Re: [PATCH] t4205: correctly test %(describe:abbrev=...)

From: Junio C Hamano <gitster@pobox.com>
Date: Jun 28, 2023, 21:30 UTC
Message-ID: <xmqqv8f7b7h1.fsf@gitster.g>
In-Reply-To: <20230628181753.10384-1-five231003@gmail.com>

Kousik Sanagavarapu <five231003@gmail.com> writes:

> The pretty format %(describe:abbrev=<number>) tells describe to use only
> <number> characters of the oid to generate the human-readable format of
> the commit-ish.

Is that *only* correct? I thought it was "at least <number> hexdigits" to allow for future growth of the project.

> This is not apparent in the test for %(describe:abbrev=...) because we
> directly tag HEAD and use that, in which case the human-readable format
> is just the tag name. So, create a new commit and use that instead.

Nice. How was this found, I have to wonder, and more importantly, how would we have written this test in the first place to avoid testing "the wrong thing", to learn from this experience?

Show 7 quoted lines

>  test_expect_success '%(describe:abbrev=...) vs git describe --abbrev=...' '
> -	test_when_finished "git tag -d tagname" &&
> -	git tag -a -m tagged tagname &&
> +	test_commit --no-tag file &&
>  	git describe --abbrev=15 >expect &&
>  	git log -1 --format="%(describe:abbrev=15)" >actual &&
>  	test_cmp expect actual

The current test checks that the output in the case where the number of commits since the tag is 0, and "describe --abbrev=15" and "log --format='%(describe:abbrev=15)'" give exactly the same result. Which is a good thing to test.

But we *also* want to test a more typical case where there are commits between HEAD and the tag that is used to describe it.

And we *also* want to make sure that the hexadecimal object name suffix used in the description is at least 15 hexdigits long, if not more.

The updated test drops test #1 (which is questionable), adds test #2 (which is good), and still omits test #3 (which is not so good).

So, perhaps

    test_when_finished "git tag -d tagname" &&
-   git tag -a -m tagged tagname &&
    test_commit --no-tag file &&
    git describe --abbrev=15 >expect &&
    git log -1 --format="%(describe:abbrev=15)" >actual &&
    test_cmp expect actual &&
+   sed -e "s/^.*-g\([0-9a-f]*\)$/\1/" <actual >hexpart &&
+   test 16 -le $(wc -c <hexpart) &&
+
+   git tag -a -m tagged tagname &&
+   git describe --abbrev=15 >expect &&
+   git log -1 --format="%(describe:abbrev=15)" >actual &&
+   test_cmp expect actual &&
+   test tagname = $(cat actual)

or something along the line? First we test with a commit that is not tagged at all to have some commits between the tag and HEAD with the original comparison (this is for #2), then we make sure the length of the hexpart (new---this is for #3), and then we add the tag to see the "exact" case also works (this is for #1).

Thanks.