On Wed, Oct 18, 2023 at 03:27:11PM +0200, Han-Wen Nienhuys wrote:

On Wed, Oct 18, 2023 at 7:35 AM Patrick Steinhardt [off-list ref] wrote:

quoted

In t4207 we set up a set of replace objects via git-replace(1). Because
these references should not be impacting subsequent tests we also set up
some cleanup logic that deletes the replacement references via a call to
`rm -rf`. This reaches into the internal implementation details of the
reference backend and will thus break when we grow an alternative refdb
implementation.

Refactor the tests to delete the replacement refs via Git commands so
that we become independent of the actual refdb that's in use. As we
don't have a nice way to delete all replacements or all references in a
certain namespace, we opt for a combination of git-for-each-ref(1) and
git-update-ref(1)'s `--stdin` mode.

There is a test helper that can directly access the ref database,
t/helper/test-ref-store.c.

If you use that manipulate refs for testing purposes, you make the
test independent of behavior git-for-each-ref/git-update-ref, which is
what you want for testing replace-objects?

Is there any specific reason why we shouldn't be using
git-for-each-ref(1) or git-update-ref(1) here? Neither of those commands
are part of the system under test, as we rather care about git-log(1)
here. So as those commands are already being verified in other tests I
think it should be fine to assume that they work as intended here.

Happy to hear differing viewpoints though.

Patrick