[PATCH] revset: improve performance of _generatorset.__contains__ (issue 4201)
durham at fb.com
Tue Mar 25 17:38:08 CDT 2014
On 3/25/14 2:26 PM, "Gregory Szorc" <gregory.szorc at gmail.com> wrote:
>On 3/25/14, 12:31 PM, Augie Fackler wrote:
>> On Tue, Mar 25, 2014 at 05:42:43PM +0000, Durham Goode wrote:
>>> On 3/24/14 8:01 PM, "Gregory Szorc" <gregory.szorc at gmail.com> wrote:
>>>> # HG changeset patch
>>>> # User Gregory Szorc <gregory.szorc at gmail.com>
>>>> # Date 1395716418 25200
>>>> # Mon Mar 24 20:00:18 2014 -0700
>>>> # Node ID 6f64e244c57706cfb123c32c4fadef63690eed40
>>>> # Parent 3879ac3858ffd9bb46e19fcc3a2b31d7bb2b54c5
>>>> revset: improve performance of _generatorset.__contains__ (issue 4201)
>>> Looks good to me. I tried it in our repo using this revset:
>> Agreed. Queued.
>FWIW, I've been running with this patch all day today and I've noticed
>performance improvements all across the board. Rebases are obviously
>faster. Evolve is much faster as well. Mercurial no longer feels
>sluggish when operating on my 200,000+ changeset repo.
I just found a bug in this patch. In:
# At least a part of the list should be cached if iteration has
# started over the generatorset.
for l in self._genlist:
for item in self._consumegen():
self._iterated = True
for item in self._gen:
self._cache[item] = True
If two revsets are running _consumegen on the same revset at the same
time, they could both be stepping through self._gen inside of their
respective _consumegen calls. Causing them to see different, mutually
exclusive outputs of _consumegen.
I noticed it due to changes I'm making to improve root() and
_descendants(), so I don't have a repro for @ at the moment. I'll try to
think up a fix.
More information about the Mercurial-devel