520293 - GC should compress obj->dslots if it gets too fragmented in a dictionary-mode object

Status: RESOLVED WONTFIX

(Core :: JavaScript Engine, defect)

Description

[Andreas Gal :gal]

If we delete anything but the last slot, we don't shrink freeslot. If it is indeed the last slot, we do shrink back freeslot, which requires us to check for SLOT_IN_SCOPE after the getter/setter.

Comment 1

[Andreas Gal :gal]

Attachment: patch

Comment 2

[Andreas Gal :gal]

This patch stops recycling the last slot, which is worse than what we do right now. We should first add a way to GC slots, and then we can drop the instant recycling.

Comment 3

[Brendan Eich [:brendan]]

We don't want to recycle no slots, even if we suck by recycling only the last slot on delete and re-add.

GC'ing slots would not relieve the post-getter/setter-call code from having to validate slot in scope, somehow. Perhaps it could just check for a GC having run by sampling gcNumber, but that is pessimistic.

The current code was written to optimize more by checking more precisely that slot is still in scope *and* if so, that no one deleted any property in the runtime (this is still suboptimal but better given delete's burstiness and rarity).

But whatever happens, a getter or setter could indeed invalidate the slot to be updated with the result of the get or set (setters can change the value they are given, in our API). There's no avoiding a check here just by bloating dslots.

GC'ing slots would be great, can we morph this bug into that.

/be

Comment 4

[Brendan Eich [:brendan]]

(In reply to comment #3)
> GC'ing slots would be great, can we morph this bug into that.

IOW, the summary of this bug is invalid. We need to morph this bug or close it.

/be

Comment 5

[Brendan Eich [:brendan]]

Recycling slots is patched in bug 535629. Re-summarizing this bug as promised in comment 4.

/be

Comment 6

[Nicholas Nethercote [inactive]]

This is so old now, it's clearly not going anywhere.