are not meant to be > +included in the squash. > + > +For example, if we have this commit graph: > + > +A(...) > +B(parent=A) > +C(parent=B) > + > +...and we amend B to produce D, we’d get: > + > +A(...) > +B(parent=A) > +C(parent=B) > +D(parent=B) > + > +...along with a new change branch indicating D should be squashed with its > +parents when submitted: > + > +metas/changeB = D > +metas/changeC = C > + > +We’d also create a nosquash tag for A indicating that A shouldn’t be included > +when changeB is squashed. > + > +If a user amends the change again, they’d get: > + > +A(...) > +B(parent=A) > +C(parent=B) > +D(parent=B) > +E(parent=D) > + > +metas/changeB = E > +metas/changeC = C > + > +Pros: > +- Good GC behavior. > +- Provides a natural way to share changes (they’re just normal branches). > +- Merge-base works automatically without special cases. > +- Rewriting the obslog would be easy using existing git commands. > +- No new data types needed. > +Cons: > +- No way to connect the squashed version of a change to the original, so no way > + to automatically clean up old changes. This also means users lose all benefits > + of the evolve command if they prematurely squash their commits. This may occur > + if a user thinks a change is ready for submission, squashes it, and then later > + discovers an additional change to make. > +- Histories would look very cluttered (users would see all previous edits to > + their commit in the commit log, and all previous rebases would show up as > + merges). Could be quite hard for users to tell what is going on. (Possible > + fix: also implement a new smart log feature that displays the log as though > + the squashes had occurred). > +- Need to change the current behavior of current commands (like amend and > + rebase) in ways that will be unexpected to many users. > -- > gitgitgadget >
"Stefan Xenos via GitGitGadget" <gitgitgadget@gmail.com> writes: > From: Stefan Xenos <sxenos@google.com> [the above address should bounce, and has been removed from cc: list] > +Background > +========== > +Imagine you have three sequential changes up for review and you receive feedback > +that requires editing all three changes. We'll define the word "change" > +formally later, but for the moment let's say that a change is a work-in-progress > +whose final version will be submitted as a commit in the future. > + > +While you're editing one change, more feedback arrives on one of the others. > +What do you do? > + > +The evolve command is a convenient way to work with chains of commits that are > +under review. Whenever you rebase or amend a commit, the repository remembers > +that the old commit is obsolete and has been replaced by the new one. Then, at > +some point in the future, you can run "git evolve" and the correct sequence of > +rebases will occur in the correct order such that no commit has an obsolete > +parent. > + > +Part of making the "evolve" command work involves tracking the edits to a commit > +over time, which is why we need an change graph. However, the change > +graph will also bring other benefits: It would be assuring to hear that "change graph" will also be defined and explained formally later, just like "change" will in the previous paragraph. We will later see mention of "metacommits" and "meta-commits" in this document, and I am guessing both of them are quasi-synonyms to "change graph". If that is true, it is better to stick to a single terminology. > +Goals > +----- > +Legend: Goals marked with P0 are required. Goals marked with Pn should be > +attempted unless they interfere with goals marked with Pn-1. > + > +P0. All commands that modify commits (such as the normal commit --amend or > + rebase command) should mark the old commit as being obsolete and replaced by > + the new one. No additional commands should be required to keep the > + change graph up-to-date. > +P0. Any commit that may be involved in a future evolve command should not be > + garbage collected. Specifically: > + - Commits that obsolete another should not be garbage collected until > + user-specified conditions have occurred and the change has expired from > + the reflog. User specified conditions for removing changes include: > + - The user explicitly deleted the change. > + - The change was merged into a specific branch. > + - Commits that have been obsoleted by another should not be garbage > + collected if any of their replacements are still being retained. > +P0. A commit can be obsoleted by more than one replacement (called divergence). > +P0. Users must be able to resolve divergence (convergence). P0: a single parent commit should keep only one parent. IOW, the "change graph" implementation should not contaminate the end-result commit in the regular part of the history, which is the product of the final iteration of a "change" IOW ... > +P2. It should be possible to discard part or all of the change graph > + without discarding the commits themselves that are already present in > + branches and the reflog. ... this item should be P0. > +Overview > +======== > +We introduce the notion of “meta-commits” which describe how one commit was Random appearance of smart quotes are annoying. We'll be formatting the doc via AsciiDoc, so let's stick to vanilla double or single quotes. > +created from other commits. A branch of meta-commits is known as a change. > +Changes are created and updated automatically whenever a user runs a command > +that creates a commit. They are used for locating obsolete commits, providing a > +list of a user’s unsubmitted work in progress, and providing a stable name for > +each unsubmitted change. Can "change graph" also be defined and explained here, too? Or if it is pretty much a synonym to "a branch of meta-commits", then perhaps the document does not have to introduce the term "change graph" and still stay understandable? > +Detailed design > +=============== > +Obsolescence information is stored as a graph of meta-commits. A meta-commit is > +a specially-formatted merge commit that describes how one commit was created > +from others. > + > +Meta-commits look like this: > + > +$ git cat-file -p <example_meta_commit> > +tree 4b825dc642cb6eb9a060e54bf8d69288fbee4904 > +parent aa7ce55545bf2c14bef48db91af1a74e2347539a > +parent d64309ee51d0af12723b6cb027fc9f195b15a5e9 > +parent 7e1bbcd3a0fa854a7a9eac9bf1eea6465de98136 > +author Stefan Xenos <sxenos@gmail.com> 1540841596 -0700 > +committer Stefan Xenos <sxenos@gmail.com> 1540841596 -0700 > +parent-type c r o > + > +This says “commit aa7ce555 makes commit d64309ee obsolete. It was created by > +cherry-picking commit 7e1bbcd3”. > + > +The tree for meta-commits is always the empty tree, but future versions of git > +may attach other trees here. For forward-compatibility fsck should ignore such > +trees if found on future repository versions. This will allow future versions of > +git to add metadata to the meta-commit tree without breaking forwards > +compatibility. Not clear why "trees" need to be ignored only by fsck but not others like fetch/push, and I'd strongly advise against making such a special case. If they are missing, they are missing and should be reported as corruptoin, and if you do not like it, do not add a missing tree. > +Parent-type > +----------- > +The “parent-type” field in the commit header identifies a commit as a > +meta-commit and indicates the meaning for each of its parents. It is never > +present for normal commits. It contains a space-deliminated list of enum values > +whose order matches the order of the parents. Possible parent types are: > +- c: (content) the content parent identifies the commit that this meta-commit is > + describing. > +- r: (replaced) indicates that this parent is made obsolete by the content > + parent. > +- o: (origin) indicates that the content parent was generated by cherry-picking > + this parent. > +- a: (abandoned) used in place of a content parent for abandoned changes. Points > + to the final content commit for the change at the time it was abandoned. Don't be cute with parent-type using single letters. You'll thank me later when you need two types that share the first letter. > +A meta-commit can have zero or more origin parents. A cherry-pick creates a > +single origin parent. Certain types of squash merge will create multiple origin > +parents. Origin parents don't directly cause their origin to become obsolete, > +but are used when computing blame or locating a merge base. The section > +on obsolescence over cherry-picks describes how the evolve command uses > +origin parents. Should it make a difference among doing these operations? - running "commit --amend" after "cherry-pick --no-commit" possibly with editing - running "commit --amend" after manually editing the same way, and - running "commit --amend" after "cherry-pick", possibly with editing? It seems that the first two will not be captured while the last one leaves 'origin'. What should happen after running "commit --amend" after "apply --index" possibly with editing? What's the point of giving 'origin' only for "cherry-pick" and squash merge? I am wondering if we want to record contributions sourced from an e-mailed patch from elsewhere (currently people use external services like patchwork to do this)? For the purpose of discussing "evolve", should "rebase" (with or without "-i") be treated pretty much the same as a series of "cherry-pick" mixed with "commit --amend" (possibly preceded with a manual edit), followed by finally replacing the tip of the branch? In the end result, the replaced commits after a "rebase" become accessible only from reflog, but other than that, these two bulk transplanting operations shouldn't be all that different. > +The parent-type field needs to go after the committer field since git's rules > +for forwards-compatibility require that new fields to be at the end of the > +header. Putting a new field in the middle of the header would break fsck. You can do without introducing a new header to avoid compatibility issue by recording the information in the body of the commit object, which would be even cleaner. > +Change deletion > +--------------- > +Changes are normally only interesting to a user while a commit is still in > +development and under review. Once the commit has submitted wherever it is > +going, its change can be discarded. > + > +The normal way of deleting changes makes this easy to do - changes are deleted > +by the evolve command when it detects that the change is present in an upstream > +branch. It does this in two ways: if the latest commit in a change either shows > +up in the branch history or the change becomes empty after a rebase, it is > +considered merged and the change is discarded. In this context, an “upstream > +branch” is any branch passed in as the upstream argument of the evolve command. > + > +In case this sometimes deletes a useful change, such automatic deletions are > +recorded in the reflog allowing them to be easily recovered. Deleting a useful change is recorded in the reflog? Isn't a change recorded as a ref in metas/ hierarchy? Doesn't the removal of such a ref remove its reflog as well? I guess the above silly questions come from the fact that the document does not make it clear reflog of what ref it is recorded. > +Modify commands > +--------------- > +Modification commands (commit --amend, rebase) will mark the old commit as > +obsolete by creating a new meta-commit that references the old one as a > +replaced parent. In the event that multiple changes point to the same commit, > +this is done independently for every such change. > + > +More specifically, modifications work like this: > + > +1. Locate all existing changes for which the old commit is the content for the > + head of the change branch. If no such branch exists, create one that points > + to the old commit. Changes that include this commit in their history but not > + at their head are explicitly not included. > +2. For every such change, create a new meta-commit that references the new > + commit as its content and references the old head of the change as a > + replaced parent. > +3. Move the change branch forward to point to the new meta-commit. > + > +Copy commands > +------------- > +Copy commands (cherry-pick, merge --squash) create a new meta-commit that > +references the old commits as origin parents. Besides the fact that the new > +parents are tagged differently, copy commands work the same way as modify > +commands. It is unclear what benefit we will get by separating "Copy" commands from "Modify" commands. "checkout A && cherry-pick B" may make a new copy of the edit the commit at the tip of branch B wanted to make at the tip of branch A, but "commit --amend" is the same, in that it makes a new copy of the edit the commit at the tip of the current branch wanted to make, and the original copy is available in both cases. It is just that the original of "cherry-pick B" is slightly easier to access (i.e. it is still at the tip of branch B, until the branch gains new commits on top of it) than the original of "commit --amend" (i.e. the user needs to know that @{1} is the previous state). Shouldn't all commands that create a new commit object using some existing material (i.e. not from scratch) be treated equally, without splitting them into two camps? IOW, the above explains that the new parents are tagged differently, but it does not explain why it is a good idea to do so.
"Stefan Xenos via GitGitGadget" <gitgitgadget@gmail.com> writes: > +Rebase > +------ > +In general the rebase command is treated as a modify command. When a change is > +rebased, the new commit replaces the original. > + > +Rebase --abort is special. Its intent is to restore git to the state it had > +prior to running rebase. It should move back any changes to point to the refs > +they had prior to running rebase and delete any new changes that were created as > +part of the rebase. To achieve this, rebase will save the state of all changes > +in refs/metas prior to running rebase and will restore the entire namespace > +after rebase completes (deleting any newly-created changes). Newly-created > +metacommits are left in place, but will have no effect until garbage collected > +since metacommits are only used if they are reachable from refs/metas. One thing that makes me nervous is how well your analysis capture "unusual" but still reasonable ways to use these commands, as the workflows of people are quite different. For example, I almost never do "git checkout topic && git rebase origin"; instead I would do "git checkout topic && git rebase origin HEAD^0" to first make a detached HEAD out of the topic, in order to have two copies explicitly available to be compared after "rebase" finishes. After doing so and get satisfied by the result of comparison between topic and HEAD, I may do "git checkout -B topic" to update. Would that leave exactly the same set of metacommits as the case where I didn't do the "first rebase the detached HEAD and then update the bracnh for real" and instead "rebase the topic" directly?
On 28/09/2022 23:20, Junio C Hamano wrote: > "Stefan Xenos via GitGitGadget" <gitgitgadget@gmail.com> writes: > >> +Rebase >> +------ >> +In general the rebase command is treated as a modify command. When a change is >> +rebased, the new commit replaces the original. >> + >> +Rebase --abort is special. Its intent is to restore git to the state it had >> +prior to running rebase. It should move back any changes to point to the refs >> +they had prior to running rebase and delete any new changes that were created as >> +part of the rebase. To achieve this, rebase will save the state of all changes >> +in refs/metas prior to running rebase and will restore the entire namespace >> +after rebase completes (deleting any newly-created changes). That wont work now that we have multiple worktrees. If a user starts two rebases of two different branches in two different worktrees and aborts one of them we loose the new meta-commits of both. Each rebase will need to track which refs under refs/metas/ it has updated (that will also save the overhead of copying the entire refs/metas/ subtree). > Newly-created >> +metacommits are left in place, but will have no effect until garbage collected >> +since metacommits are only used if they are reachable from refs/metas. > > One thing that makes me nervous is how well your analysis capture > "unusual" but still reasonable ways to use these commands, as the > workflows of people are quite different. > > For example, I almost never do "git checkout topic && git rebase > origin"; instead I would do "git checkout topic && git rebase origin > HEAD^0" to first make a detached HEAD out of the topic, in order to > have two copies explicitly available to be compared after "rebase" > finishes. After doing so and get satisfied by the result of > comparison between topic and HEAD, I may do "git checkout -B topic" > to update. Would that leave exactly the same set of metacommits as > the case where I didn't do the "first rebase the detached HEAD and > then update the bracnh for real" and instead "rebase the topic" > directly? As I understand it we have a ref under refs/metas/ for each commit. If that understanding is correct the two cases you describe should be recorded identically I think. Best Wishes Phillip
"Stefan Xenos via GitGitGadget" <gitgitgadget@gmail.com> writes: > +Background > +========== > +Imagine you have three sequential changes up for review and you receive feedback > +that requires editing all three changes. We'll define the word "change" > +formally later, but for the moment let's say that a change is a work-in-progress > +whose final version will be submitted as a commit in the future. [snip] > +Part of making the "evolve" command work involves tracking the edits to a commit > +over time, which is why we need an change graph. Reading later, I thought that a "change" is a connected subset of elements in the set of metacommits, so a "change" is already a graph. I'll mentally substitute "the concept of a change" for "an [sic] change graph" for now - hopefully that's correct. > +- It can be used as part of other high-level commands that combine or split > + changes. Is the current concept of a change suitable for combining and splitting? There is divergence, but that seems like a commit being modified in 2 different ways, not a commit being split into 2. > +Goals > +----- [snip] > +P0. A commit can be obsoleted by more than one replacement (called divergence). > +P0. Users must be able to resolve divergence (convergence). Is divergence important? It seems to me that both the internals and the UX could be simplified if we don't allow divergence, and systems like Gerrit don't have it either (as far as I know, in Gerrit, all commits bearing the same Change-Id just form a sequence in the order that they were pushed to the server, with no branching). > +Overview > +======== > +We introduce the notion of “meta-commits” which describe how one commit was > +created from other commits. A branch of meta-commits is known as a change. "Branch" here is confusing. Do you mean a connected set of meta-commits? ("Branch" has a specific meaning in Git - a ref of the form refs/heads/??. If you mean that refs of the form refs/metas/?? point to changes in a 1:1 manner, then the term "ref" is appropriate.) > +Example usage > +------------- > +# First create three dependent changes > +$ echo foo>bar.txt && git add . > +$ git commit -m "This is a test" > +created change metas/this_is_a_test > +$ echo foo2>bar2.txt && git add . > +$ git commit -m "This is also a test" > +created change metas/this_is_also_a_test > +$ echo foo3>bar3.txt && git add . > +$ git commit -m "More testing" > +created change metas/more_testing > + > +# List all our changes in progress > +$ git change list > +metas/this_is_a_test > +metas/this_is_also_a_test > +* metas/more_testing > +metas/some_change_already_merged_upstream > + > +# Now modify the earliest change, using its stable name > +$ git reset --hard metas/this_is_a_test > +$ echo morefoo>>bar.txt && git add . && git commit --amend --no-edit So up to here, I thought that we would have 2 refs metas/this_is_a_test2 and metas/this_is_a_test, with the latter's commit being one of the parents of the former's commit. (This is because we presumably need to be able to represent the situation in which the user checks out the original "This is a test" commit and modifies it, so we still need to hang on to the metas/this_is_a_test.) > +# Use git-evolve to fix up any dependent changes > +$ git evolve > +rebasing metas/this_is_also_a_test onto metas/this_is_a_test > +rebasing metas/more_testing onto metas/this_is_also_a_test > +Done So I'm surprised that there's no mention of this_is_a_test2 here. In addition, linearizing a change like this doesn't seem to be described in this document. Having said that, I don't think that linearizing changes is important to the goals of this "evolve" concept, so maybe one thing we can do is to not support it at all. Fast-forward... > +# Fetch the latest code from origin/master and use git-evolve > +# to rebase all dependent changes. > +$ git fetch origin master > +$ git evolve origin/master > +deleting metas/some_change_already_merged_upstream > +rebasing metas/this_is_a_test onto origin/master > +rebasing metas/this_is_also_a_test onto metas/this_is_a_test > +rebasing metas/more_testing onto metas/this_is_also_a_test > +rebasing metas/unrelated_change onto origin/master > +Conflict detected! Resolve it and then use git evolve --continue to resume. > + > +# Sort out the conflict > +$ git mergetool > +$ git evolve origin/master > +Done This is what I expected from the evolve mechanism, so that's great :-) The conflict resolution needs to be discussed further, though. It is superficially similar to rebase, but with rebase, the ref being rebased is only rewritten at the end of the process, so it is always possible to abort halfway. Here, multiple refs are written during the process, so it is not as easy to abort. > +Parent-type > +----------- > +The “parent-type” field in the commit header identifies a commit as a > +meta-commit and indicates the meaning for each of its parents. It is never > +present for normal commits. It contains a space-deliminated list of enum values > +whose order matches the order of the parents. Possible parent types are: > + > +- c: (content) the content parent identifies the commit that this meta-commit is > + describing. > +- r: (replaced) indicates that this parent is made obsolete by the content > + parent. > +- o: (origin) indicates that the content parent was generated by cherry-picking > + this parent. > +- a: (abandoned) used in place of a content parent for abandoned changes. Points > + to the final content commit for the change at the time it was abandoned. How would the "o" parent be useful to the user? > +Changes > +------- [snip] > +Changes are also stored in the refs/hiddenmetas namespace. Hiddenmetas holds > +metadata for historical changes that are not currently in progress by the user. > +Commands like filter-branch and other bulk import commands create metadata in > +this namespace. > + > +Note that the changes in hiddenmetas get special treatment in several ways: > + > +- They are not cleaned up automatically once merged, since it is expected that > + they refer to historical changes. > +- User commands that modify changes don't append to these changes as they would > + to a change in refs/metas. > +- They are not displayed when the user lists their local changes. The presence of refs/hiddenmetas further muddies the already unclear lifecycle of meta-commits and their refs. Non-hidden meta-commits get cleaned up when their latest commit appears upstream, so they may get deleted when the user doesn't expect it (especially if the user is using, say, a prefetch mechanism that downloads refs at night). We're adding to this a class of refs that don't get cleaned up at all. Besides the disk space taken by the meta-commits, having more refs typically reduces performance e.g. because all refs generally take part in packfile negotiation during fetching. (And they probably should continue with this behavior because sharing meta-commits is one of the features we want.) So I think that having a clear cleanup strategy is a good idea, and permanent archiving probably shouldn't be it. > +Change creation > +--------------- > +Changes are created automatically whenever the user runs a command like “commit” > +that has the semantics of creating a new change. They also move forward > +automatically even if they’re not checked out. For example, whenever the user > +runs a command like “commit --amend” that modifies a commit, all branches in > +refs/metas that pointed to the old commit move forward to point to its > +replacement instead. What happens in the following? $ echo "hello" >hello.txt $ git add hello.txt $ git commit -m "hello" $ git tag hello $ echo "one" >hello.txt $ git commit -a --amend # this updates refs/metas/hello $ git checkout hello $ echo "one" >hello.txt $ git commit -a --amend # does this update refs/metas/hello too? > +Sharing changes > +--------------- > +Change histories are shared by pushing or fetching meta-commits and change > +branches. This provides users with a lot of control of what to share and > +repository implementations with control over what to retain. > + > +Users that only want to share the content of a commit can do so by pushing the > +commit itself as they currently would. Users that want to share an edit history > +for the commit can push its change, which would point to a meta-commit rather > +than the commit itself if there is any history to share. Note that multiple > +changes can refer to the same commits, so it’s possible to construct and push a > +different history for the same commit in order to remove sensitive or irrelevant > +intermediate states. It looks difficult to remove such intermediate states, but maybe that doesn't have to be dealt with in the initial design. > +Checkout > +-------- > +Running checkout on a change by name has the same effect as checking out a > +detached head pointing to the latest commit on that change-branch. There is no > +need to ever have HEAD point to a change since changes always move forward when > +necessary, no matter what branch the user has checked out > + > +Meta-commits themselves cannot be checked out by their hash. This is the same behavior as for annotated tags, but I guess we can't use them because those can only have one referent.
diff --git a/Documentation/technical/evolve.txt b/Documentation/technical/evolve.txt new file mode 100644 index 00000000000..68ee2457e52 --- /dev/null +++ b/Documentation/technical/evolve.txt @@ -0,0 +1,1051 @@ +Evolve +====== + +Objective +========= +Create an "evolve" command to help users craft a high quality commit history. +Users can improve commits one at a time and in any order, then run git evolve to +rewrite their recent history to ensure everything is up-to-date. We track +amendments to a commit over time in a change graph. Users can share their +progress with others by exchanging their change graphs using the standard push, +fetch, and format-patch commands. + +Status +====== +This proposal has not been implemented yet. + +Background +========== +Imagine you have three sequential changes up for review and you receive feedback +that requires editing all three changes. We'll define the word "change" +formally later, but for the moment let's say that a change is a work-in-progress +whose final version will be submitted as a commit in the future. + +While you're editing one change, more feedback arrives on one of the others. +What do you do? + +The evolve command is a convenient way to work with chains of commits that are +under review. Whenever you rebase or amend a commit, the repository remembers +that the old commit is obsolete and has been replaced by the new one. Then, at +some point in the future, you can run "git evolve" and the correct sequence of +rebases will occur in the correct order such that no commit has an obsolete +parent. + +Part of making the "evolve" command work involves tracking the edits to a commit +over time, which is why we need an change graph. However, the change +graph will also bring other benefits: + +- Users can view the history of a change directly (the sequence of amends and + rebases it has undergone, orthogonal to the history of the branch it is on). +- It will be possible to quickly locate and list all the changes the user + currently has in progress. +- It can be used as part of other high-level commands that combine or split + changes. +- It can be used to decorate commits (in git log, gitk, etc) that are either + obsolete or are the tip of a work in progress. +- By pushing and pulling the change graph, users can collaborate more + easily on changes-in-progress. This is better than pushing and pulling the + commits themselves since the change graph can be used to locate a more + specific merge base, allowing for better merges between different versions of + the same change. +- It could be used to correctly rebase local changes and other local branches + after running git-filter-branch. +- It can replace the change-id footer used by gerrit. + +Goals +----- +Legend: Goals marked with P0 are required. Goals marked with Pn should be +attempted unless they interfere with goals marked with Pn-1. + +P0. All commands that modify commits (such as the normal commit --amend or + rebase command) should mark the old commit as being obsolete and replaced by + the new one. No additional commands should be required to keep the + change graph up-to-date. +P0. Any commit that may be involved in a future evolve command should not be + garbage collected. Specifically: + - Commits that obsolete another should not be garbage collected until + user-specified conditions have occurred and the change has expired from + the reflog. User specified conditions for removing changes include: + - The user explicitly deleted the change. + - The change was merged into a specific branch. + - Commits that have been obsoleted by another should not be garbage + collected if any of their replacements are still being retained. +P0. A commit can be obsoleted by more than one replacement (called divergence). +P0. Users must be able to resolve divergence (convergence). +P1. Users should be able to share chains of obsolete changes in order to + collaborate on WIP changes. +P2. Such sharing should be at the user’s option. That is, it should be possible + to directly share a change without also sharing the file states or commit + comments from the obsolete changes that led up to it, and the choice not to + share those commits should not require changing any commit hashes. +P2. It should be possible to discard part or all of the change graph + without discarding the commits themselves that are already present in + branches and the reflog. +P2. Provide sufficient information to replace gerrit's Change-Id footers. + +Similar technologies +-------------------- +There are some other technologies that address the same end-user problem. + +Rebase -i can be used to solve the same problem, but users can't easily switch +tasks midway through an interactive rebase or have more than one interactive +rebase going on at the same time. It can't handle the case where you have +multiple changes sharing the same parent when that parent needs to be rebased +and won't let you collaborate with others on resolving a complicated interactive +rebase. You can think of rebase -i as a top-down approach and the evolve command +as the bottom-up approach to the same problem. + +Several patch queue managers have been built on top of git (such as topgit, +stgit, and quilt). They address the same user need. However they also rely on +state managed outside git that needs to be kept in sync. Such state can be +easily damaged when running a git native command that is unaware of the patch +queue. They also typically require an explicit initialization step to be done by +the user which creates workflow problems. + +Mercurial implements a very similar feature in its EvolveExtension. The behavior +of the evolve command itself is very similar, but the storage format for the +change graph differs. In the case of mercurial, each change set can have one or +more obsolescence markers that point to other changesets that they replace. This +is similar to the "Commit Headers" approach considered in the other options +appendix. The approach proposed here stores obsolescence information in a +separate metacommit graph, which makes exchanging of obsolescence information +optional. + +Mercurial's default behavior makes it easy to find and switch between +non-obsolete changesets that aren't currently on any branch. We introduce the +notion of a new ref namespace that enables a similar workflow via a different +mechanism. Mercurial has the notion of changeset phases which isn't present +in git and creates new ways for a changeset to diverge. Git doesn't need +to deal with these issues, but it has to deal with the problems of picking an +upstream branch as a target for rebases and protecting obsolescence information +from GC. We also introduce some additional transformations (see +obsolescence-over-cherry-pick, below) that aren't present in the mercurial +implementation. + +Semi-related work +----------------- +There are other technologies that address different problems but have some +similarities with this proposal. + +Replacements (refs/replace) are superficially similar to obsolescences in that +they describe that one commit should be replaced by another. However, they +differ in both how they are created and how they are intended to be used. +Obsolescences are created automatically by the commands a user runs, and they +describe the user’s intent to perform a future rebase. Obsolete commits still +appear in branches, logs, etc like normal commits (possibly with an extra +decoration that marks them as obsolete). Replacements are typically created +explicitly by the user, they are meant to be kept around for a long time, and +they describe a replacement to be applied at read-time rather than as the input +to a future operation. When a replaced commit is queried, it is typically hidden +and swapped out with its replacement as though the replacement has already +occurred. + +Git-imerge is a project to help make complicated merges easier, particularly +when merging or rebasing long chains of patches. It is not an alternative to +the change graph, but its algorithm of applying smaller incremental merges +could be used as part of the evolve algorithm in the future. + +Overview +======== +We introduce the notion of “meta-commits” which describe how one commit was +created from other commits. A branch of meta-commits is known as a change. +Changes are created and updated automatically whenever a user runs a command +that creates a commit. They are used for locating obsolete commits, providing a +list of a user’s unsubmitted work in progress, and providing a stable name for +each unsubmitted change. + +Users can exchange edit histories by pushing and fetching changes. + +New commands will be introduced for manipulating changes and resolving +divergence between them. Existing commands that create commits will be updated +to modify the meta-commit graph and create changes where necessary. + +Example usage +------------- +# First create three dependent changes +$ echo foo>bar.txt && git add . +$ git commit -m "This is a test" +created change metas/this_is_a_test +$ echo foo2>bar2.txt && git add . +$ git commit -m "This is also a test" +created change metas/this_is_also_a_test +$ echo foo3>bar3.txt && git add . +$ git commit -m "More testing" +created change metas/more_testing + +# List all our changes in progress +$ git change list +metas/this_is_a_test +metas/this_is_also_a_test +* metas/more_testing +metas/some_change_already_merged_upstream + +# Now modify the earliest change, using its stable name +$ git reset --hard metas/this_is_a_test +$ echo morefoo>>bar.txt && git add . && git commit --amend --no-edit + +# Use git-evolve to fix up any dependent changes +$ git evolve +rebasing metas/this_is_also_a_test onto metas/this_is_a_test +rebasing metas/more_testing onto metas/this_is_also_a_test +Done + +# Use git-obslog to view the history of the this_is_a_test change +$ git log --obslog +93f110 metas/this_is_a_test@{0} commit (amend): This is a test +930219 metas/this_is_a_test@{1} commit: This is a test + +# Now create an unrelated change +$ git reset --hard origin/master +$ echo newchange>unrelated.txt && git add . +$ git commit -m "Unrelated change" +created change metas/unrelated_change + +# Fetch the latest code from origin/master and use git-evolve +# to rebase all dependent changes. +$ git fetch origin master +$ git evolve origin/master +deleting metas/some_change_already_merged_upstream +rebasing metas/this_is_a_test onto origin/master +rebasing metas/this_is_also_a_test onto metas/this_is_a_test +rebasing metas/more_testing onto metas/this_is_also_a_test +rebasing metas/unrelated_change onto origin/master +Conflict detected! Resolve it and then use git evolve --continue to resume. + +# Sort out the conflict +$ git mergetool +$ git evolve origin/master +Done + +# Share the full history of edits for the this_is_a_test change +# with a review server +$ git push origin metas/this_is_a_test:refs/for/master +# Share the lastest commit for “Unrelated change”, without history +$ git push origin HEAD:refs/for/master + +Detailed design +=============== +Obsolescence information is stored as a graph of meta-commits. A meta-commit is +a specially-formatted merge commit that describes how one commit was created +from others. + +Meta-commits look like this: + +$ git cat-file -p <example_meta_commit> +tree 4b825dc642cb6eb9a060e54bf8d69288fbee4904 +parent aa7ce55545bf2c14bef48db91af1a74e2347539a +parent d64309ee51d0af12723b6cb027fc9f195b15a5e9 +parent 7e1bbcd3a0fa854a7a9eac9bf1eea6465de98136 +author Stefan Xenos <sxenos@gmail.com> 1540841596 -0700 +committer Stefan Xenos <sxenos@gmail.com> 1540841596 -0700 +parent-type c r o + +This says “commit aa7ce555 makes commit d64309ee obsolete. It was created by +cherry-picking commit 7e1bbcd3”. + +The tree for meta-commits is always the empty tree, but future versions of git +may attach other trees here. For forward-compatibility fsck should ignore such +trees if found on future repository versions. This will allow future versions of +git to add metadata to the meta-commit tree without breaking forwards +compatibility. + +The commit comment for a meta-commit is an auto-generated user-readable string +describing the command that produced the meta commit. These strings are shown +to the user when they view the obslog. + +Parent-type +----------- +The “parent-type” field in the commit header identifies a commit as a +meta-commit and indicates the meaning for each of its parents. It is never +present for normal commits. It contains a space-deliminated list of enum values +whose order matches the order of the parents. Possible parent types are: + +- c: (content) the content parent identifies the commit that this meta-commit is + describing. +- r: (replaced) indicates that this parent is made obsolete by the content + parent. +- o: (origin) indicates that the content parent was generated by cherry-picking + this parent. +- a: (abandoned) used in place of a content parent for abandoned changes. Points + to the final content commit for the change at the time it was abandoned. + +There must be exactly one content or abandoned parent for each meta-commit and +it is always the first parent. The content commit will always be a normal commit +and not a meta-commit. However, future versions of git may create meta-commits +for other meta-commits and the fsck tool must be aware of this for forwards +compatibility. + +A meta-commit can have zero or more replaced parents. An amend operation creates +a single replaced parent. A merge used to resolve divergence (see divergence, +below) will create multiple replaced parents. A meta-commit may have no +replaced parents if it describes a cherry-pick or squash merge that copies one +or more commits but does not replace them. + +A meta-commit can have zero or more origin parents. A cherry-pick creates a +single origin parent. Certain types of squash merge will create multiple origin +parents. Origin parents don't directly cause their origin to become obsolete, +but are used when computing blame or locating a merge base. The section +on obsolescence over cherry-picks describes how the evolve command uses +origin parents. + +A replaced parent or origin parent may be either a normal commit (indicating +the oldest-known version of a change) or another meta-commit (for a change that +has already been modified one or more times). + +The parent-type field needs to go after the committer field since git's rules +for forwards-compatibility require that new fields to be at the end of the +header. Putting a new field in the middle of the header would break fsck. + +The presence of an abandoned parent indicates that the change should be pruned +by the evolve command, and removed from the repository's history. Any follow-up +changes should rebased onto the parent of the pruned commit. The abandoned +parent points to the version of the change that should be restored if the user +attempts to restore the change. + +Changes +------- +A branch of meta-commits describes how a commit was produced and what previous +commits it is based on. It is also an identifier for a thing the user is +currently working on. We refer to such a meta-branch as a change. + +Local changes are stored in the new refs/metas namespace. Remote changes are +stored in the refs/remote/<remotename>/metas namespace. + +The list of changes in refs/metas is more than just a mechanism for the evolve +command to locate obsolete commits. It is also a convenient list of all of a +user’s work in progress and their current state - a list of things they’re +likely to want to come back to. + +Strictly speaking, it is the presence of the branch in the refs/metas namespace +that marks a branch as being a change, not the fact that it points to a +metacommit. Metacommits are only created when a commit is amended or rebased, so +in the case where a change points to a commit that has never been modified, the +change points to that initial commit rather than a metacommit. + +Changes are also stored in the refs/hiddenmetas namespace. Hiddenmetas holds +metadata for historical changes that are not currently in progress by the user. +Commands like filter-branch and other bulk import commands create metadata in +this namespace. + +Note that the changes in hiddenmetas get special treatment in several ways: + +- They are not cleaned up automatically once merged, since it is expected that + they refer to historical changes. +- User commands that modify changes don't append to these changes as they would + to a change in refs/metas. +- They are not displayed when the user lists their local changes. + +Obsolescence +------------ +A commit is considered obsolete if it is reachable from the “replaces” edges +anywhere in the history of a change and it isn’t the head of that change. +Commits may be the content for 0 or more meta-commits. If the same commit +appears in multiple changes, it is not obsolete if it is the head of any of +those changes. + +Note that there is an exception to this rule. The metas namespace takes +precedence over the hiddenmetas namespace for the purpose of obsolescence. That +is, if a change appears in a replaces edge of a change in the metas namespace, +it is obsolete even if it also appears as the head of a change in the +hiddenmetas namespace. + +This special case prevents the hiddenmetas namespace from creating divergence +with the user's work in progress, and allows the user to resolve historical +divergence by creating new changes in the metas namespace. + +Divergence +---------- +From the user’s perspective, two changes are divergent if they both ask for +different replacements to the same commit. More precisely, a target commit is +considered divergent if there is more than one commit at the head of a change in +refs/metas that leads to the target commit via an unbroken chain of “replaces” +parents. + +Much like a merge conflict, divergence is a situation that requires user +intervention to resolve. The evolve command will stop when it encounters +divergence and prompt the user to resolve the problem. Users can solve the +problem in several ways: + +- Discard one of the changes (by deleting its change branch). +- Merge the two changes (producing a single change branch). +- Copy one of the changes (keep both commits, but one of them gets a new + metacommit appended to its history that is connected to its predecessor via an + origin edge rather than a replaces edge. That new change no longer obsoletes + the original.) + +Obsolescence across cherry-picks +-------------------------------- +By default the evolve command will treat cherry-picks and squash merges as being +completely separate from the original. Further amendments to the original commit +will have no effect on the cherry-picked copy. However, this behavior may not be +desirable in all circumstances. + +The evolve command may at some point support an option to look for cases where +the source of a cherry-pick or squash merge has itself been amended, and +automatically apply that same change to the cherry-picked copy. In such cases, +it would traverse origin edges rather than ignoring them, and would treat a +commit with origin edges as being obsolete if any of its origins were obsolete. + +Garbage collection +------------------ +For GC purposes, meta-commits are normal commits. Just as a commit causes its +parents and tree to be retained, a meta-commit also causes its parents to be +retained. + +Change creation +--------------- +Changes are created automatically whenever the user runs a command like “commit” +that has the semantics of creating a new change. They also move forward +automatically even if they’re not checked out. For example, whenever the user +runs a command like “commit --amend” that modifies a commit, all branches in +refs/metas that pointed to the old commit move forward to point to its +replacement instead. This also happens when the user is working from a detached +head. + +This does not mean that every commit has a corresponding change. By default, +changes only exist for recent locally-created commits. Users may explicitly pull +changes from other users or keep their changes around for a long time, but +either behavior requires a user to opt-in. Code review systems like gerrit may +also choose to keep changes around forever. + +Note that the changes in refs/metas serve a dual function as both a way to +identify obsolete changes and as a way for the user to keep track of their work +in progress. If we were only concerned with identifying obsolete changes, it +would be sufficient to create the change branch lazily the first time a commit +is obsoleted. Addressing the second use - of refs/metas as a mechanism for +keeping track of work in progress - is the reason for eagerly creating the +change on first commit. + +Change naming +------------- +When a change is first created, the only requirement for its name is that it +must be unique. Good names would also serve as useful mnemonics and be easy to +type. For example, a short word from the commit message containing no numbers or +special characters and that shows up with low frequency in other commit messages +would make a good choice. + +Different users may prefer different heuristics for their change names. For this +reason a new hook will be introduced to compute change names. Git will invoke +the hook for all newly-created changes and will append a numeric suffix if the +name isn’t unique. The default heuristics are not specified by this proposal and +may change during implementation. + +Change deletion +--------------- +Changes are normally only interesting to a user while a commit is still in +development and under review. Once the commit has submitted wherever it is +going, its change can be discarded. + +The normal way of deleting changes makes this easy to do - changes are deleted +by the evolve command when it detects that the change is present in an upstream +branch. It does this in two ways: if the latest commit in a change either shows +up in the branch history or the change becomes empty after a rebase, it is +considered merged and the change is discarded. In this context, an “upstream +branch” is any branch passed in as the upstream argument of the evolve command. + +In case this sometimes deletes a useful change, such automatic deletions are +recorded in the reflog allowing them to be easily recovered. + +Sharing changes +--------------- +Change histories are shared by pushing or fetching meta-commits and change +branches. This provides users with a lot of control of what to share and +repository implementations with control over what to retain. + +Users that only want to share the content of a commit can do so by pushing the +commit itself as they currently would. Users that want to share an edit history +for the commit can push its change, which would point to a meta-commit rather +than the commit itself if there is any history to share. Note that multiple +changes can refer to the same commits, so it’s possible to construct and push a +different history for the same commit in order to remove sensitive or irrelevant +intermediate states. + +Imagine the user is working on a change “mychange” that is currently the latest +commit on master. They have two ways to share it: + +# User shares just a commit without its history +> git push origin master + +# User shares the full history of the commit to a review system +> git push origin metas/mychange:refs/for/master + +# User fetches a collaborator’s modifications to their change +> git fetch remotename metas/mychange +# Which updates the ref remote/remotename/metas/mychange + +This will cause more intermediate states to be shared with the server than would +have been shared previously. A review system like gerrit would need to keep +track of which states had been explicitly pushed versus other intermediate +states in order to de-emphasize (or hide) the extra intermediate states from the +user interface. + +Merge-base +---------- +Merge-base will be changed to search the meta-commit graph for common ancestors +as well as the commit graph, and will generally prefer results from the +meta-commit graph over the commit graph. Merge-base will consider meta-commits +from all changes, and will traverse both origin and obsolete edges. + +The reason for this is that - when merging two versions of the same commit +together - an earlier version of that same commit will usually be much more +similar than their common parent. This should make the workflow of collaborating +on unsubmitted patches as convenient as the workflow for collaborating in a +topic branch by eliminating repeated merges. + +Configuration +------------- +The core.enableChanges configuration variable enables the creation and update +of change branches. This is enabled by default. + +User interface +-------------- +All git porcelain commands that create commits are classified as having one of +four behaviors: modify, create, copy, or import. These behaviors are discussed +in more detail below. + +Modify commands +--------------- +Modification commands (commit --amend, rebase) will mark the old commit as +obsolete by creating a new meta-commit that references the old one as a +replaced parent. In the event that multiple changes point to the same commit, +this is done independently for every such change. + +More specifically, modifications work like this: + +1. Locate all existing changes for which the old commit is the content for the + head of the change branch. If no such branch exists, create one that points + to the old commit. Changes that include this commit in their history but not + at their head are explicitly not included. +2. For every such change, create a new meta-commit that references the new + commit as its content and references the old head of the change as a + replaced parent. +3. Move the change branch forward to point to the new meta-commit. + +Copy commands +------------- +Copy commands (cherry-pick, merge --squash) create a new meta-commit that +references the old commits as origin parents. Besides the fact that the new +parents are tagged differently, copy commands work the same way as modify +commands. + +Create commands +--------------- +Creation commands (commit, merge) create a new commit and a new change that +points to that commit. The do not create any meta-commits. + +Import commands +--------------- +Import commands (fetch, pull) do not create any new meta-commits or changes +unless that is specifically what they are importing. For example, the fetch +command would update remote/origin/metas/change35 and fetch all referenced +meta-commits if asked to do so directly, but it wouldn’t create any changes or +meta-commits for commits discovered on the master branch when running “git fetch +origin master”. + +Other commands +-------------- +Some commands don’t fit cleanly into one of the above categories. + +Semantically, filter-branch should be treated as a modify command, but doing so +is likely to create a lot of irrelevant clutter in the changes namespace and the +large number of extra change refs may introduce performance problems. We +recommend treating filter-branch as an import command initially, but making it +behave more like a modify command in future follow-up work. One possible +solution may be to treat commits that are part of existing changes as being +modified but to avoid creating changes for other rewritten changes. Another +solution may be to record the modifications as changes in the hiddenmetas +namespace. + +Once the evolve command can handle obsolescence across cherry-picks, such +cherry-picks will result in a hybrid move-and-copy operation. It will create +cherry-picks that replace other cherry-picks, which will have both origin edges +(pointing to the new source commit being picked) and replacement edges (pointing +to the previous cherry-pick being replaced). + +Evolve +------ +The evolve command performs the correct sequence of rebases such that no change +has an obsolete parent. The syntax looks like this: + +git evolve [upstream…] + +It takes an optional list of upstream branches. All changes whose parent shows +up in the history of one of the upstream branches will be rebased onto the +upstream branch before resolving obsolete parents. + +Any change whose latest state is found in an upstream branch (or that ends up +empty after rebase) will be deleted. This is the normal mechanism for deleting +changes. Changes are created automatically on the first commit, and are deleted +automatically when evolve determines that they’ve been merged upstream. + +Orphan commits are commits with obsolete parents. The evolve command then +repeatedly rebases orphan commits with non-orphan parents until there are either +no orphan commits left, or a merge conflict is discovered. It will also +terminate if it detects a divergent parent or a cycle that can't be resolved +using any of the enabled transformations. + +When evolve discovers divergence, it will first check if it can resolve the +divergence automatically using one of its enabled transformations. Supported +transformations are: + +- Check if the user has already merged the divergent changes in a follow-up + change. That is, look for an existing merge in a follow-up change where all + the parents are divergent versions of the same change. Squash that merge with + its parents and use the result as the resolution for the divergence. + +- Attempt to auto-merge all the divergent changes (disabled by default). + +Each of the transformations can be enabled or disabled by command line options. + +Cycles can occur when two changes reference one another as parents. This can +happen when both changes use an obsolete version of the other change as their +parent. Although there are never cycles in the commit graph, users can create +cycles in the change graph by rebasing changes onto obsolete commits. The evolve +command has a transformation that will detect and break cycles by arbitrarily +picking one of the changes to go first. If this generates a merge conflict, +it tries each of the other changes in sequence to see if any ordering merges +cleanly. If no possible ordering merges cleanly, it picks one and terminates +to let the user resolve the merge conflict. + +If the working tree is dirty, evolve will attempt to stash the user's changes +before applying the evolve and then reapply those changes afterward, in much +the same way as rebase --autostash does. + +Checkout +-------- +Running checkout on a change by name has the same effect as checking out a +detached head pointing to the latest commit on that change-branch. There is no +need to ever have HEAD point to a change since changes always move forward when +necessary, no matter what branch the user has checked out + +Meta-commits themselves cannot be checked out by their hash. + +Reset +----- +Resetting a branch to a change by name is the same as resetting to the content +(or abandoned) commit at that change’s head. + +Commit +------ +Commit --amend gets modify semantics and will move existing changes forward. The +normal form of commit gets create semantics and will create a new change. + +$ touch foo && git add . && git commit -m "foo" && git tag A +$ touch bar && git add . && git commit -m "bar" && git tag B +$ touch baz && git add . && git commit -m "baz" && git tag C + +This produces the following commits: +A(tree=[foo]) +B(tree=[foo, bar], parent=A) +C(tree=[foo, bar, baz], parent=B) + +...along with three changes: +metas/foo = A +metas/bar = B +metas/baz = C + +Running commit --amend does the following: +$ git checkout B +$ touch zoom && git add . && git commit --amend -m "baz and zoom" +$ git tag D + +Commits: +A(tree=[foo]) +B(tree=[foo, bar], parent=A) +C(tree=[foo, bar, baz], parent=B) +D(tree=[foo, bar, zoom], parent=A) +Dmeta(content=D, obsolete=B) + +Changes: +metas/foo = A +metas/bar = Dmeta +metas/baz = C + +Merge +----- +Merge gets create, modify, or copy semantics based on what is being merged and +the options being used. + +The --squash version of merge gets copy semantics (it produces a new change that +is marked as a copy of all the original changes that were squashed into it). + +The “modify” version of merge replaces both of the original commits with the +resulting merge commit. This is one of the standard mechanisms for resolving +divergence. The parents of the merge commit are the parents of the two commits +being merged. The resulting commit will not be a merge commit if both of the +original commits had the same parent or if one was the parent of the other. + +The “create” version of merge creates a new change pointing to a merge commit +that has both original commits as parents. The result is what merge produces now +- a new merge commit. However, this version of merge doesn’t directly resolve +divergence. + +To select between these two behaviors, merge gets new “--amend” and “--noamend” +options which select between the “create” and “modify” behaviors respectively, +with noamend being the default. + +For example, imagine we created two divergent changes like this: + +$ touch foo && git add . && git commit -m "foo" && git tag A +$ touch bar && git add . && git commit -m "bar" && git tag B +$ touch baz && git add . && git commit --amend -m "bar and baz" +$ git tag C +$ git checkout B +$ touch bam && git add . && git commit --amend -m "bar and bam" +$ git tag D + +At this point the commit graph looks like this: + +A(tree=[foo]) +B(tree=[bar], parent=A) +C(tree=[bar, baz], parent=A) +D(tree=[bar, bam], parent=A) +Cmeta(content=C, obsoletes=B) +Dmeta(content=D, obsoletes=B) + +There would be three active changes with heads pointing as follows: + +metas/changeA=A +metas/changeB=Cmeta +metas/changeB2=Dmeta + +ChangeB and changeB2 are divergent at this point. Lets consider what happens if +perform each type of merge between changeB and changeB2. + +Merge example: Amend merge +One way to resolve divergent changes is to use an amend merge. Recall that HEAD +is currently pointing to D at this point. + +$ git merge --amend metas/changeB + +Here we’ve asked for an amend merge since we’re trying to resolve divergence +between two versions of the same change. There are no conflicts so we end up +with this: + +E(tree=[bar, baz, bam], parent=A) +Emeta(content=E, obsoletes=[Cmeta, Dmeta]) + +With the following branches: + +metas/changeA=A +metas/changeB=Emeta +metas/changeB2=Emeta + +Notice that the result of the “amend merge” is a replacement for C and D rather +than a new commit with C and D as parents (as a normal merge would have +produced). The parents of the amend merge are the parents of C and D which - in +this case - is just A, so the result is not a merge commit. Also notice that +changeB and changeB2 are now aliases for the same change. + +Merge example: Noamend merge +Consider what would have happened if we’d used a noamend merge instead. Recall +that HEAD was at D and our branches looked like this: + +metas/changeA=A +metas/changeB=Cmeta +metas/changeB2=Dmeta + +$ git merge --noamend metas/changeB + +That would produce the sort of merge we’d normally expect today: + +F(tree=[bar, baz, bam], parent=[C, D]) + +And our changes would look like this: +metas/changeA=A +metas/changeB=Cmeta +metas/changeB2=Dmeta +metas/changeF=F + +In this case, changeB and changeB2 are still divergent and we’ve created a new +change for our merge commit. However, this is just a temporary state. The next +time we run the “evolve” command, it will discover the divergence but also +discover the merge commit F that resolves it. Evolve will suggest converting F +into an amend merge in order to resolve the divergence and will display the +command for doing so. + +Rebase +------ +In general the rebase command is treated as a modify command. When a change is +rebased, the new commit replaces the original. + +Rebase --abort is special. Its intent is to restore git to the state it had +prior to running rebase. It should move back any changes to point to the refs +they had prior to running rebase and delete any new changes that were created as +part of the rebase. To achieve this, rebase will save the state of all changes +in refs/metas prior to running rebase and will restore the entire namespace +after rebase completes (deleting any newly-created changes). Newly-created +metacommits are left in place, but will have no effect until garbage collected +since metacommits are only used if they are reachable from refs/metas. + +Change +------ +The “change” command can be used to list, rename, reset or delete change. It has +a number of subcommands. + +The "list" subcommand lists local changes. If given the -r argument, it lists +remote changes. + +The "rename" subcommand renames a change, given its old and new name. If the old +name is omitted and there is exactly one change pointing to the current HEAD, +that change is renamed. If there are no changes pointing to the current HEAD, +one is created with the given name. + +The "forget" subcommand deletes a change by deleting its ref from the metas/ +namespace. This is the normal way to delete extra aliases for a change if the +change has more than one name. By default, this will refuse to delete the last +alias for a change if there are any other changes that reference this change as +a parent. + +The "update" subcommand adds a new state to a change. It uses the default +algorithm for assigning change names. If the content commit is omitted, HEAD is +used. If given the optional --force argument, it will overwrite any existing +change of the same name. This latter form of "update" can be used to effectively +reset changes. + +The "update" command can accept any number of --origin and --replace arguments. +If any are present, the resulting change branch will point to a metacommit +containing the given origin and replacement edges. + +The "abandon" command deletes a change using obsolescence markers. It marks the +change as being obsolete and having been replaced by its parent. If given no +arguments, it applies to the current commit. Running evolve will cause any +abandoned changes to be removed from the branch. Any child changes will be +reparented on top of the parent of the abandoned change. If the current change +is abandoned, HEAD will move to point to its parent. + +The "restore" command restores a previously-abandoned change. + +The "prune" command deletes all obsolete changes and all changes that are +present in the given branch. Note that such changes can be recovered from the +reflog. + +Combined with the GC protection that is offered, this is intended to facilitate +a workflow that relies on changes instead of branches. Users could choose to +work with no local branches and use changes instead - both for mailing list and +gerrit workflows. + +Log +--- +When a commit is shown in git log that is part of a change, it is decorated with +extra change information. If it is the head of a change, the name of the change +is shown next to the list of branches. If it is obsolete, it is decorated with +the text “obsolete, <n> commits behind <changename>”. + +Log gets a new --obslog argument indicating that the obsolescence graph should +be followed instead of the commit graph. This also changes the default +formatting options to make them more appropriate for viewing different +iterations of the same commit. + +Pull +---- + +Pull gets an --evolve argument that will automatically attempt to run "evolve" +on any affected branches after pulling. + +We also introduce an "evolve" enum value for the branch.<name>.rebase config +value. When set, the evolve behavior will happen automatically for that branch +after every pull even if the --evolve argument is not used. + +Next +---- + +The "next" command will reset HEAD to a non-obsolete commit that refers to this +change as its parent. If there is more than one such change, the user will be +prompted. If given the --evolve argument, the next commit will be evolved if +necessary first. + +The "next" command can be thought of as the opposite of +"git reset --hard HEAD^" in that it navigates to a child commit rather than a +parent. + +Prev +---- + +The "prev" command will reset HEAD to the latest version of the parent change. +If the parent change isn't obsolete, this is equivalent to +"git reset --hard HEAD^". If the parent commit is obsolete, it resets to the +latest replacement for the parent commit. + +Other options considered +======================== +We considered several other options for storing the obsolescence graph. This +section describes the other options and why they were rejected. + +Commit header +------------- +Add an “obsoletes” field to the commit header that points backwards from a +commit to the previous commits it obsoletes. + +Pros: +- Very simple +- Easy to traverse from a commit to the previous commits it obsoletes. +Cons: +- Adds a cost to the storage format, even for commits where the change history + is uninteresting. +- Unconditionally prevents the change history from being garbage collected. +- Always causes the change history to be shared when pushing or pulling changes. + +Git notes +--------- +Instead of storing obsolescence information in metacommits, the metacommit +content could go in a new notes namespace - say refs/notes/metacommit. Each note +would contain the list of obsolete and origin parents. An automerger could +be supplied to make it easy to merge the metacommit notes from different remotes. + +Pros: +- Easy to locate all commits obsoleted by a given commit (since there would only + be one metacommit for any given commit). +Cons: +- Wrong GC behavior (obsolete commits wouldn’t automatically be retained by GC) + unless we introduced a special case for these kinds of notes. +- No way to selectively share or pull the metacommits for one specific change. + It would be all-or-nothing, which would be expensive. This could be addressed + by changes to the protocol, but this would be invasive. +- Requires custom auto-merging behavior on fetch. + +Tags +---- +Put the content of the metacommit in a message attached to tag on the +replacement commit. This is very similar to the git notes approach and has the +same pros and cons. + +Simple forward references +------------------------- +Record an edge from an obsolete commit to its replacement in this form: + +refs/obsoletes/<A> + +pointing to commit <B> as an indication that B is the replacement for the +obsolete commit A. + +Pros: +- Protects <B> from being garbage collected. +- Fast lookup for the evolve operation, without additional search structures + (“what is the replacement for <A>?” is very fast). + +Cons: +- Can’t represent divergence (which is a P0 requirement). +- Creates lots of refs (which can be inefficient) +- Doesn’t provide a way to fetch only refs for a specific change. +- The obslog command requires a search of all refs. + +Complex forward references +-------------------------- +Record an edge from an obsolete commit to its replacement in this form: + +refs/obsoletes/<change_id>/obs<A>_<B> + +Pointing to commit <B> as an indication that B is the replacement for obsolete +commit A. + +Pros: +- Permits sharing and fetching refs for only a specific change. +- Supports divergence +- Protects <B> from being garbage collected. + +Cons: +- Creates lots of refs, which is inefficient. +- Doesn’t provide a good lookup structure for lookups in either direction. + +Backward references +------------------- +Record an edge from a replacement commit to the obsolete one in this form: + +refs/obsolescences/<B> + +Cons: +- Doesn’t provide a way to resolve divergence (which is a P0 requirement). +- Doesn’t protect <B> from being garbage collected (which could be fixed by + combining this with a refs/metas namespace, as in the metacommit variant). + +Obsolescences file +------------------ +Create a custom file (or files) in .git recording obsolescences. + +Pros: +- Can store exactly the information we want with exactly the performance we want + for all operations. For example, there could be a disk-based hashtable + permitting constant time lookups in either direction. + +Cons: +- Handling GC, pushing, and pulling would all require custom solutions. GC + issues could be addressed with a repository format extension. + +Squash points +------------- +We treat changes like topic branches, and use special squash points to mark +places in the commit graph that separate changes. + +We create and update change branches in refs/metas at the same time we +would have in the metacommit proposal. However, rather than pointing to a +metacommit branch they point to normal commits and are treated as “squash +points” - markers for sequences of commits intended to be squashed together on +submission. + +Amends and rebases work differently than they do now. Rather than actually +containing the desired state of a commit, they contain a delta from the previous +version along with a squash point indicating that the preceding changes are +intended to be squashed on submission. Specifically, amends would become new +changes and rebases would become merge commits with the old commit and new +parent as parents. + +When the changes are finally submitted, the squashes are executed, producing the +final version of the commit. + +In addition to the squash points, git would maintain a set of “nosquash” tags +for commits that were used as ancestors of a change that are not meant to be +included in the squash. + +For example, if we have this commit graph: + +A(...) +B(parent=A) +C(parent=B) + +...and we amend B to produce D, we’d get: + +A(...) +B(parent=A) +C(parent=B) +D(parent=B) + +...along with a new change branch indicating D should be squashed with its +parents when submitted: + +metas/changeB = D +metas/changeC = C + +We’d also create a nosquash tag for A indicating that A shouldn’t be included +when changeB is squashed. + +If a user amends the change again, they’d get: + +A(...) +B(parent=A) +C(parent=B) +D(parent=B) +E(parent=D) + +metas/changeB = E +metas/changeC = C + +Pros: +- Good GC behavior. +- Provides a natural way to share changes (they’re just normal branches). +- Merge-base works automatically without special cases. +- Rewriting the obslog would be easy using existing git commands. +- No new data types needed. +Cons: +- No way to connect the squashed version of a change to the original, so no way + to automatically clean up old changes. This also means users lose all benefits + of the evolve command if they prematurely squash their commits. This may occur + if a user thinks a change is ready for submission, squashes it, and then later + discovers an additional change to make. +- Histories would look very cluttered (users would see all previous edits to + their commit in the commit log, and all previous rebases would show up as + merges). Could be quite hard for users to tell what is going on. (Possible + fix: also implement a new smart log feature that displays the log as though + the squashes had occurred). +- Need to change the current behavior of current commands (like amend and + rebase) in ways that will be unexpected to many users.