Do not force GC on each objects tracking sample.

Do not force GC on each objects tracking sample.

GC takes significant part of a sample timeframe. There's no real need
to make it on every sample.
Default GC invocations made by v8 should be enough for objects tracking
purposes.

[alph, 2013-06-05 13:39:18]

Folks, could you please take a look.

[yurys, 2013-06-05 13:45:36]

lgtm

[ulan, 2013-06-05 14:47:11]

lgtm

[alph, 2013-06-05 17:23:10]

As discussed offline, made it call GC once per second. Please take a look.

[loislo, 2013-06-05 17:29:59]

lgtm

[not_yurys, 2013-06-05 17:31:35]

On 2013/06/05 17:23:10, alph wrote:
> As discussed offline, made it call GC once per second. Please take a look.

If we want to do this to be done in v8 there should be a way to configure the
interval through its public API but I think we could trigger full gc from the
devtools front-end instead of hardwiring the interval in v8, wdyt?

[loislo, 2013-06-06 13:22:52]

On 2013/06/05 17:31:35, yurys wrote:
> On 2013/06/05 17:23:10, alph wrote:
> > As discussed offline, made it call GC once per second. Please take a look.
> 
> If we want to do this to be done in v8 there should be a way to configure the
> interval through its public API but I think we could trigger full gc from the
> devtools front-end instead of hardwiring the interval in v8, wdyt?

I'm not sure that it is good idea. We are pooling v8 from the backend code 20
times per second
and send events about the heap state. It'd be strange to force gc from frontend
in this scenario.

[yurys, 2013-06-06 13:35:19]

On 2013/06/06 13:22:52, loislo wrote:
> On 2013/06/05 17:31:35, yurys wrote:
> > On 2013/06/05 17:23:10, alph wrote:
> > > As discussed offline, made it call GC once per second. Please take a look.
> > 
> > If we want to do this to be done in v8 there should be a way to configure
the
> > interval through its public API but I think we could trigger full gc from
the
> > devtools front-end instead of hardwiring the interval in v8, wdyt?
> 
> I'm not sure that it is good idea. We are pooling v8 from the backend code 20
> times per second
> and send events about the heap state. It'd be strange to force gc from
frontend
> in this scenario.

Then it seems natural to trigger GC from the same place where we poll v8,
doesn't it? My point is that if we implement the functionality in v8 then it
should be configurable through the public API but it seems that it would be more
logical to implement this in Blink.