[PATCH 1 of 2 STABLE V2] context: add 'changectx.dirs()' to recognize directory matching pattern

Matt Mackall mpm at selenic.com
Tue Feb 21 16:36:07 CST 2012


On Tue, 2012-02-21 at 16:40 +0900, FUJIWARA Katsunori wrote:
> At Mon, 20 Feb 2012 14:12:20 -0600,
> Matt Mackall wrote:
> > 
> > On Sun, 2012-02-19 at 19:25 +0900, FUJIWARA Katsunori wrote:
> > > # HG changeset patch
> > > # User FUJIWARA Katsunori <foozy at lares.dti.ne.jp>
> > > # Date 1329646846 -32400
> > > # Branch stable
> > > # Node ID 59430cdb845b58c916547dd5623ee438abcb2814
> > > # Parent  0e0060bf2f440d5cc33e5f36d99868a5380debd4
> > > context: add 'changectx.dirs()' to recognize directory matching pattern
> > 
> > I don't want util.adddirs (and if I did want it, I would want it in its
> > own patch!). It's way too special-purpose for util and indeed only has
> > one user.
> > 
> > > +    @propertycache
> > > +    def _dirs(self):
> > > +        dirs = set()
> > > +        for f in self._manifest:
> > > +            util.adddirs(f, dirs)
> > > +        return dirs
> > 
> > You've added a scope lookup and a function call (to a single use
> > function) to the inner loop. Both unfortunate things for a language with
> > effectively no optimization. Much better to just inline adddirs here.
> > 
> > I still think we should have a workingctx version of this method that
> > reuses the dirstate directory data rather than building another copy.
> 
> Thank you for your comments !
> 
> I want to confirm about 'workingctx' version of this.
> 
> For this bug fix, 'match.bad()' is invoked only in 'working' case, in
> which 'ctx1' should be 'changectx', not 'workingctx'.
> 
> In my plan to use 'chagectx.dirs()' for largefiles bug fix, I'll use
> 'dirstate' directly for working context, because 'changectx.dirs()' is
> not enough for some reasons.
> 
> So, I plan to define 'dirs()' only in 'changectx' now.
> 
> But should I define it also in 'workingctx' to reduce building cost
> for latent purposes ? If so, I'll define it like:
> 
>     def dirs(self):
>         return self._repo.dirstate.dirs()
> 
> # this requires 'dirstate.dirs()' as public interface of 'dirstate._dirs'

Yes, sounds good.

-- 
Mathematics is the supreme nostalgia of our time.




More information about the Mercurial-devel mailing list