Re: [PATCH] refs: unify parse_worktree_ref() and ref_type()

[Junio C Hamano <gitster@pobox.com>]

Junio C Hamano <gitster@pobox.com> writes:

> Otherwise we fall through with worktree_ref that we have stripped
> main-worktree/ prefix, which means the original input
>
> 	main-worktree/worktrees/foo/blah
>
> is now 
>
> 	worktrees/foo/blah
>
> and the next skip_prefix() will see that it begins with "worktrees/".
> Of course, if the initial input were
>
> 	worktrees/foo/blah
>
> then we wouldn't have skipped main-worktree/ prefix from it, and go
> to the next skip_prefix().  So from here on, we cannot tell which
> case the original input was.
>
> But that is OK.  Asking "give me the ref 'blah' in the worktree 'foo'"
> in the current worktree should yield the same answer to the question
> "give me the ref 'blah' in the worktree 'foo', as if I asked you to
> do so in the main worktree".

This makes me wonder...

I wonder if it makes the resulting code clearer to go fully
recursive, unlike the posted code that says "if a recursive call
says it is for current, that means it is for main worktree, and
otherwise pretend as if the input did not have the prefix".

That is, something like

parse_worktree_ref(...) {

	... prepare name, name_len and ref ...

	if (skip_prefix(worktree_ref, "main-worktree/", &worktree_ref)) {
		parsed = parse_worktree_ref(worktree_ref, name, name_len, ref);
        	switch (parsed) {
		case REF_WORKTREE_CURRENT:
                case REF_WORKTREE_MAIN:
			return REF_WORKTREE_MAIN;
		case REF_WORKTREE_OTHER:
		case REF_WORKTREE_SHARED:
			return parsed;
		}
	}

	if (skip_prefix(worktree_ref, "worktrees/", &worktree_ref)) {
		slash = strchr(worktree_ref, '/');

		parsed = parse_worktree_ref(slash + 1, name, name_len, ref);
        	switch (parsed) {
		case REF_WORKTREE_CURRENT:
			return WORKTREE_OTHER; /* iffy */
                case REF_WORKTREE_MAIN:
			return REF_WORKTREE_MAIN;
		case REF_WORKTREE_OTHER:
		case REF_WORKTREE_SHARED:
			return parsed;
		}
	}

	/* Otherwise the input is like HEAD, MERGE_HEAD, refs/$BLAH */
	... do whatever for these trivial cases ...

}

And while composing this follow-up, I found another thing that is
iffy.  When you ask for "worktrees/foo/refs/bisect/good", the
recursive call in "worktrees/foo" for the remainder says "current",
and the posted patch and the above pseudocode would take CURRENT to
mean "that other worktree's", and turns it into OTHER i.e. "not
ours".  But the code does so without even checking what our worktree
is called.  What happens if we are the owner of the "worktrees/foo"
namespace?  The same thing for "The main worktree says the given ref
is 'current', so we can turn that into 'main'"---if our worktree is
'main', it is also 'current' and does not have to be turned into
'main' (even though it does not hurt).