Default enable the Blink GC plugin.

Default enable the Blink GC plugin.

The Blink GC plugin helps to ensure correctness of the GC infrastructure.
Errors are described at: http://www.chromium.org/developers/blink-gc-plugin-errors
The plugin can be disabled using GYP_DEFINE blink_gc_plugin.

R=abarth@chromium.org, ager@chromium.org, thakis@chromium.org
BUG=

Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=173367

[zerny-chromium, 2014-04-30 13:59:49]

Now that we are shipping oilpan-enabled modules we should default enable the GC
plugin to help check correctness.

The plugin adds almost no overhead (<4s for a total build time of >9m). This is
because we mostly just check properties of the declaration structure. Only for
trace methods and destructors do we check statement level nodes.

Local compile times:

rm -r out/Release
GYP_DEFINES="clang=1 blink_gc_plugin=0" gclient runhooks
time ninja -j 100 -C out/Release blink_tests
...
[11304/11304] STAMP obj/build/blink_tests.actions_depends.stamp

real	9m15.325s
user	274m53.260s
sys	10m21.500s

rm -r out/Release
GYP_DEFINES="clang=1 blink_gc_plugin=1" gclient runhooks
time ninja -j 100 -C out/Release blink_tests
...
[11304/11304] STAMP obj/build/blink_tests.actions_depends.stamp

real	9m19.017s
user	276m1.310s
sys	10m31.670s

[wibling-chromium, 2014-04-30 14:26:07]

lgtm

[abarth-chromium, 2014-04-30 16:55:58]

lgtm

[Nico, 2014-04-30 17:53:23]

How is "-j 100" possibly a local build?

[abarth-chromium, 2014-04-30 19:19:16]

On 2014/04/30 17:53:23, Nico wrote:
> How is "-j 100" possibly a local build?

Nico, should we roll out this CL while we answer that question?

[Nico, 2014-04-30 19:20:11]

On 2014/04/30 19:19:16, abarth wrote:
> On 2014/04/30 17:53:23, Nico wrote:
> > How is "-j 100" possibly a local build?
> 
> Nico, should we roll out this CL while we answer that question?

Was it landed already?

[abarth-chromium, 2014-04-30 19:28:14]

On 2014/04/30 19:20:11, Nico wrote:
> On 2014/04/30 19:19:16, abarth wrote:
> > On 2014/04/30 17:53:23, Nico wrote:
> > > How is "-j 100" possibly a local build?
> > 
> > Nico, should we roll out this CL while we answer that question?
> 
> Was it landed already?

Ah, you're right.  /me was confused.

[zerny-chromium, 2014-05-01 06:35:44]

On 2014/04/30 17:53:23, Nico wrote:
> How is "-j 100" possibly a local build?

Yeah, the number is too high (copy/paste from my typical goma run), but that
will just starts more jobs than can be directly scheduled. In any case, -j 30
gives comparable results (9m17.590s vs 9m14.516s). When running goma, times are
so noisy that I can't get any useful comparison (often I get a faster compile
with the plugin enabled than with it disabled).

[Mads Ager (chromium), 2014-05-02 06:24:43]

LGTM

[Nico, 2014-05-05 18:42:21]

lgtm

I did 2 end-to-end local clobber builds of "blink_tests" with this off (37m22s,
36m50s) and on (37m52s, 39m7.339s) while not using the computer doing the build
for anything other than email. The difference feels low enough to qualify as
noise.

I wondered if this means that the cost of non-blink files (where the plugin
never runs) hides the cost, so I also did min-of-5 runs of doing

   rm
out_local/Release/obj/third_party/WebKit/Source/core/gen/blink/bindings/webcore_generated.V8GeneratedModulesBindings01.o
  time caffeinate ninja -C out_local/Release
obj/third_party/WebKit/Source/core/gen/blink/bindings/webcore_generated.V8GeneratedModulesBindings01.o

with both the plugin on and off; this required around 7.3s with both plugin on
and off too.

[zerny-chromium, 2014-05-06 07:00:21]

The CQ bit was checked by zerny@chromium.org

[commit-bot: I haz the power, 2014-05-06 07:00:35]

CQ is trying da patch. Follow status at
 https://chromium-status.appspot.com/cq/zerny@chromium.org/262533008/1

[zerny-chromium, 2014-05-06 07:02:32]

On 2014/05/05 18:42:21, Nico wrote:
> lgtm
> 
> I did 2 end-to-end local clobber builds of "blink_tests" with this off
(37m22s,
> 36m50s) and on (37m52s, 39m7.339s) while not using the computer doing the
build
> for anything other than email. The difference feels low enough to qualify as
> noise.
> 
> I wondered if this means that the cost of non-blink files (where the plugin
> never runs) hides the cost, so I also did min-of-5 runs of doing
> 
>    rm
>
out_local/Release/obj/third_party/WebKit/Source/core/gen/blink/bindings/webcore_generated.V8GeneratedModulesBindings01.o
>   time caffeinate ninja -C out_local/Release
>
obj/third_party/WebKit/Source/core/gen/blink/bindings/webcore_generated.V8GeneratedModulesBindings01.o
> 
> with both the plugin on and off; this required around 7.3s with both plugin on
> and off too.

Thanks for verifying the times. I too, get only minor differences when
recompiling after touching platform/heap/Handle.h which hits the most plugin
intensive parts of the code base.

[commit-bot: I haz the power, 2014-05-06 08:07:17]

Change committed as 173367

[Reid Kleckner, 2014-05-06 20:03:03]

This broke the Clang on Windows fyi builder:
http://build.chromium.org/p/chromium.fyi/builders/Chromium%20Win%20Clang/buil...

Can you disable this plugin on Windows for now?

[Nico, 2014-05-06 20:10:46]

 https://codereview.chromium.org/268833009/ is fixing the windows fyi bot

On Tue, May 6, 2014 at 1:03 PM,  <rnk@chromium.org> wrote:
> This broke the Clang on Windows fyi builder:
>
http://build.chromium.org/p/chromium.fyi/builders/Chromium%20Win%20Clang/buil...
>
> Can you disable this plugin on Windows for now?
>
> https://codereview.chromium.org/262533008/

To unsubscribe from this group and stop receiving emails from it, send an email
to blink-reviews+unsubscribe@chromium.org.