Re: [PATCH] ref-filter: don't look for objects when outside of a repository

[Jeff King <peff@peff.net>]

On Fri, Nov 16, 2018 at 02:09:07PM +0900, Junio C Hamano wrote:

> >> I see problems in both directions:
> >> 
> >>  - sorting by "objectname" works now, but it's marked with SOURCE_OBJ,
> >>    and would be forbidden with your patch.  I'm actually not sure if
> >>    SOURCE_OBJ is accurate; we shouldn't need to access the object to
> >>    show it (and we are probably wasting effort loading the full contents
> >>    for tools like for-each-ref).
> >> 
> >>    However, that's not the full story. For objectname:short, it _does_ call
> >>    find_unique_abbrev(). So we expect to have an object directory.
> >
> > Oops, I'm apparently bad at reading. It is in fact SOURCE_OTHER, which
> > makes sense (outside of this whole "--sort outside a repo thing").
> >
> > But we'd ideally distinguish between "objectname" (which should be OK
> > outside a repo) and "objectname:short" (which currently segfaults).
> 
> Arguably, use of ref-filter machinery in ls-remote, whether it is
> given from inside or outside a repo, was a mistake in 1fb20dfd
> ("ls-remote: create '--sort' option", 2018-04-09), as the whole
> point of "ls-remote" is to peek the list of refs and it is perfectly
> normal that the objects listed are not available.

I think it's conceptually reasonable to use the ref-filter machinery.
It's just that it was underprepared to handle this out-of-repo case. I
think we're not too far off, though.

> "ls-remote --sort=authorname" that is run in a repository may not
> segfault on a ref that points at a yet-to-be-fetched commit, but it
> cannot be doing anything sensible.  Is it still better to silently
> produce a nonsense result than refusing to --sort no matter what the
> sort keys are, whether we are inside or outside a repository?

I don't think we produce silent nonsense in the current code (or after
any of the discussed solutions), either in a repo or out. We say "fatal:
missing object ..." inside a repo if the request cannot be fulfilled.
That's not incredibly illuminating, perhaps, but it means we fulfill
whatever we _can_ on behalf of the user's request, and bail otherwise.

If you are arguing that even in a repo we should reject "authorname"
early (just as we would outside of a repo), I could buy that.
Technically we can make it work sometimes (if we happen to have fetched
everything the other side has), but behaving consistently (and with a
decent error message) may trump that.

-Peff