gyp: cache conditional ASTs for 3 CPU seconds (5%) gain.

gyp: cache conditional ASTs for 3 CPU seconds (5%) gain.

This is in the parallel part so net gain is somewhere between
5/core_count% and 5%. In chromium on my computer I save roughly
1.5 seconds in total runtime.

BUG=362075
r1900

[Daniel Bratell, 2014-04-10 14:21:57]

This is one of a series (maybe 15-20 changes in total). They are almost all
small with a limited gain but add up to 30% faster gyp.

[scottmg, 2014-04-10 17:14:35]

input.py -> mark

[Daniel Bratell, 2014-04-11 08:24:37]

On 2014/04/10 17:14:35, scottmg wrote:
> input.py -> mark

I suspect this was a change someone had planned all along so maybe there was a
reason it didn't happen. There is no use splitting eval into compile+eval unless
you plan to cache the compile result.

[Daniel Bratell, 2014-04-14 09:51:22]

mark, can you please take a look? This is one of the bigger improvements, and
smaller patches, so I think it's worth looking at.

[Daniel Bratell, 2014-04-15 16:14:26]

I should mention that this patch is smaller than it looks. It's this big because
I extracted a block into a separate method to make the code easier to read and
easier to profile. The actual change is only a few lines (a dict with ASTs,
lookup and storage in it).

[Mark Mentovai, 2014-04-15 21:49:26]

LGTM

[Nico, 2014-04-16 15:37:41]

r1900