1155796 - Our DependsOn values may not be quite the right set after all

Status: NEW

(Core :: DOM: Core & HTML, defect)

Description

[Boris Zbarsky [:bzbarsky]]

Not sure how important this is... but see bug 1155342.

So what _can_ we flag a [NewObject] method with?  We could use DependsOn=DeviceState if it's totally not dependent on the state of the DOM or anything, but I'd guess that's pretty rare: why would you return a new object each time if the contents of that object don't depend on anything at all about the DOM?

We can't use DependsOn=DOMState, because that would allow LICM and CSE, which is clearly not OK for [NewObject] stuff.

We can do DependsOn=Everything, but it turns out jitinfo has no useful thing it can do with DependsOn=Everything,Affects=Nothing because JitInfo::AliasSet has no useful value for that.  Maybe we should fix _that_.

Or maybe we should add a way to do DependsOn=(DeviceState,DOMState), since the former would just feed into movability while the latter feeds into the alias set.  Or something.

Opinions welcome.

Comment 1

[Boris Zbarsky [:bzbarsky]]

Oh, once we do this, we may want to change what annotations setlike/maplike is using.  Right now it's doing [DependsOn=Everything,Affects=Nothing] for some of its stuff.