Re: [PATCH v3] reflog: close leak of reflog expire entry
- From
Jacob Keller <jacob.e.keller@intel.com>
- Date
- Jul 22, 2025, 23:10 UTC
- Message-ID
- <fd14c857-63a8-41e7-8361-bc816d4a47c4@intel.com>
- In-Reply-To
- <20250722045456.GA824456@coredump.intra.peff.net>
On 7/21/2025 9:54 PM, Jeff King wrote:
Show 24 quoted lines
> On Mon, Jul 21, 2025 at 04:39:37PM -0700, Jacob Keller wrote: > >> Changes in v3: >> - Remove the incorrect call in reflog_expiry_cleanup() >> - Add a call in reflog_expire_condition() >> - Link to v2: https://lore.kernel.org/r/20250709-jk-fix-leak-reflog-expire-config-v2-1-f9af934be8c1@gmail.com > > This looks correct to me except... > >> diff --git a/builtin/gc.c b/builtin/gc.c >> index 845876ff0286..37f543736599 100644 >> --- a/builtin/gc.c >> +++ b/builtin/gc.c >> @@ -346,6 +346,7 @@ static int reflog_expire_condition(struct gc_config *cfg UNUSED) >> count_reflog_entries, &data); >> >> reflog_expiry_cleanup(&data.policy); >> + reflog_clear_expire_config(&data.policy); >> return data.count >= data.limit; >> } >> > > This needs to pass &data.policy.opts, no? >
You're right... I think I fixed that and forgot to actually commit it before sending. Ugh.
Show 6 quoted lines
> I think we might also want this test on top (or I'd be happy to see it > squashed in). It shows off your fix when built with SANITIZE=leak, and > also catches the bug that v2 of your patch had. > > -Peff >
Sounds good. I'll send a v4 which squashes this in.
Show 55 quoted lines
> -- >8 --
> Subject: [PATCH] t1410: add test of gc.<pattern>.reflogExpire config
>
> We have long supported the ability to set reflog expiration config for
> individual, going back to 3cb22b8efe (Per-ref reflog expiry
> configuration, 2008-06-15). But we have never had any tests.
>
> Let's add a very basic one that checks that we apply the config
> correctly to a subset of refs (and not elsewhere). This also
> triggers the leaky code fixed by the previous commit.
>
> Signed-off-by: Jeff King <peff@peff.net>
> ---
> t/t1410-reflog.sh | 28 ++++++++++++++++++++++++++++
> 1 file changed, 28 insertions(+)
>
> diff --git a/t/t1410-reflog.sh b/t/t1410-reflog.sh
> index 42b501f163..362e90d7d6 100755
> --- a/t/t1410-reflog.sh
> +++ b/t/t1410-reflog.sh
> @@ -320,6 +320,34 @@ test_expect_success 'git reflog expire unknown reference' '
> test_grep "error: reflog could not be found: ${SQ}does-not-exist${SQ}" stderr
> '
>
> +test_expect_success 'expire with pattern config' '
> + # Split refs/heads/ into two roots so we can apply config to each. Make
> + # two branches per root to verify that config is applied correctly
> + # multiple times.
> + git branch root1/branch1 &&
> + git branch root1/branch2 &&
> + git branch root2/branch1 &&
> + git branch root2/branch2 &&
> +
> + test_config "gc.reflogexpire" "never" &&
> + test_config "gc.refs/heads/root2/*.reflogExpire" "now" &&
> + git reflog expire \
> + root1/branch1 root1/branch2 \
> + root2/branch1 root2/branch2 &&
> +
> + cat >expect <<-\EOF &&
> + root1/branch1@{0}
> + root1/branch2@{0}
> + EOF
> + git log -g --branches="root*" --format=%gD >actual.raw &&
> + # The sole reflog entry of each branch points to the same commit, so
> + # the order in which they are shown is nondeterministic. We just care
> + # about the what was expired (and what was not), so sort to get a known
> + # order.
> + sort <actual.raw >actual.sorted &&
> + test_cmp expect actual.sorted
> +'
> +
> test_expect_success 'checkout should not delete log for packed ref' '
> test $(git reflog main | wc -l) = 4 &&
> git branch foo &&