Move microtask queueing logic from JavaScript to C++

Move microtask queueing logic from JavaScript to C++

This avoids the appearence of a leak due to storing a JSObject
as the microtask_state in the strong root list, and allows callers
to call Isolate::RunMicrotasks() without having any v8::Context
available (as at least Blink has interest in doing).

The queue is now a strong root, represented as a FixedArray of JSFunctions
(or empty_fixed_array, if it's empty); it doubles in size when it needs to grow.
The number of elements in the queue is stored in Isolate::pending_microtask_count().

LOG=Y
R=dcarney@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=21356

[dcarney, 2014-05-16 17:41:50]

lgtm, nits

https://codereview.chromium.org/290633010/diff/1/src/heap.h
File src/heap.h (right):

https://codereview.chromium.org/290633010/diff/1/src/heap.h#newcode186
src/heap.h:186: V(JSObject, observation_state, ObservationState)                
            \
unrelated, but is this another context leak?

https://codereview.chromium.org/290633010/diff/1/src/isolate.cc
File src/isolate.cc (right):

https://codereview.chromium.org/290633010/diff/1/src/isolate.cc#newcode2273
src/isolate.cc:2273: while (pending_microtask_count() > 0) {
put handlescope here

https://codereview.chromium.org/290633010/diff/1/src/isolate.cc#newcode2280
src/isolate.cc:2280: for (int i = 0; i < num_tasks; i++) {
and here maybe

[dcarney, 2014-05-17 06:54:09]

https://codereview.chromium.org/290633010/diff/1/src/isolate.cc
File src/isolate.cc (right):

https://codereview.chromium.org/290633010/diff/1/src/isolate.cc#newcode2254
src/isolate.cc:2254: queue = factory()->NewFixedArray(1);
1 might be a little conservative here, maybe 4, 8 or whatever - depending on
what you see in the wild

[Yang, 2014-05-17 12:57:19]

On 2014/05/17 06:54:09, dcarney wrote:
> https://codereview.chromium.org/290633010/diff/1/src/isolate.cc
> File src/isolate.cc (right):
> 
> https://codereview.chromium.org/290633010/diff/1/src/isolate.cc#newcode2254
> src/isolate.cc:2254: queue = factory()->NewFixedArray(1);
> 1 might be a little conservative here, maybe 4, 8 or whatever - depending on
> what you see in the wild

Note that this would make running multiple microtasks somewhat more expensive.

[adamk, 2014-05-19 07:47:57]

https://codereview.chromium.org/290633010/diff/1/src/heap.h
File src/heap.h (right):

https://codereview.chromium.org/290633010/diff/1/src/heap.h#newcode186
src/heap.h:186: V(JSObject, observation_state, ObservationState)                
            \
On 2014/05/16 17:41:51, dcarney wrote:
> unrelated, but is this another context leak?

My reading of the code suggests it's not: Factory::NewMap() returns a map with a
null |constructor| field, so the JSObject that's put into observation_state
doesn't leak the context via its constructor (but it does lead to further
questions about whether "creation context" is a meaningful thing for all
objects).

https://codereview.chromium.org/290633010/diff/1/src/isolate.cc
File src/isolate.cc (right):

https://codereview.chromium.org/290633010/diff/1/src/isolate.cc#newcode2254
src/isolate.cc:2254: queue = factory()->NewFixedArray(1);
On 2014/05/17 06:54:09, dcarney wrote:
> 1 might be a little conservative here, maybe 4, 8 or whatever - depending on
> what you see in the wild

Let's go with 8.

https://codereview.chromium.org/290633010/diff/1/src/isolate.cc#newcode2273
src/isolate.cc:2273: while (pending_microtask_count() > 0) {
On 2014/05/16 17:41:51, dcarney wrote:
> put handlescope here

Done.

https://codereview.chromium.org/290633010/diff/1/src/isolate.cc#newcode2280
src/isolate.cc:2280: for (int i = 0; i < num_tasks; i++) {
On 2014/05/16 17:41:51, dcarney wrote:
> and here maybe

Done.

[adamk, 2014-05-19 07:57:25]

Committed patchset #2 manually as r21356 (presubmit successful).