[PATCH] mq: refactor usage of repo.branchmap().iteritems() with itervalues()

[Brodie Rao]

On Wed, Sep 25, 2013 at 2:49 PM, Kevin Bullock
<kbullock+mercurial at ringworld.org> wrote:
> On 25 Sep 2013, at 4:38 PM, Brodie Rao wrote:
>
>> # HG changeset patch
>> # User Brodie Rao <brodie at sf.io>
>> # Date 1364871676 25200
>> #      Mon Apr 01 20:01:16 2013 -0700
>> # Node ID 5ddfc6e253794f76691b27f805b1275d54708298
>> # Parent  50d721553198cea51c30f53b76d41dc919280097
>> mq: refactor usage of repo.branchmap().iteritems() with itervalues()
>>
>> diff --git a/hgext/mq.py b/hgext/mq.py
>> --- a/hgext/mq.py
>> +++ b/hgext/mq.py
>> @@ -1222,9 +1222,7 @@ class queue(object):
>>         diffopts = self.diffopts()
>>         wlock = repo.wlock()
>>         try:
>> -            heads = []
>> -            for b, ls in repo.branchmap().iteritems():
>> -                heads += ls
>> +            heads = [h for hs in repo.branchmap().itervalues() for h in hs]
>
> The double-`for` is really confusing to me here.

If you insert some imaginary punctuation, it might make more sense:

[h <- for hs in repo.branchmap().itervalues(): for h in hs:]

That said, I just realized this could theoretically make the code
microscopically slower:

>>> timeit.timeit('things = [n for ns in stuff.itervalues() for n in ns]', 'stuff = {n: range(10 * n) for n in range(10)}') / 1000000.0
2.6555580139160157e-05

>>> timeit.timeit('things = []\nfor ns in stuff.itervalues():\n    things += ns\n', 'stuff = {n: range(10 * n) for n in range(10)}') / 1000000.0
4.299727916717529e-06

I guess the latter is faster because the final list ends up doing less
resizing as it's added to. I'm not sure if the performance difference
really matters though.