[patch] syntax:plain for .hgignore

[Johannes Hofmann]

Just realized, that the culprit for the bad performance in our case
is the '.*' that is added in front of each pattern by regex() in
util.py.
Using syntax:regexp and preceeding each fixed pattern with a '^'
makes things fast.

  Johannes


On Thu, Sep 13, 2007 at 02:32:55PM -0500, Matt Mackall wrote:
> On Thu, Sep 13, 2007 at 08:18:46PM +0200, Guido Ostkamp wrote:
> > On Wed, 12 Sep 2007, Matt Mackall wrote:
> > >Can you mail your .hgignore file? Privately and/or obfuscated is fine.
> > 
> > I hope my private email with the hgignore has reached you in the meantime.
> > 
> > May I kindly ask, whether you have been able to reproduce the problem on 
> > your system?
> 
> Just took a stab at it. Adding syntax:regex before all the fixed
> patterns took it from .610s to .329s for an empty repo. Our glob to
> regex conversion could use some optimizing, apparently. Folding a
> bunch of fixed patterns together by hand took that down to .239s. Time
> to run hg -v is .078, so the ignore pattern building got 3.3 times
> faster and the resulting ignore file is 73 lines and 13k vs 624 lines
> and 23k in the original.
> 
> Now running this ignore file in my kernel repo:
> 
> no ignore file: 1.301s
> your original:  2.874s
> with regexp:    2.151s
> with folding:   1.535s
> 
> Assuming your ignore file isn't trimming substantial work from the
> tree walk, we can figure that 2.874 - 1.301 - (.610 - .078) = 1.041s
> for matching in your original, vs 1.535 - 1.301 - (.239 - .078) =
> .073s for matching for my improved version (14.2x).
> 
> Does this kind of speed improvement even make sense? Yes. All of the
> regex matching of the form foo/bar/baz is a simple string match. If we
> compare foo/bar/baz|foo/bar/quux with foo/bar/(baz|quux), we can very
> often eliminate an entire match immediately. As my optimized ignore
> file has about a tenth of the full path patterns as yours, it's doing
> a lot less backtracking work.
> 
> -- 
> Mathematics is the supreme nostalgia of our time.
> _______________________________________________
> Mercurial-devel mailing list
> Mercurial-devel at selenic.com
> http://selenic.com/mailman/listinfo/mercurial-devel