bug: `git pull` does not clean up after itself if it fails

What did you do before the bug happened? (Steps to reproduce your issue) In your local repo have a directory that you cannot write to. Pull from a remote that has changes/new files in that directory.

What did you expect to happen? (Expected behavior) The pull fails and the local repo in the same state as before.

What happened instead? (Actual behavior) All changes to writeable files are applied and all new files in writeable dirs are added. The local HEAD however remained the same.

What's different between what you expected and what actually happened? A failed `git pull` inadvertently creates changes and untracked files.

Anything else you want to add: I created this bugreport on my local machine, but I first noticed this on git version 2.25.1 under Ubuntu (uname: Linux s-c05-mvi-civdb 5.4.0-174-generic #193-Ubuntu SMP Thu Mar 7 14:29:28 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux)

[System Info] git version: git version 2.45.2 cpu: x86_64 no commit associated with this build sizeof-long: 8 sizeof-size_t: 8 shell-path: /bin/sh uname: Linux 6.9.7-arch1-1 #1 SMP PREEMPT_DYNAMIC Fri, 28 Jun 2024 04:32:50 +0000 x86_64 compiler info: gnuc: 14.1 libc info: glibc: 2.39 $SHELL (typically, interactive shell): /bin/zsh

On Mon, Jul 15, 2024 at 01:19:32PM +0200, Nikolai Zaki wrote:

A pull is basically fetch followed by merge. So I'd expect us to fail in the merge step, at which point pull would generally leave the state as-is, because merge failures tend to be conflicts.

So I guess doing what you expect would require a merge which hits a non-conflict error (like an unwriteable directory) to clean up after itself. That's probably not impossible, but it may be rather tricky, especially if we are concerned about unrelated changes in the working tree.

-Peff

Hi Peff,

thanks for your reply. You're right, the problem happens during merge.

Since merge already aborts if the index doesn't match HEAD though, couldn't it be as simple as saving the state of the worktree and then either restoring it in case of failure or applying the diff in case of success?

In terms of git commands I think the following would work:

When `git merge` is run first run `git stash --all` then `git merge` as usual.

If there are non-conflict errors: `git checkout .` and `git clean -f;

If there are no errors `git stash pop`.

On 7/17/24 07:55, Jeff King wrote: