patcht7701: make annotated tag unreachable

In 4dc16e2cb0 (gc: introduce `gc.recentObjectsHook`, 2023-06-07), we added tests to ensure that prune-able (i.e. unreachable and with mtime older than the cutoff) objects which are marked as recent via the new `gc.recentObjectsHook` configuration are unpacked as loose with `--unpack-unreachable`.

In that test, we also ensure that objects which are reachable from other unreachable objects which were *not* pruned are kept as well, regardless of their mtimes. For this, we use an annotated tag pointing at a blob ($obj2) which would otherwise be pruned.

But after pruning, that object is kept around for two reasons. One, the tag object's mtime wasn't adjusted to be beyond the 1-hour cutoff, so it would be kept as due to its recency regardless. The other reason is because the tag itself is reachable.

Use mktag to write the tag object directly without pointing a reference at it, and adjust the mtime of the tag object to be older than the cutoff to ensure that our `gc.recentObjectsHook` configuration is working as intended.

Noticed-by: Jeff King <peff@peff.net>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
Peff pointed this out to me after the gc.recentObjectsHook stuff was in
'next', but this should go on top. This patch is based off of the tip of
"tb/gc-recent-object-hook".

 t/t7701-repack-unpack-unreachable.sh | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)

On Wed, Jun 21, 2023 at 06:21:10AM -0400, Taylor Blau wrote:

Thanks, this should fix the problem (which you can notice by making the recentObjectsHook in the test a noop and seeing that it still passes).

But there's one more thing...

Since we are using "mktag" here, the call to "git tag" can go away entirely. It's not only redundant, but it erroneously keeps reachability to $obj2. The test still does something useful (it makes sure that $obj2_tag was saved), but it is failing to test what it wanted to about $obj2, namely that it is kept because the hook mentioned an object that points to it.

So we'd want to squash in:

In 4dc16e2cb0 (gc: introduce `gc.recentObjectsHook`, 2023-06-07), we added tests to ensure that prune-able (i.e. unreachable and with mtime older than the cutoff) objects which are marked as recent via the new `gc.recentObjectsHook` configuration are unpacked as loose with `--unpack-unreachable`.

In that test, we also ensure that objects which are reachable from other unreachable objects which were *not* pruned are kept as well, regardless of their mtimes. For this, we use an annotated tag pointing at a blob ($obj2) which would otherwise be pruned.

But after pruning, that object is kept around for two reasons. One, the tag object's mtime wasn't adjusted to be beyond the 1-hour cutoff, so it would be kept as due to its recency regardless. The other reason is because the tag itself is reachable.

Use mktag to write the tag object directly without pointing a reference at it, and adjust the mtime of the tag object to be older than the cutoff to ensure that our `gc.recentObjectsHook` configuration is working as intended.

Noticed-by: Jeff King <peff@peff.net>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
---
Fixes a trivial oversight from an earlier version of this patch noticed
by Peff.

Range-diff against v1:
1:  48d3c2c187 ! 1:  259b1b5591 t7701: make annotated tag unreachable
    @@ t/t7701-repack-unpack-unreachable.sh: test_expect_success 'gc.recentObjectsHook'
      	git cat-file -p $obj2 &&
      	git cat-file -p $obj3 &&

    +-	git tag -a -m tag obj2-tag $obj2 &&
    +-	obj2_tag="$(git rev-parse obj2-tag)" &&
     +	# make an unreachable annotated tag object to ensure we rescue objects
     +	# which are reachable from non-pruned unreachable objects
    - 	git tag -a -m tag obj2-tag $obj2 &&
    --	obj2_tag="$(git rev-parse obj2-tag)" &&
     +	obj2_tag="$(git mktag <<-EOF
     +	object $obj2
     +	type blob

 t/t7701-repack-unpack-unreachable.sh | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)

On Sat, Jun 24, 2023 at 10:33:47AM -0400, Taylor Blau wrote:

Thanks, this looks great to me.

-Peff