[PATCH 2 of 2] Fix income/pull with bundle and -R (issue 820)

Peter Arrenbrecht peter.arrenbrecht at gmail.com
Wed Dec 5 02:34:06 CST 2007


OK, sending another patch using setconfig().
-peter

On Dec 4, 2007 7:41 PM, Matt Mackall <mpm at selenic.com> wrote:
> On Tue, Dec 04, 2007 at 09:06:37AM +0100, Peter Arrenbrecht wrote:
> > On Dec 4, 2007 8:51 AM, Matt Mackall <mpm at selenic.com> wrote:
> > >
> > > On Tue, Dec 04, 2007 at 07:25:56AM +0100, Peter Arrenbrecht wrote:
> > > > On Dec 3, 2007 10:53 PM, Matt Mackall <mpm at selenic.com> wrote:
> > > > > On Mon, Dec 03, 2007 at 08:49:35PM +0100, Peter Arrenbrecht wrote:
> > > > > > +import changegroup, util, os, struct, bz2, tempfile, mdiff, hg
> > > > >
> > > > > I think that'll give us a circular include, which becomes problematic
> > > > > when demandloading is disabled. Ugh.
> > > >
> > > > So we go with the additional field in the ui class, then?
> > >
> > > Yeah, it at least won't have that problem. But it'll be an ugly
> > > layering violation. I'm tempted to say use setconfig("bundle","base").
> >
> > How would "use setconfig("bundle","base")" be less of a layering
> > problem than a new field?
>
> It isn't, it just looks less like one.
>
> The real problem is that -bundlerepos- are the layering violation in
> the current architecture. No two repo objects should need to know
> about each other's internals directly. That's a good thing. It forces
> everything to go through the wire protocol for one thing. But a
> bundlerepo intrinsically needs to know about the main repo because
> it's incomplete.
>
> Up until now bundlerepos have just been -guessing- where their base
> repo is, so eventually we'll have to introduce something gross like a
> ui hack or a global variable or bite the bullet and say "yes, it
> generally makes sense for the second repo in an operation to be able
> to poke at the interals of the first repo."
>
> So what's the value of making it -look- like less of layering
> violation? Basically, it keeps the ui class from growing barnacles.
>
> And it's really not so bad: we use setconfig to turn other command
> line options into config settings already. If we think of -R as just
> an alias for our weird config option, we can rationalize this pretty
> well.
>
>
> --
> Mathematics is the supreme nostalgia of our time.
>


More information about the Mercurial-devel mailing list