Re: [PATCH v3 4/8] packed-backend: add "packed-refs" header consistency check

[Junio C Hamano <gitster@pobox.com>]

Patrick Steinhardt <ps@pks.im> writes:

> On Thu, Feb 06, 2025 at 01:59:04PM +0800, shejialuo wrote:
>> diff --git a/refs/packed-backend.c b/refs/packed-backend.c
>> index 6401cecd5f..683cfe78dc 100644
>> --- a/refs/packed-backend.c
>> +++ b/refs/packed-backend.c
>> @@ -1749,12 +1749,76 @@ static struct ref_iterator *packed_reflog_iterator_begin(struct ref_store *ref_s
>> +static int packed_fsck_ref_header(struct fsck_options *o,
>> +				  const char *start, const char *eol)
>> +{
>> +	if (!starts_with(start, "# pack-refs with:")) {
>> +		struct fsck_ref_report report = { 0 };
>> +		report.path = "packed-refs.header";
>> +
>> +		return fsck_report_ref(o, &report,
>> +				       FSCK_MSG_BAD_PACKED_REF_HEADER,
>> +				       "'%.*s' does not start with '# pack-refs with:'",
>> +				       (int)(eol - start), start);
>> +	}
>> +
>> +	return 0;
>> +}
>
> Okay. We still complain about bad headers, but only if there is a line
> starting with "#" and only if the prefix doesn't match. This addresses
> Junio's comment that packfiles don't have to have a header, and that
> they may contain capabilities that we don't understand.

We'd want to also ensure that there is a single trailing whitespace
after that colon, which we have always written after "with:", no?

>> diff --git a/t/t0602-reffiles-fsck.sh b/t/t0602-reffiles-fsck.sh
>> index 42c8d4ca1e..da321f16c6 100755
>> --- a/t/t0602-reffiles-fsck.sh
>> +++ b/t/t0602-reffiles-fsck.sh
>> @@ -639,4 +639,29 @@ test_expect_success SYMLINKS 'the filetype of packed-refs should be checked' '
>>  	)
>>  '
>>  
>> +test_expect_success 'packed-refs header should be checked' '
>> +	test_when_finished "rm -rf repo" &&
>> +	git init repo &&
>> +	(
>> +		cd repo &&
>> +		test_commit default &&
>> +
>> +		git refs verify 2>err &&
>> +		test_must_be_empty err &&
>> +
>> +		for bad_header in "# pack-refs wit: peeled fully-peeled sorted " \
>> +				  "# pack-refs with traits: peeled fully-peeled sorted " \
>> +				  "# pack-refs with a: peeled fully-peeled"
>
> Instead of verifying thrice that we complain about bad header prefixes,
> should we maybe replace two of these with instances where we check a
> packed-refs file _without_ a header and one with capabilities that we
> don't understand?

Yup.  I also notice that refs/packed-backend.c:create_snapshot()
would accept "# pack-refs with:peeled" if I am not reading it
correctly, which is an unrelated bug.

Thanks.