Re: [PATCH] diff: do not reuse_worktree_file for submodules

From: Thomas Rast <tr@thomasrast.ch>
Date: Feb 23, 2014, 12:46 UTC
Message-ID: <87bnxyq9n6.fsf@thomasrast.ch>
In-Reply-To: <8738jbtmji.fsf@thomasrast.ch>

Thomas Rast <tr@thomasrast.ch> writes:

Show 27 quoted lines

> Junio C Hamano <gitster@pobox.com> writes:
>
>> Thomas Rast <tr@thomasrast.ch> writes:
>>
>>> @@ -2845,8 +2845,9 @@ static struct diff_tempfile *prepare_temp_file(const char *name,
>>>  		remove_tempfile_installed = 1;
>>>  	}
>>>  
>>> -	if (!one->sha1_valid ||
>>> -	    reuse_worktree_file(name, one->sha1, 1)) {
>>> +	if (!S_ISGITLINK(one->mode) &&
>>> +	    (!one->sha1_valid ||
>>> +	     reuse_worktree_file(name, one->sha1, 1))) {
>>
>> I agree with the goal/end result, but I have to wonder if the
>> reuse_worktree_file() be the helper function that ought to
>> encapsulate such a logic?
>>
>> Instead of feeding it an object name and a path, if we passed a
>> diff_filespec to the helper, it would have access to the mode as
>> well.  It would result in a more intrusive change, so I'd prefer to
>> see your patch applied first and then build such a refactor on top,
>> perhaps like the attached.
>
> I see that you already queued 721e727, which has the change you
> described plus moving the S_ISGITLINK test into reuse_worktree_file.
> The change looks good to me.

I spoke too soon; it breaks the test I wrote to cover this case, for a reason that gives me a headache.

When we hit the conditional

Show 5 quoted lines

>>> -	if (!one->sha1_valid ||
>>> -	    reuse_worktree_file(name, one->sha1, 1)) {
>>> +	if (!S_ISGITLINK(one->mode) &&
>>> +	    (!one->sha1_valid ||
>>> +	     reuse_worktree_file(name, one->sha1, 1))) {

sha1_valid=0 for the submodule on the worktree side of the diff. The reason is that we start out with sha1_valid=0 and sha1=000..000 for the worktree side of all dirty entries, which makes sense at that point. We later set the sha1 by looking inside the submodule in diff_fill_sha1_info(), but we never set sha1_valid. So the above conditional will now trigger on the !one->sha1_valid arm, completely defeating the change to reuse_worktree_file().

We can fix it like below, but it feels a bit wrong to me. Are submodules the only case where it makes sense to set sha1_valid when we fill the sha1?

 diff.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git i/diff.c w/diff.c
index dabf913..cf7281d 100644
--- i/diff.c
+++ w/diff.c
@@ -3081,6 +3082,8 @@ static void diff_fill_sha1_info(struct diff_filespec *one)
 				die_errno("stat '%s'", one->path);
 			if (index_path(one->sha1, one->path, &st, 0))
 				die("cannot hash %s", one->path);
+			if (S_ISGITLINK(one->mode))
+				one->sha1_valid = 1;
 		}
 	}
 	else

-- 
Thomas Rast
tr@thomasrast.ch