We appear to trace through JSFunction::atom. We only ever initialized it, so we don't need incremental barriers on it, but we do need generation barriers here.
Created attachment 607327 [details] [diff] [review] v0 Does this make sense? It would be less code to just trigger the atom post barrier manually in the two places where it is needed, but this way seems cleaner.
I think this misses the barrier in the js_XDRAtom case. The right way to handle that is similar to what we do with script in that function. Basically, define a local variable |atom|. In the encode case, set it to fun->atom. In the decode case, set it to NULL. Then pass &atom to js_XDRAtom. Later, in the JSXDR_DECODE section, call fun->atom.init(atom). Related to this, I think that the fun->setScript(script) statement can be changed to initScript.
Created attachment 607341 [details] [diff] [review] v1: fixed xdring of the wrapped atom