git diff-files bug

Thank you for filling out a Git bug report! Please answer the following questions to help us understand your issue.

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

⬢[jreed@toolbox example-project]$ git diff-files --exit-code --patch; echo $? 0 ⬢[jreed@toolbox example-project]$ chmod g+w README.md ⬢[jreed@toolbox example-project]$ git diff-files --exit-code --patch; echo $? 1 ⬢[jreed@toolbox example-project]$ git diff --exit-code --patch; echo $? 0 ⬢[jreed@toolbox example-project]$ git diff-files --exit-code --patch; echo $? 0

What did you expect to happen? (Expected behavior)

Git diff-files should likely ignore group permissions changes, or at least its output should be stable across the same worktree/index state.

What happened instead? (Actual behavior)

The command `git diff-files --exit-code --patch` fails with no exit code when a file mode has changed in a way that the rest of git commands ignored. Furthermore, subsequent git commands seem to change the behavior of `git diff-files` in this regard, so it's hard to tell what is the expected behavior.

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

The `git diff-files --exit-code` command is inconsistent in how it behaves. I suspect it should ignore irrelavant mode changes like `g+w`, but even if it should report them, they should produce a patch or at least have stable results when we re-run the command.

Anything else you want to add:

Please review the rest of the bug report below. You can delete any lines you don't wish to share.

[System Info] git version: git version 2.41.0 cpu: x86_64 no commit associated with this build sizeof-long: 8 sizeof-size_t: 8 shell-path: /bin/sh uname: Linux 6.5.10-200.fc38.x86_64 #1 SMP PREEMPT_DYNAMIC Thu Nov 2 19:59:55 UTC 2023 x86_64 compiler info: gnuc: 13.1 libc info: glibc: 2.37 $SHELL (typically, interactive shell): /usr/bin/fish

[Enabled Hooks] pre-commit pre-push

Hi,

On Tue, Dec 19, 2023 at 8:56 AM Josh Reed <jriddy@gmail.com> wrote:

Let's extend your example a bit to point out when/if the index is updated...

$ stat --format=%y .git/index 2023-12-19 09:30:55.867055501 -0800

This is the modification time of the index at the very beginning.

$ stat --format=%y .git/index 2023-12-19 09:30:55.867055501 -0800

As expected, the index has not changed just because some file changed.

$ stat --format=%y .git/index 2023-12-19 09:30:55.867055501 -0800

Note that the index still has not changed; diff-files will not update it when it notices differences. But diff-files shows us differences between the working tree and index, and the working tree has changed. Those changes mean that the stat information in the index no longer matches the working tree. That alone is enough for diff-files to give a non-zero exit status. This does tend to trip up folks, though.

$ stat --format=%y .git/index 2023-12-19 09:33:46.773896615 -0800

Now the index _was_ updated. So, not only are there no content differences, and no relevant filemode differences, but the stat information in the index now matches what is in the working tree. So, diff returns an exit code of 0. (Note that this means "git diff" is not a read-only operation, which sometimes trips up people in the other direction.)

Right, contents match, relelvant filemodes match, and the stat information in the index now matches what is in the working tree. And:

$ stat --format=%y .git/index 2023-12-19 09:33:46.773896615 -0800

...no further changes to the index from this command, of course.

It does ignore group permission changes; the problem was that stat information in the index did not match what was in the working tree. If you run

$ git update-index --refresh

before calling diff-files, that'll make sure the stat information is up-to-date.

This has bit a few folks before; I wonder if there's a reasonable update we can make to the documentation to address this.