This is a series of three patches that moves PersistentRooted and Rooteds mechanisms closer together in simple, and hopefully, obviously-correct steps.

Use FOR_EACH_GC_LAYOUT to reduce the code volume.

Move the two blocks of code adjacent so that it's easier to mutate them in parallel and maybe merge them once they have the same set of mechanisms.

Comment on attachment 8632294 [details] [diff] [review] 2_coalesce_persistentrooted-v0.diff Review of attachment 8632294 [details] [diff] [review]: ----------------------------------------------------------------- Uhg, actually it looks like we already have a user of PersistentRooted<JSPropertyDescriptor>, but for some reason it's not used in my local builds, only on Try. I must be skipping something in IPC somehow? In any case, I need to redo this.

Nope, I'm not missing anything. It must be a compiler version issue or something -- the ones on try are all managing to instantiate PermanentRooted<JSPropertyDescriptor> when passing Rooted<JSPropertyDescriptor> to a MutableHandle<JSPropertyDescriptor> argument. I've no idea why they think that's useful, but that's what they're doing. In any case, it's a total artifact, so I've just cut that case out of the static assert and added a runtime assert to cover the hole. Let's see if that helps: https://treeherder.mozilla.org/#/jobs?repo=try&revision=fd71fc94f2c8

Comment on attachment 8632293 [details] [diff] [review] 1_move_persistentrooted_to_rootlists-v0.diff Gah! Now I'm having intermittent trouble with browser shutdown because we appear to finalize contexts way before the runtime. I think the right solution here is to make the context have a RootLists* and store everything in PerThreadData. I'm going to just totally re-do this series.

There was a ton more subtlety here than I really expected, so I've broken this down into separate bugs with smaller patches, largely orthogonal to these.