Normative: When an object dies, all references die at once #122
Conversation
This patch rephrases that logic path into an imperative algorithm, which is described to behave atomically. The PR attempts to resolve the issue documented in tc39#121 (comment) Related issue: tc39#105
|
cc @mhofman |
| 1. Set _ref_.[[Target]] to ~empty~. | ||
| 1. For each FinalizationGroup _fg_ such that _fg_.[[Cells]] contains _cell_, such that _cell_.[[Target]] is _obj_, | ||
| 1. Set _cell_.[[Target]] to ~empty~. | ||
| 1. Optionally, perform ! HostCleanupFinalizationGroup(_fg_). |
There was a problem hiding this comment.
Is this calling https://tc39.github.io/proposal-weakrefs/#sec-cleanup-finalization-group, which means it could throw?
There was a problem hiding this comment.
Nope, it's calling a host hook, which later (in HTML, in a queued task) will call that.
| At any time, if an object _obj_ is not live, an ECMAScript implementation may perform the following steps atomically: | ||
|
|
||
| <emu-alg> | ||
| 1. For each WeakRef _ref_ such that _ref_.[[Target]] is _obj_, |
There was a problem hiding this comment.
Does this automatically apply to all realms in all agents, or should they be explicitly iterated?
There was a problem hiding this comment.
There are no cross-agent references. I don't think we have a concept of an object being "in a Realm", so I'm not sure what it would mean to enumerate them.
|
@tschneidereit and I talked this patch over. We were both impressed by @mhofman 's good catch here--this case had not occurred to either of us. @tschneidereit said that this atomicity would hold in the planned implementation strategy for SpiderMonkey, and supported merging this patch. |
This patch rephrases that logic path into an imperative algorithm,
which is described to behave atomically. The PR attempts to resolve
the issue documented in
#121 (comment)
Related issue: #105