Retire GC lockout on initializing hash buckets.

Retire GC lockout on initializing hash buckets.

Do not disallow GCs across initialization of hash table buckets, be they
from freshly allocated tables or reuse of deleted entries. It serves
no real purpose to disallow GCs for the reasons given in the code
(the key isn't stored separately), and if the bucket is a deleted entry
being initialized, we already run the risk of having GCs with those
in the table. Hence, disallowing it extra here serves no purpose.

R=haraken
BUG=

Committed: https://crrev.com/514b91aef76eb537eea23900427d32d888c8d4ae
Cr-Commit-Position: refs/heads/master@{#351781}

[sof, 2015-10-01 08:36:13]

sigbjornf@opera.com changed reviewers:
+ oilpan-reviews@chromium.org, tkent@chromium.org

[sof, 2015-10-01 08:36:14]

please take a look.

I've been going over this one for the past day, trying to understand & justify
the explicit GC lockout done during bucket init..and I honestly can't see what
problem this is trying to solve. Hence, go bold & remove.

(The extra bookkeeping churn that this generates accounts for a 5% slowdown on a
hash table heavy test like blink_perf.svg.SvgNestedUse)

[haraken, 2015-10-01 08:39:29]

On 2015/10/01 08:36:14, sof wrote:
> please take a look.
> 
> I've been going over this one for the past day, trying to understand & justify
> the explicit GC lockout done during bucket init..and I honestly can't see what
> problem this is trying to solve. Hence, go bold & remove.
> 
> (The extra bookkeeping churn that this generates accounts for a 5% slowdown on
a
> hash table heavy test like blink_perf.svg.SvgNestedUse)

You cannot remove enterGCForbiddenScope since it is used in other places (e.g.,
HashTable.h, ListHashSet.h)?

[haraken, 2015-10-01 08:39:50]

OK, PS2 LGTM

[sof, 2015-10-01 08:41:51]

On 2015/10/01 08:39:29, haraken wrote:
> On 2015/10/01 08:36:14, sof wrote:
> > please take a look.
> > 
> > I've been going over this one for the past day, trying to understand &
justify
> > the explicit GC lockout done during bucket init..and I honestly can't see
what
> > problem this is trying to solve. Hence, go bold & remove.
> > 
> > (The extra bookkeeping churn that this generates accounts for a 5% slowdown
on
> a
> > hash table heavy test like blink_perf.svg.SvgNestedUse)
> 
> You cannot remove enterGCForbiddenScope since it is used in other places
(e.g.,
> HashTable.h, ListHashSet.h)?

Indeed, sorry for not catching that before uploading ps#1. I do understand the
need for those during swap().

[sof, 2015-10-01 12:23:51]

The CQ bit was checked by sigbjornf@opera.com

[commit-bot: I haz the power, 2015-10-01 12:23:56]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1378273002/20001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1378273002/20001

[commit-bot: I haz the power, 2015-10-01 12:28:42]

Committed patchset #2 (id:20001)

[commit-bot: I haz the power, 2015-10-01 12:29:24]

Patchset 2 (id:??) landed as
https://crrev.com/514b91aef76eb537eea23900427d32d888c8d4ae
Cr-Commit-Position: refs/heads/master@{#351781}