Re: [PATCH 2/3] diff-merges: cleanup set_diff_merges()

From: Sergey Organov <sorganov@gmail.com>
To: Junio C Hamano <gitster@pobox.com>
Cc: git@vger.kernel.org
Subject: Re: [PATCH 2/3] diff-merges: cleanup set_diff_merges()
Date: Fri, 16 Sep 2022 16:38:45 +0300	[thread overview]
Message-ID: <87pmfvjv6i.fsf@osv.gnss.ru> (raw)
In-Reply-To: <xmqq35csmkuq.fsf@gitster.g> (Junio C. Hamano's message of "Thu, 15 Sep 2022 13:41:17 -0700")

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

> Sergey Organov <sorganov@gmail.com> writes:
>
>> Get rid of special-casing of 'suppress' in set_diff_merges(). Instead
>> set 'merges_need_diff' flag correctly in every option handling
>> function.
>>
>> Signed-off-by: Sergey Organov <sorganov@gmail.com>
>> ---
>>  diff-merges.c | 30 +++++++++++++++++++-----------
>>  1 file changed, 19 insertions(+), 11 deletions(-)
>
> Looks OK to me.
>
> Everybody else says set_X() but set_none() has nothing to do ther
> than calling suppress(), so the change does not really make a
> functional difference (as you said in the cover letter).
>
> Is the idea that the original value in .merges_need_diff member does
> not matter because in every case it is set to either 0 or 1?  Most
> cases call common_setup() to set it to 1 like this here...

Yes, exactly.

>
>> +static void common_setup(struct rev_info *revs)
>> +{
>> +	suppress(revs);
>> +	revs->merges_need_diff = 1;
>> +}
>
> ... but this does not touch (in other words, it does not explicitly
> clear) it, ...

>
>> +static void set_none(struct rev_info *revs)
>> +{
>> +	suppress(revs);
>> +}
>
>  ... so we still rely on somebody to set the .merges_need_diff
> to 0 initially, right?

No, the suppress() does clear everything, .merges_need_diff included.
The suppress() ensures every next diff-merges option on the command-line
actually overwrites and correctly sets everything.

>
>> -
>> -	/* NOTE: the merges_need_diff flag is cleared by func() call */
>> -	if (func != suppress)
>> -		revs->merges_need_diff = 1;
>>  }
>
> It is very good to see this one go.

Yep, that was a kludge that bothered me for a while indeed.

>
>>  /*
>> @@ -115,6 +122,7 @@ int diff_merges_parse_opts(struct rev_info *revs, const char **argv)
>>  
>>  	if (!suppress_m_parsing && !strcmp(arg, "-m")) {
>>  		set_to_default(revs);
>> +		revs->merges_need_diff = 0;
>
> I am wondering how this becomes necessary?  Is it because
> set_to_default() would flip the member to 1 unconditionally, or
> something?

Yes, as all "native" -diff-merge= options set this bit to 1.

> If it weren't for this hunk, the lossage of the previous
> hunk is a very good clean-up, but if we need to do this, I cannot
> shake the feeling that we mostly shifted the dirt around, without
> really cleaning it?  I dunno.

Yes, I believe it does cleanup things in both these places.

The "-m" option indeed differs from the rest of the options in exactly
this thing: it wants .merges_need_diff=0 to suppress output unless '-p'
is given as well.

The -c/--cc are in fact similar, but they don't need this line as they
imply '-p' that in turn ignores .merges_need_diff, and generates output
anyway, so -m code has:

  revs->merges_need_diff = 0;

whereas -c and --cc code both have:

  revs->merges_imply_patch = 1;

Thanks,
-- Sergey Organov

>
>> @@ -125,7 +133,7 @@ int diff_merges_parse_opts(struct rev_info *revs, const char **argv)
>>  		set_remerge_diff(revs);
>>  		revs->merges_imply_patch = 1;
>>  	} else if (!strcmp(arg, "--no-diff-merges")) {
>> -		suppress(revs);
>> +		set_none(revs);
>
> We do not need to explicitly set .merges_need_diff to 0 here,
> presumably because it is initialized to 0 and nobody touched it,
> right?

No, we rather do set it to 0 here, in suppress() called from set_none().

Thanks,
-- Sergey Organov