patchgit-diff-*: Allow "--name-only -z" as alias for "--name-only-z"

Junio C Hamano wrote:

Speaking as a user, I would get rather frustrated when I try to do that and it doesn't work... so the attached patch allows that usage. Please apply.

---

I've considered it, but what happens if you give -z first and then name-only?

Hi,

Junio C Hamano:

Exactly the same thing as vice versa. Or, even more exactly, my patch *makes* that happen. ;-)

Matthias Urlichs <smurf@smurf.noris.de> writes:

Ah, I was not being careful enough.  Sorry.

That said, I have been hating that diff options parsing for quite a while, and I've been thinking about cleaning it up along the lines I'll outline here, but have not done anything about it. Care to help me out?

 - In diff.h introduce these new stuff:

     struct diff_opts {
     int output_format;
     int detect_rename;
     ...
     };
     void diff_opts_init(struct diff_opts *);
     int diff_opts_parse(const char *, struct diff_opts *);
     int diff_opts_final(struct diff_opts *);

 - In diff-* brothers:

   - replace individual diff option variables with a single
     "static struct diff_opts diff_opts";

   - change the argument parsing code to do the following:

     diff_opts_init(&diff_opts);
     for each arg {
         /* common options to diff brothers are handled by
          * diff_opts_parse()
          */
         switch (diff_opts_parse(arg, &diff_opts)) {
         case 1: /* was a diff option and was parsed successfully */
         	continue;
         case -1: /* error */
                usage(diff_*_usage);
         }
         if (!strcmp())
              ... parsing of other options
     }
     if (diff_opts_final(&diff_opts))
         /* defaulting to HUMAN format when nothing specified,
          * complaining if find-copies-harder is specified but
          * -C was not, etc. is done in diff_opts_final().
          *
          * The complex if() chains that checks if we are in
          * name or in raw mode and switch output_format around
          * properly is what I missed in your patch, but I think
          * you can lose that by recording z-ness of the output
          * independently from the output format in diff_opts_parse()
          * and combining diff-raw vs diff-name and z vs non-z
          * in diff_opts_final().  That would make the code much
          * simpler.
          */
         usage(diff_*_usage);

 - In diff.h and diff.c, replace individual option parameters
   for the following functions to a single pointer to struct
   diff_opts:

     diff_setup(), diffcore_std(), diffcore_std_no_resolve(), diff_flush().  

We probably can make diff_scoreopt_parse() function static to diff.c once this is done.

We may want to rip out the independeant pickaxe, orderfile and filter support for diff-helper while we are at it, making it truly just a "diff-raw to diff-patch" converter.

Hmm?

Hi,

Junio C Hamano:

I saw the problem...

Sure -- assuming I find some time to actually do it over the next few days.

The problem is that this has been a problem lately. :-/

On Thu, 14 Jul 2005, Junio C Hamano wrote:

I didn't do what you suggested, but I _did_ split the "format" up into "format + line_termination", which in my opinion cleaned up part of it a _lot_.

So now "-z" only sets "line_termination" to NUL. "format" starts out as "DIFF_FORMAT_RAW" (which is the old HUMAN/MACHINE format - the difference between those two are now the line termination) but can be "PATCH" and "NAME".

Now, DIFF_FORMAT_PATCH + -z wouldn't seem to make any sense at all, but you can actually do so, and it actually makes some amount of sense for the case of

	git-diff-tree -v -p -z HEAD

where the "-z" means that the commit _message_ will be terminated by a NUL character, while the "-v" obviously means that the commit message will be printed at all, and the "-p" means that the diff gets printed as a patch.

But the diff obviously gets printed with newlines (as does any newlines _within_ the commit message), not with lines terminated by NUL's.

So "--name-only-z" no longer exists. It's "-z --name-only" (in any order, quite naturally).

			Linus

Linus Torvalds <torvalds@osdl.org> writes:

Agreed 100%. Regardless of the further cleanup I suggested, what you did was something I should have done in the first place. Thanks for the cleanup.