Re: diff-index --cc no longer permitted, gitk is now broken (slightly)

From: Sergey Organov <sorganov@gmail.com>
Date: Sep 16, 2021, 22:41 UTC
Message-ID: <874kaki1nb.fsf@osv.gnss.ru>
In-Reply-To: <xmqqy27wrzmj.fsf@gitster.g>

Junio C Hamano <gitster@pobox.com> writes:

Show 9 quoted lines

> Sergey Organov <sorganov@gmail.com> writes:
>
>> I'm afraid this issue is left up in the air after application of the
>> fix-up patch, as usage of --cc in the diff-index is still undocumented.
>
> Yeah, I do think documentation update is needed, but being buried by
> other topics I haven't had a chance to revisit the way how --cc is
> described in the wider context in order to make an intellgent
> suggestion on how to present it in the context of "diff-index".

I tend to believe yet another description of --cc is needed for diff-index, separate from the current one. Just saying.

Show 8 quoted lines

>
>> I.e., the fix-up just restores the historical status quo that has a
>> problem by itself.
>
> I do agree "show -p" on merge is an oddball that trips new people,
> because it does not imply the "do present the changes for merges"
> bit unlike "show -c/--cc" do, and from that point of view, the
> generalization --diff-merges tried to bring us was a good thing.

I'm not sure I follow. What "show -p" has to do with "diff-index --cc"?

My only point here is that usage of *--cc* in *diff-index* is entirely undocumneted, and that needs to be somehow resolved.

Show 5 quoted lines

>
> But "-c/--cc" are explicit enough in what they want to do.  It does
> want to present the changes to compare a single end state with
> possibly more than one starting state (e.g. a merge) and not
> requiring an explicit "-m" is quite natural.

Doesn't seem to be relevant for "diff-index --cc" lacking documentation, but -m and -c and --cc are rather *mutually exclusive*. I.e., they all set different formats for output of diffs for merges, so "--cc -m" == "-m", and "-m --cc" == "--cc", i.e., the latest overrides the format to be used. Therefore "requiring explicit -m for -cc" simply doesn't make sense.

Show 5 quoted lines

> Even more, when it compares the end state with only one starting state
> (e.g. showing a single parent commit), there is only one pairwise
> result to "combine", so it is also natural that it ends up showing the
> same output as "-p". So I do not quite see the behaviour of
> "diff*/show --cc" as a problem, though.

I don't see it as a problem as well, so whom you are arguing with?

The only problem in this particular case I see is that "diff-index --cc" is undocumented (and untested), and this has nothing to do with log/diff/show, unless I miss your point.

> IOW, the use pattern in gitk is more than just "historical status
> quo", but is quite sensible, I would have to say.

"diff-index --cc" in gitk is a bug, as according to Git documentation "diff-index" does not accept "--cc", period.

gitk trying to make sense of what is neither documented, nor tested, nor guaranteed is the problem, but I was talking even not about that problem, but rather about the cause of this: some undocumented processing of "git diff-index --cc".

Thanks, -- Sergey Organov