[Development] Proposal for "container-oriented deterministic memory manager"

[Phil Bouchard]

On 12/26/2016 06:43 AM, Thiago Macieira wrote:
> Em segunda-feira, 26 de dezembro de 2016, às 04:09:00 BRST, Phil Bouchard
> escreveu:
>> I can't say which version of WebKit is faster because this is
>> confidential but Blink uses a mark and sweep type of GC, which is pretty
>> much the cutting edge technology:
>> https://chromium.googlesource.com/chromium/src/+/master/third_party/WebKit/S
>> ource/platform/heap/BlinkGCAPIReference.md
>
> So you cannot tell us if Blink is faster on the same CPU as you tested WebKit?

Well there's WebKit, WebKit2, Blink, WebKit for Wayland but I can't 
share the results of the benchmarks.  The renderer differs so some are 
faster but the bottom line is they all use a mark and sweep GC running 
in a different thread.

> So how can we know that this cutting-edge GC will make things better?

I am not saying Blink GC will make things better.

> Note: I am not saying that we shouldn't adopt better technologies that improve
> performance, even if just a little. I am simply debating your claim that the
> GC is the whole reason behind the slowness.

Personally I just want the Javascript / QML to be blinking fast and a 
mixture of one of the aforementioned renderer, QtQuickCompiler 
on-demand, a decent CPU architecture and root_ptr could definitely get 
the job done once and for all.