Re: [PATCH v9 1/2] t1092: add tests for `git diff-files`

From: Victoria Dye <vdye@github.com>
Date: May 3, 2023, 16:37 UTC
Message-ID: <2fa835b8-1c9b-67d3-aa4a-70a978b5f20d@github.com>
In-Reply-To: <xmqqjzxqzgbi.fsf@gitster.g>

Junio C Hamano wrote:

Show 18 quoted lines

> Shuqi Liang <cheskaqiqi@gmail.com> writes:
> 
>> +test_expect_success 'diff-files with pathspec outside sparse definition' '
>> +	init_repos &&
>> +
>> +	test_sparse_match test_must_fail git diff-files folder2/a &&
> 
> In "sparse" directories at this point of test, "folder2" is outside
> the cone(s) of interest and is not instantiated.  The reason why
> the command fails is because the command line parsing that is
> generic to all users of the revision machinery requires you to have
> a disambiguating double-dash before such a pathspec that tries to
> match a path that does not exist in the working tree and is not
> specific to "diff-files".
> 
> I wonder how interesting and useful this test is.  Without
> accompanying test that uses disambiguating double-dash properly,
> e.g. "git diff-files -- folder2", I doubt it is very much useful.

I agree, this test isn't helpful as-is. With the '--', 'diff-files' shouldn't fail, so in the interest of avoiding unexpected regressions in the future (masked by 'test_must_fail'), I think this test should be updated as you described.

Show 17 quoted lines

>> +	# file present on-disk without modifications
>> +	# use `--stat` to ignore file creation time differences in
>> +	# unrefreshed index
>> +	test_all_match git diff-files --stat &&
>> +	test_all_match git diff-files --stat folder1/a &&
>> +	test_all_match git diff-files --stat "folder*/a" &&
> 
> Because in all three repositories, "folder1/a" exists in the working
> tree, the "you need to disambiguate" error like the first test
> (whose utility I questioned) would not trigger.
> 
> What does this demonstrate, though?  That instantiating a file on
> the working tree, even outside the cone(s) of interest in a sparsely
> checked out working tree, makes it part of the interesting set
> automatically?  As there is no difference between the indexed
> contents and what is in the working tree, we cannot tell from this
> test if that is the case (not a complaint, just an observation).

This was meant [1] to check whether there are any issues expanding the index (specifically, the 'folder1/' sparse directory) before comparing to the now-on-disk 'folder1/a'.

However...

[1] https://lore.kernel.org/git/b537d855-edb7-4f67-de08-d651868247a5@github.com/

Show 17 quoted lines

> 
> But ...
> 
>> +	# file present on-disk with modifications
>> +	run_on_all ../edit-contents folder1/a &&
>> +	test_all_match git diff-files &&
>> +	test_all_match git diff-files folder1/a &&
>> +	test_all_match git diff-files "folder*/a"
> 
> ... it is shown by doing the same test with modified contents?
> 
> For consistency with the earlier "the same contents" test, we should
> use "--stat" here, too.  Or even "--stat -p".
> 
> Alternatively, we could refresh the index before running diff-files
> (here and also before the earlier "the same contents" test), I
> guess.

...to your point, we probably don't need both the "unmodified folder1/a diff-files" *and* "modified folder1/a diff-files" tests. In fact, the empty output of "unmodified folder1/a" could be caused by either "this file is unmodified" or "this file isn't in the index", so the test might pass even if there's an issue with index expansion. That isn't a problem in the "modified folder1/a" case, since we're expecting to see - and comparing the contents of - a diff.

I think we can drop the 'diff-files --stat' tests and go straight to the 'run_on_all ../edit-contents folder1/a'. Adding '--' here to disambiguate the pathspecs might be nice as well.

Show 6 quoted lines

> 
>> +'
>> +
>>  test_done
> 
> Thanks.

Thanks for the detailed review, apologies for missing these issues earlier.