Re: [PATCH v2 1/4] t1400: Avoid touching refs on filesystem

From: Patrick Steinhardt <ps@pks.im>
Date: Nov 10, 2020, 12:34 UTC
Message-ID: <X6qIidgbnhYa3S0o@ncase>
In-Reply-To: <xmqqpn4mb9bv.fsf@gitster.c.googlers.com>

On Mon, Nov 09, 2020 at 11:48:20AM -0800, Junio C Hamano wrote:

Show 19 quoted lines

> Patrick Steinhardt <ps@pks.im> writes:
> > diff --git a/t/t1400-update-ref.sh b/t/t1400-update-ref.sh
> > index 4c01e08551..957bef272d 100755
> > --- a/t/t1400-update-ref.sh
> > +++ b/t/t1400-update-ref.sh
> > @@ -14,6 +14,12 @@ n=$n_dir/fixes
> >  outside=refs/foo
> >  bare=bare-repo
> >  
> > +# Some of the tests delete HEAD, which causes us to not treat the current
> > +# working directory as a Git repository anymore. To avoid using any potential
> > +# parent repository to be discovered, we need to set up the ceiling directories.
> > +GIT_CEILING_DIRECTORIES="$PWD/.."
> > +export GIT_CEILING_DIRECTORIES
> > +
> 
> Interesting.  The current tests do not need to do this because the
> HEAD-less broken state is transitory and is corrected without using
> "git" at all (e.g. echoing a valid value into .git/HEAD), I presume?

Correct.

Show 22 quoted lines

> > @@ -80,26 +86,26 @@ test_expect_success "fail to delete $m (by HEAD) with stale ref" '
> >  	test $B = $(git show-ref -s --verify $m)
> >  '
> >  test_expect_success "delete $m (by HEAD)" '
> > -	test_when_finished "rm -f .git/$m" &&
> > +	test_when_finished "git update-ref -d $m" &&
> >  	git update-ref -d HEAD $B &&
> > -	test_path_is_missing .git/$m
> > +	test_must_fail git show-ref --verify -q $m
> >  '
> 
> During the above test, we are on the branch ${m#refs/heads/}, so
> "update-ref -d HEAD" is removing the underlying branch ref, making
> it an unborn branch, without destroying the repository, so this is
> perfectly sensible.
> 
> This is a tangent, but what makes this test doubly interesting is
> that "git update-ref -d HEAD" would have allowed us to make it a
> non-repository if HEAD were detached, and it seems that we do not
> require "--force" to do so.  We probably should forbid removing HEAD
> that id detached without "--force", which is such a destructive
> operation.

That'd make sense to me, yes. It also took me quite some time to actually figure out why tests were misbehaving when I converted it. Until I finally realized: this is not a Git repo anymore, and refs have now been modified directly in the real git.git repository. Just this morning I still found some invalid refs created by the test in git.git.

Show 12 quoted lines

> >  cp -f .git/HEAD .git/HEAD.orig
> >  test_expect_success 'delete symref without dereference' '
> >  	test_when_finished "cp -f .git/HEAD.orig .git/HEAD" &&
> >  	git update-ref --no-deref -d HEAD &&
> > -	test_path_is_missing .git/HEAD
> > +	test_must_fail git show-ref --verify -q HEAD
> >  '
> 
> This is an example of breaking the repository.  I am not sure if the
> test_must_fail is a good replacement--it would fail even if you say
> "git show-ref --verify -q refs/heads/$branch" where $branch is a
> name of a branch that exists, no?

Right, it's not. We're not detecting HEAD deletion by means of checking that it doesn't exist anymore, but only detect it because the repo is not a repo anymore. Which would in fact cause most git commands to fail now.

Show 16 quoted lines

> For now, I think this is fine, but we'd probably clean it up so that
> without --force update-ref would not corrupt the repository like
> this.  When used with --force, or before adding such a safety
> measure, how we test if we successfully corrupted the repository is
> an interesting matter.  I'd say
> 
> 	git update-ref --force --no-deref -d HEAD &&
> 	test_must_fail git show-ref --verify -q HEAD &&
> 	cp -f .git/HEAD.orig .git/HEAD &&
> 	git show-ref --verify -q "$m"
> 
> to ensure that other than removing HEAD and corrupting the
> repository, it did not cause permanent damage by removing the
> underlying ref, perhaps.
> 
> We may want to add some NEEDSWORK comment around such tests.

I think the proper way to do the test would be to create a non-mandatory symref and delete it. That'd also be a nice preparation for the `--force` parameter already.

Show 10 quoted lines

> >  test_expect_success 'delete symref without dereference when the referred ref is packed' '
> > @@ -208,7 +214,7 @@ test_expect_success 'delete symref without dereference when the referred ref is
> >  	git commit -m foo &&
> >  	git pack-refs --all &&
> >  	git update-ref --no-deref -d HEAD &&
> > -	test_path_is_missing .git/HEAD
> > +	test_must_fail git show-ref --verify -q HEAD
> >  '
> 
> Does this share the same issue as the above?

Yup, it does.

Patrick