Re: git clone of empty repositories doesn't preserve hash

From: Junio C Hamano <gitster@pobox.com>
Date: Apr 25, 2023, 22:24 UTC
Message-ID: <xmqqcz3rzjkg.fsf@gitster.g>
In-Reply-To: <ZEhHsJh20gtiDBd9@tapette.crustytoothpaste.net>

"brian m. carlson" <sandals@crustytoothpaste.net> writes:

Show 28 quoted lines

> So it looks like this has made it into master and works for v2 but
> breaks things for v0 and v1.  I noticed because the Git LFS testsuite is
> broken and the following test demonstrates it:
>
> ----- %< -----
> diff --git a/t/t5700-protocol-v1.sh b/t/t5700-protocol-v1.sh
> index 6c8d4c6cf1..3be5057579 100755
> --- a/t/t5700-protocol-v1.sh
> +++ b/t/t5700-protocol-v1.sh
> @@ -244,6 +244,17 @@ test_expect_success 'push with ssh:// using protocol v1' '
>  	grep "push< version 1" log
>  '
>  
> +test_expect_success 'clone propagates object-format from empty repo' '
> +	test_when_finished "rm -fr src256 dst256" &&
> +
> +	echo sha256 >expect &&
> +	git init --object-format=sha256 src256 &&
> +	git clone src256 dst256 &&
> +	git -C dst256 rev-parse --show-object-format >actual &&
> +
> +	test_cmp expect actual
> +'
> +
>  # Test protocol v1 with 'http://' transport
>  #
>  . "$TEST_DIRECTORY"/lib-httpd.sh
> ----- %< -----

Interesting.  I applied

 - the above addition to the t/t5700

 - reverse of 96f4113a (Merge branch 'jc/clone-object-format-from-void',
   2023-04-11), but only the part outside t/

on top of today's 'master', and tried to run t5700 and t5702 (the latter has new tests added by 96f4113a to protect the change in the topic from future breakage for v2). It seems both t5700 and t5702 fails.

The latter failing is very much expected; the code change reverted by the above experiment is what made it work in the first place.

But the former not working, after reverting the change, is totally unexpected---doesn't it mean that the change in question does not have much to do in the breakage?

In fact, the new test to 5700 applied on top of v2.40.1 fails, too.

Or are you complaining that the fix merged to 'master' only covers the current protocol and not historical v0/v1 protocol? I couldn't tell, and didn't get that impression from the phrasing "breaks", implying whatever that was "broken" used to be working.

Puzzled.