On Mon, Feb 25, 2019 at 12:37 PM Rusty Russell <rusty@xxxxxxxxxxxxxxx> wrote:
>
> Jeff King <peff@xxxxxxxx> writes:
> > On Fri, Feb 08, 2019 at 12:18:57PM +1030, Rusty Russell wrote:
> >
> >> This broke my "is this clean?" sanity check and very much violates
> >> the man page description.
> >
> > Wow, this one had me scratching my head for a minute. What you're
> > describing here:
> >
> >> $ git status -u no
> >> On branch guilt/repro
> >
> > ...seems horribly broken, and the behavior goes back to the beginnings
> > of "-u". So I wondered how we would not have noticed for so long.
> >
> > But what is going on is quite subtle. The "-u" option takes an optional
> > argument, and so must be used in its "stuck" form. I.e.,
> >
> > git status -uno
>
> Urgh, manual page even *says* this, and I missed it:
>
> It is optional: it
> defaults to all, and if specified, it must be stuck to the option (e.g. -uno, but not -u
> no).
>
> Note that optional arguments are fundamentally broken like this: you
> should *never* add them to your programs. If -u suddently wanted an
> argument, a new option should have been added.
This -uno thing is bugging me too and I've seen a couple more options
like this. Maybe we can start moving to -u=no or something like that.
It's still broken if we accept -= as a valid option (I don't think we
do as of now), but at least it does not look as misleading as -uno,
which reads -u -n -o by everybody else who has not read all the man
pages.
> But despite that "well, I wouldn't start from here!" advice, I think
> your patch seems really helpful.
>
> Thanks for clue contribution!
> Rusty.
>
> > does do what you expect. We cannot allow the separated form here,
> > because that would conflict with another actual option, like:
> >
> > git status -u --porcelain
> >
> > So why does it behave as it does? We accept "no" as its own pathspec
> > argument, and thus we limit the status to that path. And that's why
> > there's "nothing to commit"; there's nothing in that pathspec.
> >
> > This is a pretty horrible UI trap. Most of the time with pathspecs we
> > require them to be on the right-hand side of a "--" unless the paths
> > actually exist in the filesystem. But then, in most of those cases we're
> > making sure they're not ambiguous between revisions and paths. So maybe
> > it's overkill here. I dunno. But the patch below certainly makes what's
> > going on in your case less subtle:
> >
> > $ git status -u no
> > fatal: no: no such path in the working tree.
> > Use 'git <command> -- <path>...' to specify paths that do not exist locally.
> >
> > You do still have to figure out why it wasn't stuck to "-u", though.
> >
> > ---
> > diff --git a/builtin/commit.c b/builtin/commit.c
> > index 2986553d5f..7177d7d82f 100644
> > --- a/builtin/commit.c
> > +++ b/builtin/commit.c
> > @@ -1300,6 +1300,16 @@ static int git_status_config(const char *k, const char *v, void *cb)
> > return git_diff_ui_config(k, v, NULL);
> > }
> >
> > +static void verify_pathspec(const char *prefix, const char **argv)
> > +{
> > + while (*argv) {
> > + const char *arg = *argv++;
> > + if (!strcmp(arg, "--"))
> > + return;
> > + verify_filename(prefix, arg, 0);
> > + }
> > +}
> > +
> > int cmd_status(int argc, const char **argv, const char *prefix)
> > {
> > static int no_renames = -1;
> > @@ -1351,7 +1361,7 @@ int cmd_status(int argc, const char **argv, const char *prefix)
> > status_init_config(&s, git_status_config);
> > argc = parse_options(argc, argv, prefix,
> > builtin_status_options,
> > - builtin_status_usage, 0);
> > + builtin_status_usage, PARSE_OPT_KEEP_DASHDASH);
> > finalize_colopts(&s.colopts, -1);
> > finalize_deferred_config(&s);
> >
> > @@ -1362,6 +1372,7 @@ int cmd_status(int argc, const char **argv, const char *prefix)
> > s.show_untracked_files == SHOW_NO_UNTRACKED_FILES)
> > die(_("Unsupported combination of ignored and untracked-files arguments"));
> >
> > + verify_pathspec(prefix, argv);
> > parse_pathspec(&s.pathspec, 0,
> > PATHSPEC_PREFER_FULL,
> > prefix, argv);
--
Duy
