Re: [Monotone-devel] net.venge.monotone.deregexp

[Zack Weinberg]

On 2/28/07, Nathaniel Smith <address@hidden> wrote:
> > >if we are keeping some regexp engine around, that
> > >might simplify some other things.  (Don't have to rewrite the globish
> > >matcher by hand, etc.)
> >
> > I was thinking about grabbing the fnmatch() implementation from gnulib for
> > that.
> > [ http://www.gnu.org/software/gnulib/MODULES.html#module=fnmatch ]
>
> That might make sense for .mtn-ignore as well (or in any case),
> assuming we want to use something like real globs there.  globish
> is... not quite like real globs.  (No [] character matching, but with
> {} alternations.  And we can't trivially change it, because it's in
> the network protocol.)

Urk.  That makes me somewhat less enthusiastic about borrowing
fnmatch.  I don't think it would be a good idea if we have to modify
it at all, and it has no facility for disabling [] matching.  Nor do I
want to have two glob implementations.  And I think people would miss
[] matching in .mtn-ignore.

What would the consequences be to the network protocol if globish changed?

[... omitting stuff about pcre and the stack ...]

So I'm wondering what you see as the way forward, here.  There's a
flag day when we change .mtn-ignore, and possibly there's also a flag
day on the network protocol.  I think I've done all of the coding that
can be done without making decisions about that.

zw