Re: [PATCH 2/3] revision.c: use skip_prefix() in handle_revision_opt()

[Jeff King <peff@peff.net>]

On Fri, Jun 09, 2017 at 08:17:28PM +0200, SZEDER Gábor wrote:

> > would let us do:
> >
> >   if (match_opt(arg, "--early-output"), &optarg)) {
> >         int count = optarg ? atoi(optarg) : 100;
> >         ...
> >   }
> >
> > which is a little nicer and could maybe help other options (I didn't see
> > any, though).
> 
> Besides '--show-linear-break' and '--pretty', other options that could
> benefit from this, i.e. long options with an optional argument, are
> '--expand-tabs', '--abbrev' and '--no-walk'.  These are handled
> differently than '--early--output' and '--show-linear-break': each is
> covered by two if branches, one with and one without the optional 
> argument, i.e.:
> 
>   } else if (!strcmp(arg, "--option")) {
>     ...
>   } else if (starts_with(arg, "--option=")) {
>     ...
>   } else ...

I think those multi-branch cases end up as an improvement with a helper:

  if (match_opt(arg, "--no-walk", &optarg)) {
	if (!optarg || !strcmp(optarg, "sorted"))
		revs->no_walk = REVISION_WALK_NO_WALK_SORTED;
	else if (!strcmp(optarg, "unsorted"))
		revs->no_walk = REVISION_WALK_NO_WALK_UNSORTED;
	else
		return error(...);
  }

> '--pretty=' couldn't benefit, though, because it is special in that
> it's equivalent with '--format=', and the two are handled in the same
> branch.

I think you could still handle them both in the same branch, like:

  if (match_opt(arg, "--pretty", &optarg) ||
      skip_prefix(arg, "--format=", &optarg)) {
       revs->verbose_header = 1;
       revs->pretty-given = 1;
       /* OK to pass NULL for --pretty case */
       get_commit_format(optarg, revs);
  }

> So inherently there are a few repeated option names and variable
> assignments, and that's not so good.  However, refactoring these to
> use match_opt() adds 40% more lines than it removes and, more
> importantly, increases the number of nested conditions.  Subjectively
> I don't think it's better, so I went with the "follow the conventions
> of the surrounding code" rule for the update.

I care less about lines of boilerplate code and more about repeated
logic. In the --pretty example above, the first two lines of the block
are common to both --pretty and --pretty=. If they ever need to change,
somebody has to update two spots.

Anyway. I certainly don't insist on you working on this, especially if
you don't agree with the aesthetics. Just fixing the actual bugs would
be sufficient for my review. ;)

> As far as I can tell, parse-options doesn't handle options with an
> optional argument by itself, but only with callback functions, so it
> is no help here as it is.

There's a flag, PARSE_OPT_OPTARG, which would do what you want. But I
agree that converting the whole thing to parse-options would be a lot of
work (quite a few of these really aren't just "this is a string", but
would need independent callback functions.

-Peff