From: Jeff King <peff@peff.net>
Date: Tue, 17 Feb 2009 19:56:58 GMT
Subject: Re: diff.defaultOptions implementation design [was diff.primer]
Message-ID: <20090217195658.GC16067@coredump.intra.peff.net>
In-Reply-To: <alpine.GSO.2.00.0902162312030.17111@kiwi.cs.ucla.edu>

On Mon, Feb 16, 2009 at 11:24:33PM -0800, Keith Cascio wrote:

> I like the idea of using parse-options to handle diff options and I
> too would like all switches negatable.  I will come back to the other
> ideas you mention if necessary.  You laid it all out nicely.

If you are interested in parse-optification of diff options, search the
archive for messages from Pierre Habouzit on the topic in the last 6
months or so. It was discussed at the GitTogether, and he had some
preliminary patches.

> diff_setup().  But diff_setup() must still ascertain at least one
> runtime fact: whether or not we are running one of the commands that
> respects default options {diff, log, show}.  Is there an elegant way
> to ascertain that fact from inside diff_setup()?  How do you
> recommend?  (BTW I believe my design achieves this elegantly).

You can impact the argument parsing by touching the diffopt struct
before doing the parsing. I.e., something like:

  /* we generally get diff options from a rev_info structure */
  struct rev_info rev;
  /* initialize the structures */
  init_revisions(&rev, prefix);
  /* now set any preferences specific to this command */
  DIFF_OPT_SET(&rev.diffopt, ALLOW_DEFAULT_CONFIG);
  /* and then actually parse */
  setup_revisions(argc, argv, rev, "HEAD");

See for example how cmd_whatchanged does it in builtin-log.c. Any
porcelains which wanted this feature would opt in to it.

-Peff