Tests failing in Fedora with git 2.48 (git update-ref generating an invalid reflog) #138
Comments
|
Thanks for the catch! It appears that git 2.28 introduced a bug (I believe) in the From that email:
This ends up breaking the The test could be changed such that it no longer depends on the reflog and passes on git 2.48, however the underlying issue with |
mystor
changed the title
|
Wow, thanks for tracking that down! I guess for now, there might not be a need to actually patch anything here, and hopefully they address the regression in the next release. If you think these failures don't represent any potential problems with the git-revise functionality, I can just disable those tests locally in the Fedora build until the git version is one that contains a fix. |
In git 2.48, there is an issue where using `git update-ref` add a new entry to the reflog for a branch will cause corrupted reflog entries to be created for any symbolic refs referencing the updated reference. This impacts most usage of `git-revise`, as we update the default `HEAD` reference indirectly due to using `git rev-parse --symbolic-full-name` to resolve the identity of a branch. This patch works around this issue in the common case by checking if the given name directly names a valid ref or symbolic ref, skipping the `--symbolic-full-name` resolution if it does. This means that HEAD will be directly updated by `git update-ref`, and receive valid reflog entries. This workaround shouldn't negatively impact other git versions, and is largely an internal change to how symbolic refs are handled by the odb.Reference type. See #138 for details.
In git 2.48, there is an issue where using `git update-ref` add a new entry to the reflog for a branch will cause corrupted reflog entries to be created for any symbolic refs referencing the updated reference. This impacts most usage of `git-revise`, as we update the default `HEAD` reference indirectly due to using `git rev-parse --symbolic-full-name` to resolve the identity of a branch. This patch works around this issue in the common case by checking if the given name directly names a valid ref or symbolic ref, skipping the `--symbolic-full-name` resolution if it does. This means that HEAD will be directly updated by `git update-ref`, and receive valid reflog entries. This workaround shouldn't negatively impact other git versions, and is largely an internal change to how symbolic refs are handled by the odb.Reference type. See #138 for details.
|
Unfortunately this bug does have some impact on I've posted a small patch with a work-around for the git bug (#139) to restore that functionality and fix the tests on 2.48.1. Fortunately, the git bug appears to already have a patch being reviewed, so hopefully this will be fixed in 2.48.2 |
In git 2.48, there is an issue where using `git update-ref` add a new entry to the reflog for a branch will cause corrupted reflog entries to be created for any symbolic refs referencing the updated reference. This impacts most usage of `git-revise`, as we update the default `HEAD` reference indirectly due to using `git rev-parse --symbolic-full-name` to resolve the identity of a branch. This patch works around this issue in the common case by checking if the given name directly names a valid ref or symbolic ref, skipping the `--symbolic-full-name` resolution if it does. This means that HEAD will be directly updated by `git update-ref`, and receive valid reflog entries. This workaround shouldn't negatively impact other git versions, and is largely an internal change to how symbolic refs are handled by the odb.Reference type. See #138 for details.
It looks like git was updated recently in Fedora 41 from 2.47 to 2.48, and I believe that is now causing the git-revise tests to fail. This can be seen in a Fedora scratch build (https://koschei.fedoraproject.org/build/19488570), with the build log at https://kojipkgs.fedoraproject.org/work/tasks/6173/128186173/build.log.
The output from the test sections are:
The text was updated successfully, but these errors were encountered: