JSC needs an API to allow custom objects to have aprivate GC-accessible properties

[Oliver Hunt]

JSC needs an API to allow custom objects to have aprivate GC-accessible properties

[Oliver Hunt]

Created attachment 51238 [details]
Patch

[Maciej Stachowiak]

Comment on attachment 51238 [details]
Patch

Comments given on IRC:

Docs:

the headerdoc could maybe explain the purpose of this
i.e. that it's a way to hold on to JavaScript values in a way that is safe for GC
I think probably in the set one would be the best place to explain the purpose
also maybe add a note to JSValueProtect
to make clear that if you want one JS object to keep another alive, you shouldn't use it, you should use a private property
otherwise you may get uncollectable reference cycles


Error cases:

JSObjectSetPrivateProperty seems to fail silently if the argument is not a callback object
maybe it should have a boolean return? or raise an exception?
it looks like JSObjectGetPrivateProperty will return garbage when called on a non-API object
I think it would make more sense to return undefined
Silent failure on delete seems ok

Testing:

you could try harder in the test case to avoid the possibility that a copy of aHeapRef remains on the stack
I think you should test calling these on non-API objects in the test case

[Oliver Hunt]

(In reply to comment #2)
> (From update of attachment 51238 [details])
> Comments given on IRC:
> 
> Docs:
> 
> the headerdoc could maybe explain the purpose of this
> i.e. that it's a way to hold on to JavaScript values in a way that is safe for
> GC
> I think probably in the set one would be the best place to explain the purpose
> also maybe add a note to JSValueProtect
> to make clear that if you want one JS object to keep another alive, you
> shouldn't use it, you should use a private property
> otherwise you may get uncollectable reference cycles
I'm not sure about adding references to the new API from actual API until the new API has gone through review.

> 
> 
> Error cases:
> 
> JSObjectSetPrivateProperty seems to fail silently if the argument is not a
> callback object
> maybe it should have a boolean return? or raise an exception?

i'll make it return bool as setPrivate does

> it looks like JSObjectGetPrivateProperty will return garbage when called on a
> non-API object
> I think it would make more sense to return undefined

It returns NULL

> Silent failure on delete seems ok
I'll return bool to be consistent with SetPrivateProperty


> 
> Testing:
> 
> you could try harder in the test case to avoid the possibility that a copy of
> aHeapRef remains on the stack

It does get blown from the stack -- i verified the test failed before i added marking support :D

> I think you should test calling these on non-API objects in the test case

Done.

[Oliver Hunt]

Created attachment 51240 [details]
Patch

[Maciej Stachowiak]

Comment on attachment 51240 [details]
Patch

Please add @result to the headerdoc for JSObjectSetPrivateProperty. Otherwise r=me.

[Oliver Hunt]

Committed r56314: <http://trac.webkit.org/changeset/56314>

[Eric Seidel (no email)]

This appears to have broken the windows build.

[David Kilzer (:ddkilzer)]

(In reply to comment #7)
> This appears to have broken the windows build.

Blind attempt #1 to fix the Windows build (failed):
<http://trac.webkit.org/changeset/56317>

Blind attempt #2 (worked):
<http://trac.webkit.org/changeset/56318>