1372207 - JS bytecode cache should use Idle callback to trigger the encoding.

Status: RESOLVED FIXED

(Core :: DOM: Core & HTML, enhancement)

Description

[Nicolas B. Pierron [:nbp]]

At the moment, the JSBC uses the next cycle after the load-end event and the last script tag to be executed (in case the load event were to inject a new script tag).

Instead of using the next cycle after the load-end event, we should use the next idle-time after the load-end event, such that we any code which is spinning the event loop after the load end can still be captured by the bytecode cache encoding.

Comment 1

[Nicolas B. Pierron [:nbp]]

Attachment: Schedule JSBC encoding to be executed at the next idle time after the onload event.

This replaces the NS_DispatchToCurrentThread by the document event target
for the idle callback.

This code seems to be the first code to ever use the
TaskCategory::IdleCallback, is that the proper category that I should be
using for this task?

Comment 2

[Olli Pettay [:smaug][[e-mail]]]

TaskCategory::IdleCallback doesn't mean anything yet. (It is very confusing to have it. )
We need to change idle dispatch to use all this TaskCategory stuff.

Also, should we ensure EncodeBytecode is called eventually? If so, http://searchfox.org/mozilla-central/rev/d840ebd5858a61dbc1622487c1fab74ecf235e03/xpcom/threads/nsThreadUtils.cpp#368 is probably the right method to use.

(You may want to ask farre too)

Comment 3

[Nicolas B. Pierron [:nbp]]

(In reply to Olli Pettay [:smaug] from comment #2)
> Also, should we ensure EncodeBytecode is called eventually?

We do not have to, as the code could be clean-up by the destructor of the ScriptLoader, but the NewRunnableMethod has an Owning reference to the ScriptLoader, so in the current implementation we would have to ensure that it is called such that we can later remove the ScriptLoader instance.

> If so,
> http://searchfox.org/mozilla-central/rev/
> d840ebd5858a61dbc1622487c1fab74ecf235e03/xpcom/threads/nsThreadUtils.cpp#368
> is probably the right method to use.

Ok, I will use that instead.

Comment 4

[Nicolas B. Pierron [:nbp]]

Attachment: Schedule JSBC encoding to be executed at the next idle time after the onload event.

Use NS_IdleDispatchToCurrentThread instead of NS_DispatchToCurrentThread.

Comment 5

[Olli Pettay [:smaug][[e-mail]]]

Comment on attachment 8877123 [details] [diff] [review]
Schedule JSBC encoding to be executed at the next idle time after the onload event.

This makes it more likely that EncodeBytecode never runs during the lifetime of document, but I guess that is fine, since that has been possible before too, in case document lifetime is short.

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/mozilla-inbound/rev/fccb81aa0a57
Schedule JSBC encoding to be executed at the next idle time after the onload event. r=smaug

Comment 7

[Sebastian Hengst [:aryx] (needinfo me if it's about an intermittent or backout)]

https://hg.mozilla.org/mozilla-central/rev/fccb81aa0a57