|Summary:||fourthTier: CodeBlock should be RefCounted|
|Product:||WebKit||Reporter:||Filip Pizlo <fpizlo>|
|Severity:||Normal||CC:||barraclough, ggaren, mark.lam, mhahnenberg, msaboff, oliver, sam|
|Version:||528+ (Nightly build)|
|Bug Depends on:|
Description Filip Pizlo 2013-05-04 13:43:53 PDT
This will get rid of a bunch of weird ownership rules, and make the implementation of concurrent compilation a heck of a lot easier.
Comment 1 Geoffrey Garen 2013-05-04 13:50:37 PDT
Can we use GC for this instead? Inside the GC heap, it's easier when GC objects point to other GC objects.
Comment 2 Filip Pizlo 2013-05-04 13:54:23 PDT
(In reply to comment #1) > Can we use GC for this instead? Inside the GC heap, it's easier when GC objects point to other GC objects. Probably, eventually. But right now CodeBlock has a weird relationship with the GC - the GC does have a say in when a CodeBlock dies but not in the usual sense of tracing - so we'll have to be careful when we do that. There's also the issue that how a CodeBlock is traced depends on who owns it (either Executable or the GC itself), so it's not like making it GC'd would simplify its visit methods. For now, I don't see a clear benefit to making CodeBlock GC'd. It's true that doing so would mean that Executable would not have to be destructible, but I think that's really the only benefit. On the other hand, there is a clear benefit to making it ref-counted.
Comment 3 Filip Pizlo 2013-05-09 18:28:32 PDT
Created attachment 201308 [details] the patch Not setting r? yet because I'm still running tests.
Comment 4 Filip Pizlo 2013-05-09 18:44:19 PDT
(In reply to comment #3) > Created an attachment (id=201308) [details] > the patch > > Not setting r? yet because I'm still running tests. It's perf neutral.
Comment 5 Geoffrey Garen 2013-05-09 19:41:07 PDT
Comment on attachment 201308 [details] the patch r=me