Merge bleeding_edge@1955 to 1.1 branch to improve compilation...

Merge bleeding_edge@1955 to 1.1 branch to improve compilation
cache behavior on for real web pages by allowing compiled
scripts to survive a few GCs in the cache.

Committed: http://code.google.com/p/v8/source/detail?r=1992

[Kasper Lund, 2009-05-18 08:54:47]

This is a change proposal for the 1.1 branch. It's pretty simple, but we haven't
shipped the code to the Dev channel yet, so I'm always a bit careful about these
things. Let me know how you want to proceed.

Impact: I have measured the cache hit rates after this change on certain news
sites and blogs - such as http://bug11341.blogspot.com/ - and when you click
around on the same site and go back a few times, the hit rates are usually in
the 50-70% range, which means that we avoid more than half of the expensive
script compilation. It only works when you stay with the same renderer process
(click links), but it has the potential to really speed up the browsing
experience for our users.

[Mads Ager (chromium), 2009-05-18 09:00:33]

LGTM

[Kasper Lund, 2009-05-18 09:11:52]

Update: I just verified that without this fix, we never hit the compilation
cache when browsing on http://bug11341.blogspot.com/, because the GCs we force
on page transitions effectively disable the cache.

[Kasper Lund, 2009-05-18 13:57:07]

Pushed to the 1.1 branch in revision 1992:

   http://v8.googlecode.com/svn/branches/1.1@1992

Feel free to revert this if you decide not to land this in 172.x and you need to
land other critical patches.

[Mark Larson, 2009-05-18 16:43:09]

On 2009/05/18 13:57:07, Kasper Lund wrote:
> Pushed to the 1.1 branch in revision 1992:
> 
>    http://v8.googlecode.com/svn/branches/1.1@1992
> 
> Feel free to revert this if you decide not to land this in 172.x and you need
to
> land other critical patches.

OK. The 172 release is code-frozen at the moment, so this will only go in if we
decide to delay ship (or when we do a 172 update).