Bubble up the transitions associated with PreventExtensionsWithTransition

[Closed given last comment by conradw]

Bubble up the transitions associated with PreventExtensionsWithTransition

PreventExtensionsWithTransition adds special transitions which
ReconfigureProperty struggles to handle. If a single one of these is the last
transition, it will now be temporarily removed and replaced once the ordinary
ReconfigureProperty algorithm has run.

This change preempts what would otherwise be a future regression in creating
strong mode class instances.

BUG=
LOG=N

[conradw, 2015-07-17 09:47:25]

conradw@chromium.org changed reviewers:
+ rossberg@chromium.org, verwaest@chromium.org

[conradw, 2015-07-17 09:47:25]

PTAL, will upload tests shortly

[rossberg, 2015-09-02 14:53:01]

rossberg@chromium.org changed reviewers:
+ ishell@chromium.org

[rossberg, 2015-09-02 14:53:52]

Igor, given that Toon is away, can you perhaps have a look at this CL?

[conradw, 2015-09-02 14:57:04]

On 2015/09/02 14:53:52, rossberg wrote:
> Igor, given that Toon is away, can you perhaps have a look at this CL?

I'm not completely happy with this CL just now, I was talking to Jakob about
this issue recently and it seems like there might be a better way to do it (plus
I've still not checked for sure that it is working correctly).

I'd definitely welcome comments though.

[Igor Sheludko, 2015-09-16 13:09:15]

Please also add tests for this functionality.

https://codereview.chromium.org/1239803004/diff/20001/src/objects.cc
File src/objects.cc (right):

https://codereview.chromium.org/1239803004/diff/20001/src/objects.cc#newcode2367
src/objects.cc:2367: if (!prev_object->IsUndefined()) {
How about adding "!old_map->is_extensible() &&" to avoid unnecessary transition
array queries?

https://codereview.chromium.org/1239803004/diff/20001/src/objects.cc#newcode2587
src/objects.cc:2587: }
You should add
  if (*target_map != *real_old_map) {
    real_old_map->NotifyLeafMapLayoutChange();
  }
here.

https://codereview.chromium.org/1239803004/diff/20001/src/objects.cc#newcode2865
src/objects.cc:2865: MaybeHandle<Map> transition_map =
TransitionForPreventExtensions(
It does not make sense to search for transition here because they do not exist.
At this point we do CopyInstallDescriptors at lease once which creates new map
without transitions. So I think you should just call
Map::CopyForPreventExtensions() here instead which will also call necessary
"real_old_map->NotifyLeafMapLayoutChange()".

[conradw, 2015-09-18 08:38:32]

After some iterations on strong class semantics, this change probably doesn't
need to go ahead, since instances are no longer going to be sealed.