From: Phillip Wood <phillip.wood123@gmail.com>
Date: Fri, 10 Dec 2021 14:13:45 GMT
Subject: Re: builtin add interactive regression
Message-ID: <53fdf2c2-1e01-a105-6a20-8115f88b39f6@gmail.com>
In-Reply-To: <20211209193625.GA3294@szeder.dev>

On 09/12/2021 19:36, SZEDER Gábor wrote:
> Hi,
> 
> This morning 'git add --patch' told me "Sorry, cannot split this hunk"
> even though that hunk in question was definitely splittable.  Then I
> spent my lunch break trying to reproduce the issue, and it turned out
> to be a regression in the builtin add interactive (I have
> 'add.interactive.useBuiltin=true' in my config).  The following test
> demonstrates this issue:
> 
>    ---  >8  ---
> 
> #!/bin/sh
> 
> test_description='test'
> . ./test-lib.sh
> 
> test_expect_success 'builtin interactive add cannot split hunk?!' '
> 	printf "%s\n" 1 2 >file &&
> 	echo a >zzzz &&
> 	git add file zzzz &&
> 	git commit -m initial &&
> 	cat >file <<-\EOF &&
> 	1
> 	add a line
> 	2
> 	add a line at the end (this is important!)
> 	EOF
> 	echo "modify one more file (this, too, is important!)" >zzzz &&
> 
> 	git diff file &&
> 
> 	git -c add.interactive.useBuiltin=false add -u -p >expect &&
> 	git -c add.interactive.useBuiltin=true add -u -p >actual &&
> 	# Uh-oh, "s" (for splitting the hunk) is missing!
> 	test_cmp expect actual
> '
> 
> test_done
> 
>    ---  8<  ---
> 
> This fails with:
> 
>    + git -c add.interactive.useBuiltin=false add -u -p
>    + git -c add.interactive.useBuiltin=true add -u -p
>    + test_cmp expect actual
>    --- expect	2021-12-09 19:21:23.354461170 +0000
>    +++ actual	2021-12-09 19:21:23.358461215 +0000
>    @@ -7,7 +7,7 @@
>     +add a line
>      2
>     +add a line at the end (this is important!)
>    -(1/1) Stage this hunk [y,n,q,a,d,s,e,?]?
>    +(1/1) Stage this hunk [y,n,q,a,d,e,?]?
>     diff --git a/zzzz b/zzzz
>     index 7898192..84e1b35 100644
>     --- a/zzzz
>    error: last command exited with $?=1
>    not ok 1 - builtin interactive add cannot split hunk?!
> 
> So the builtin add interactive doesn't even offer the 's - split the
> current hunk into smaller hunks' option, but my finger memory pressed
> 's' anyway, and then it told me that "Sorry..." message.  The scripted
> version can split the hunk just fine.

Thanks for the easy reproduction recipe, the hunk below makes it
pass. We're already checking marker at the very end of the loop
but we need to do it after the end of each diff (I think the perl
version only processes one diff at a time so does not have to
worry about this). If anyone wants to take this forward please
do, if not I'll try and find some time later next week to turn it
into a proper patch with a commit message and test.

Best Wishes

Phillip

---- >8 -----
diff --git a/add-patch.c b/add-patch.c
index 8c41cdfe39..4c2db78460 100644
--- a/add-patch.c
+++ b/add-patch.c
@@ -472,6 +472,14 @@ static int parse_diff(struct add_p_state *s, const struct pathspec *ps)
                         eol = pend;
  
                 if (starts_with(p, "diff ")) {
+                       if (marker == '-' || marker == '+')
+                               /*
+                                * Last hunk ended in non-context line
+                                * (i.e. it appended lines to the
+                                * file, so there are no trailing
+                                * context lines).
+                                */
+                               hunk->splittable_into++;
                         ALLOC_GROW_BY(s->file_diff, s->file_diff_nr, 1,
                                    file_diff_alloc);
                         file_diff = s->file_diff + s->file_diff_nr - 1;