Re: [PATCH v6] builtin/diff-files: convert ad hoc option parsing to parse-options API
- From
- Wasswa Edson <edsonwasswa54@gmail.com>
- Date
- Nov 2, 2026, 14:16 UTC
- Message-ID
- <CA+ew3JuD_FKhpw=pUT+p-r2750A0qyKTbTXZvxxvvNovjxLgyw@mail.gmail.com>
- In-Reply-To
- <xmqq8q3blfh1.fsf@gitster.g>
Hi Junio,
Thanks for the review. You make a good point about the value trade-off for a plumbing command - the loss of COMMON_DIFF_OPTIONS_HELP in the help output isn't worth the benefits parse-options would bring i had not seen that coming ......lol.
I'll drop this patch and look for a better target to work on.
Thanks for the guidance.
Best regards, Wasswa Edson
On Mon, 2 Nov 2026 at 13:01, Junio C Hamano <gitster@pobox.com> wrote:
Show 69 quoted lines
>
> Edson <edsonwasswa54@gmail.com> writes:
>
> > -static const char diff_files_usage[] =
> > -"git diff-files [-q] [-0 | -1 | -2 | -3 | -c | --cc] [<common-diff-options>] [<path>...]"
> > -"\n"
> > -COMMON_DIFF_OPTIONS_HELP;
> > +static const char * const diff_files_usage[] = {
> > + "git diff-files [-q] [-0 | -1 | -2 | -3 | -c | --cc] [<common-diff-options>] [<path>...]",
> > + NULL
> > +};
>
> The loss of COMMON_DIFF_OPTIONS_HELP here gives us a regression in
> the help output ("git diff-files -h"). By replacing the single string
> with a string array and passing it to parse_options() or
> show_usage_with_options_if_asked(), the COMMON_DIFF_OPTIONS_HELP
> macro is omitted.
>
> Since parse_options() or show_usage_with_options_if_asked() only
> knows about the options explicitly defined in the diff_files_options[]
> array (leaving standard diff options to be parsed by
> setup_revisions()), the common diff options (like -p, --stat, and -z)
> will no longer be documented when the user runs "git diff-files -h".
>
> You probably could shove COMMON_DIFF_OPTIONS_HELP into OPT_GROUP()
> or something that produces static text in the help output to avoid
> the regression if you really wanted to, but given that "diff-files"
> is plumbing meant for use by scripts, I am not sure how much value
> we get by rewriting its command-line parsing. Being able to say
> "git diff-files --bas" and get it understood as "--base" is
> something we may appreciate if it were a Porcelain command, but is
> it worth the hassle for a plumbing? I have to wonder.
>
> Especially when the parsing loop the patch wants to replace ...
>
> > argc = setup_revisions(argc, argv, &rev, NULL);
> > - while (1 < argc && argv[1][0] == '-') {
> > - if (!strcmp(argv[1], "--base"))
> > - rev.max_count = 1;
> > - else if (!strcmp(argv[1], "--ours"))
> > - rev.max_count = 2;
> > - else if (!strcmp(argv[1], "--theirs"))
> > - rev.max_count = 3;
> > - else if (!strcmp(argv[1], "-q"))
> > - options |= DIFF_SILENT_ON_REMOVED;
> > - else
> > - usage(diff_files_usage);
> > - argv++; argc--;
> > - }
>
> ... is not anything that is ugly or overly hard to understand.
>
> I dunno.
>
>
> > + argc = parse_options(argc, argv, prefix, diff_files_options,
> > + diff_files_usage, 0);
> > if (!rev.diffopt.output_format)
> > rev.diffopt.output_format = DIFF_FORMAT_RAW;
> > rev.diffopt.rotate_to_strict = 1;
> > @@ -73,7 +74,7 @@ int cmd_diff_files(int argc,
> > if (rev.pending.nr ||
> > rev.min_age != -1 || rev.max_age != -1 ||
> > 3 < rev.max_count)
> > - usage(diff_files_usage);
> > + usage_with_options(diff_files_usage, diff_files_options);
> >
> > /*
> > * "diff-files --base -p" should not combine merges because it