Re: [PATCH] compdescribe fix for unsorted groups (workers/34814) (was: Re: completion: git: --fixup: problem with _describe -2V and duplicate commit subjects)

[Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx>]

On May 16, 10:54pm, Daniel Shahaf wrote:
} Subject: Re: [PATCH] compdescribe fix for unsorted groups (workers/34814) 
}
} Daniel Shahaf wrote on Thu, May 14, 2015 at 14:36:27 +0000:
} > The first attached patch seems to address this.  However, I encountered
} > an odd problem with it, which I have not been able to narrow down.  When
} > I apply the first patch, Daniel Hahler's recent series, and the second
} > attached patch, and then try to complete 'git checkout 1<TAB>',¹ I get
} > first a screenful of hashes, and then a screenful of descriptions, and
} > so on alternating.  I'm not sure whether this indicates a bug in the
} > first patch or an independent problem.
} 
} To be explicit: the problem goes away if I revert the first patch, but
} keep the others applied.

You mean revert *all* of "PATCH 1/3", that is, both the compcore.c hunk
(which seems to be a functional no-op) and all the hunks of computil.c?

The only difference I see from eyballing the patch in 35127 is in the
computil.c cd_get() hunk where before -2V would be set only on an
existing -J group, but post-patch might be set on an existing -V group.

I don't know why that would matter unless there is both a -J and a -V
group in the same set of opts, in which case -2V is going to be set for
whichever group is encountered first?  Am I understanding the intent of
this code correctly?

In any case I can't reproduce your reported problem, at least not quite
as you explain it above.

Starting from commit d52bf91659522435d68f719389095001f050b6c5, I applied
DH's seven patches and then your 35127, and:

torch% git checkout 1<TAB>
torch% git checkout 15
153a99d  -- 35154: NEWS on arithmetic evaluation changes (2 days ago)
15aa99b  -- 35139: complete the new (b) parameter flag (2 days ago)
153a99d  -- 35154: NEWS on arithmetic evaluation changes (2 days ago)
15aa99b  -- 35139: complete the new (b) parameter flag (2 days ago)

That doesn't really look correct because it lists everything twice, but
I get exactly the same thing with computil.c backed out.  If I back out
the rest of 35127, I get:

torch% git checkout 1<TAB>
torch% git checkout 15
153a99d  -- [HEAD^^]  35154: NEWS on arithmetic evaluation changes (2 days a
15aa99b  -- [HEAD~6]  35139: complete the new (b) parameter flag (2 days ago
153a99d  -- [HEAD^^]  35154: NEWS on arithmetic evaluation changes (2 days a
15aa99b  -- [HEAD~6]  35139: complete the new (b) parameter flag (2 days ago

Which I guess are not strictly duplicates because the descriptions differ.

If I then revert all of DH's patches (back to an empty 'git diff'):

torch% git checkout 1<TAB>
torch% git checkout 15
153a99d  -- [153a99d] 35154: NEWS on arithmetic evaluation changes
15aa99b  -- [15aa99b] 35139: complete the new (b) parameter flag

The extra line for each of these is a side-effect from DH's patch 3/7 in
workers/35101.

Incidentally _git_recent_objects is missing a "return ret" at the end (but
repairing that doesn't change the duplication).