Proposal: move content JS interpretation to a background thread

[Honza Bambas]

Few reasons:
- I really don't believe we will soon/ever have a good OOP Firefox
implementation
- content JS execution very often kills my Firefox UI (and it is not I/O
that blocks)
- sync APIs like localStorage would not block the UI
- content JS GC would stop blocking UI as well
- generally would improve UI responsiveness


-hb-

[Kyle Huey]

You seem to say this as if this would be an easy project. There's
some highly experimental work in bug 718121, but pulling this off
requires a significant fraction of the work required to get e10s
working.

- Kyle

[Honza Bambas]

I would never think this was an easy project. I was not aware of that
bug, Jeff Muizelaar just CC'ed me on it. Thank you for enlightening me!

-hb-

[Robert Kaiser]

Honza Bambas schrieb:

> Few reasons:
> - I really don't believe we will soon/ever have a good OOP Firefox
> implementation

AFAIK, the major reason why we did abandon doing that was because moving
all our interaction with the content to be async was too much work for
the moment. Isn't that same work required for what you propose?

If so, I don't see why we should go for that instead of taking another
shot at moving content to a separate process - after all, we're using
that process separation code in B2G quite heavily as well, so it's not
that code itself that is problematic, it's making Firefox work well with
the parallelization it brings.

Robert Kaiser

[Ted Mielczarek]

On 11/9/2012 8:26 AM, Robert Kaiser wrote:
> AFAIK, the major reason why we did abandon doing that was because moving
> all our interaction with the content to be async was too much work for

> the moment. Isn't that same work required for what you propose?\

No. AIUI, the reason we abandoned electrolysis was because it meant that
a huge percentage of add-ons would have to be rewritten to work in the
new model. We could accept this as a one-time cost, but looking further
into the future, if Servo is a viable project and we decide to ship it
as Firefox we'd have to take that add-on compatibility hit again. Once
we can probably survive. Twice is too much to ask.

-Ted