includeIf not matching during `git rebase`

includeIf not matching during `git rebase`

[Michał Mirosław]

* What did you do before the bug happened? (Steps to reproduce your issue)

With ~/.gitconfig having:

[includeIf "onbranch:pr/"]
        path = .gitconfig.for-upstream

where the included config changes commit-msg hook,

git checkout pr/zzz # has multiple commits over upstream
git rebase -i
> 'edit' first commit
(modify it)
git rebase --continue

* What did you expect to happen? (Expected behavior)

The commit messages are unchanged.

* What happened instead? (Actual behavior)

The rebased commits had messages modified by default commit-msg hook.

* What's different between what you expected and what actually happened?

I'd expect includeIf to match and in effect prevent the default commit
message modifications.

[System Info]
git version:
git version 2.41.0.487.g6d72f3e995-goog
cpu: x86_64
no commit associated with this build
sizeof-long: 8
sizeof-size_t: 8
shell-path: /bin/sh
compiler info: gnuc: 12.2
libc info: glibc: 2.36
$SHELL (typically, interactive shell): /bin/bash


[Enabled Hooks]
commit-msg
pre-commit
prepare-commit-msg

Re: includeIf not matching during `git rebase`

[Emily Shaffer]

On Tue, Jul 25, 2023 at 12:49 PM Michał Mirosław <emmir@google.com> wrote:
>
> * What did you do before the bug happened? (Steps to reproduce your issue)
>
> With ~/.gitconfig having:
>
> [includeIf "onbranch:pr/"]
>         path = .gitconfig.for-upstream
>
> where the included config changes commit-msg hook,
>
> git checkout pr/zzz # has multiple commits over upstream
> git rebase -i
> > 'edit' first commit
> (modify it)
> git rebase --continue

Hm, I would guess this is why - in the middle of the rebase, the
branch ref doesn't move, it's only moved to the new tip when the
rebase is completed. However, it seems that we do have some knowledge
of which branch we are trying to rebase:

emilyshaffer@podkayne:~/git [libification-style|REBASE 3/4]$ git
branch | head -n1
* (no branch, rebasing libification-style)

It looks like to log that error, we're using a cached branch name (and
pick that up from wt_status_get_state and eventually
wt_status_check_rebase, which checks
.git/rebase-(merge|apply)/head-name).

However, when we check the current branch for includeif.onbranch
(config.c:include_by_branch()) we're using resolve_ref_unsafe("HEAD"),
which doesn't check the current rebase state the way that the
wt_status_* stuff does.

Does that mean that the config machinery should also be using
wt_status to determine which branch to use? The use case Michał is
describing sounds perfectly reasonable to me - is there reason to
think that doing conditional includes during a rebase based on "the
branch we were in the middle of rebasing" would negatively impact
someone's existing workflow?

 - Emily

Re: includeIf not matching during `git rebase`

[Junio C Hamano]

Emily Shaffer <nasamuffin@google.com> writes:

> emilyshaffer@podkayne:~/git [libification-style|REBASE 3/4]$ git
> branch | head -n1
> * (no branch, rebasing libification-style)
>
> It looks like to log that error, we're using a cached branch name (and
> pick that up from wt_status_get_state and eventually
> wt_status_check_rebase, which checks
> .git/rebase-(merge|apply)/head-name).
>
> However, when we check the current branch for includeif.onbranch
> (config.c:include_by_branch()) we're using resolve_ref_unsafe("HEAD"),
> which doesn't check the current rebase state the way that the
> wt_status_* stuff does.
>
> Does that mean that the config machinery should also be using
> wt_status to determine which branch to use?

Not really.  The low-level config machinery shouldn't rely on a
piece of information from so high a layer (making call to
wt_status.c or spawning "git status" is an absolute no-no).

But "we are not exactly on branch X, but doing work on behalf of
branch X" is a common situation during rebase and possibly bisect,
and I agree that it is a good future direction to introduce a
reliable low-level primitive to notice that condition.

I however am hesitant to fully support such an idea, because I
suspect that there may be cases such as "we are technically on
branch Y, but actually doing work on behalf of branch X" or worse
yet "we are on branch Z, but actually doing work on behalf of both
branches X and Y", where there are more than one plausible branch,
which is different from what HEAD points at, that
include_by_branch() could use.

Thanks.

Re: includeIf not matching during `git rebase`

[Glen Choo]

Junio C Hamano <gitster@pobox.com> writes:

> Emily Shaffer <nasamuffin@google.com> writes:
>
>> Does that mean that the config machinery should also be using
>> wt_status to determine which branch to use?
>
> Not really.  The low-level config machinery shouldn't rely on a
> piece of information from so high a layer (making call to
> wt_status.c or spawning "git status" is an absolute no-no).

"includes" are surprisingly high-level and tacked-on compared to the
rest of config parsing. Includes are implemented using just config
callbacks; config_with_options() does the preparation and
git_config_include() evaluates the includes. And, "includeIf"s already
use all sorts of higher-level information (onbranch: is evaluated by
resolving HEAD, gitdir knows about the repo setup). So I don't think it
would be the worst thing to introduce such a check into config...

> But "we are not exactly on branch X, but doing work on behalf of
> branch X" is a common situation during rebase and possibly bisect,
> and I agree that it is a good future direction to introduce a
> reliable low-level primitive to notice that condition.
>
> I however am hesitant to fully support such an idea, because I
> suspect that there may be cases such as "we are technically on
> branch Y, but actually doing work on behalf of branch X" or worse
> yet "we are on branch Z, but actually doing work on behalf of both
> branches X and Y", where there are more than one plausible branch,
> which is different from what HEAD points at, that
> include_by_branch() could use.

But I fully agree that this shouldn't be a one-off config-only change,
and I don't think we fully understand the ramifications of how this
change will affect _other_ parts of Git. I don't think we'll figure that
out without testing, though. Perhaps we could put this behind a feature
flag, probably even feature.experimental.