Re: Necko e10s specs

On 4/09/09 8:48 AM, Benjamin Smedberg wrote:
> I was vaguely aware, but we got a lot more aware of them today. I've posted
> some questions/proposals to WhatWG: the current global storage mutex is
> really unpleasant and I think we'd like to avoid it if it's not too painful
> to do so.

I think we need to grapple with the storage mutex at some point anyway, 
for localStorage. I think it would be really good for the Web if we 
could implement the currently specced requirements in a performant way.

The storage mutex doesn't have to be implemented as a single global 
lock. I was thinking we could define an "entangled" relation over 
documents, where two documents with the same domain are entangled, and 
two documents running in the same process are entangled, and the 
relation is transitively closed. Then we could partition those documents 
into equivalence classes and have one "storage mutex" per equivalence 
class. If we maintain this relation and the storage mutex(es) in the 
master process, how hard would this be?

Rob