1477063 - Debugger: Leave frame.live true for suspended generator frames

Status: RESOLVED FIXED

(Core :: JavaScript Engine, enhancement, P2)

Description

[Jason Orendorff [:jorendorff]]

All the state needed to recreate an actual stack frame is present in a suspended generator, so it should be possible to query a suspended generator frame like any other frame.

Wait, wait-- I'm getting a signal from ... THE FUTURE

It's me! I'm saying that... querying the suspended frame turns out to be embarrassingly easy, to the point of not even being interesting... while getting the bookkeeping right involves fixing an apparently endless long tail of assertions.

Hmm.

I'm going to try to get someone else to take this bug.

Comment 1

[Jason Orendorff [:jorendorff]]

jimb refuses to take, claiming the resulting time-travel paradox would destroy the universe :-P

Comment 2

[Jason Orendorff [:jorendorff]]

Attachment: Part 1: A few minor cleanups

DebuggerFrame::RESERVED_SLOTS is inaccurate and fortunately unused.

Comment 3

[Jason Orendorff [:jorendorff]]

Attachment: Part 2: Add a slot in async/generator Debugger.Frames that points to the GeneratorObject

For compartment correctness, the reference is via a Debugger.Object whose
referent is the actual GeneratorObject.

Comment 4

[Jason Orendorff [:jorendorff]]

Attachment: Part 3: Clear JSSLOT_DEBUGFRAME_GENERATOR when a Debugger.Frame becomes inactive

The slot won't be accessed again once the Frame is inactive, because all
operations on the Frame throw. So the sole effect of these changes is that
inactive Frame objects no longer keep the generator unnecessarily GC-alive.

Comment 5

[Jason Orendorff [:jorendorff]]

Attachment: Part 4: Document the three possible states of a DebuggerFrame and add accessors

There's also a little bit of code motion in this patch--not much. Still no
observable change in behavior.

Comment 6

[Jason Orendorff [:jorendorff]]

Attachment: Part 5: Move UpdateFrameIterPc up in anticipation of using it earlier in the file

Code motion only.

Comment 7

[Jason Orendorff [:jorendorff]]

Attachment: (WIP) Part 6: Implement DebuggerFramePtr

This is an AbstractFramePtr lookalike that can also handle suspended
generators, answering many of the same questions the debugger might ask of any
on-stack frame.

In this patch the type is implemented but not used.

TODO: See if it could just be a Variant.
TODO: Implement js::GetThisValueForDebuggerMaybeOptimizedOutGenerator.

Comment 8

[Jason Orendorff [:jorendorff]]

Attachment: (WIP) Part 7: Change methods to use DebuggerFramePtr

With this change, suspended Debugger.Frame objects are now `.live`.

Comment 9

[Jason Orendorff [:jorendorff]]

Attachment: Bug 1477063 - Part 1: A few minor cleanups. r?jimb

DebuggerFrame::RESERVED_SLOTS is inaccurate and fortunately unused.

Depends on D6990

Comment 10

[Jason Orendorff [:jorendorff]]

Attachment: Bug 1477063 - Part 2: Add a slot in async/generator Debugger.Frames that points to the GeneratorObject. r?jimb

For compartment correctness, the reference is via a Debugger.Object whose
referent is the actual GeneratorObject.

Depends on D7012

Comment 11

[Jason Orendorff [:jorendorff]]

Attachment: Bug 1477063 - Part 3: Clear JSSLOT_DEBUGFRAME_GENERATOR when a Debugger.Frame becomes inactive. r?jimb

The slot won't be accessed again once the Frame is inactive, because all
operations on the Frame throw. So the sole effect of these changes is that
inactive Frame objects no longer keep the generator unnecessarily GC-alive.

Depends on D7015

Comment 12

[Jason Orendorff [:jorendorff]]

Attachment: Bug 1477063 - Part 4: Document the three possible states of a DebuggerFrame and add accessors. r?jimb

There's also a little bit of code motion in this patch--not much. Still no
observable change in behavior.

Depends on D7016

Comment 13

[Jason Orendorff [:jorendorff]]

Attachment: Bug 1477063 - Part 5: Move UpdateFrameIterPc up in anticipation of using it earlier in the file. r?jimb

Code motion only.

Depends on D7018

Comment 14

[Jason Orendorff [:jorendorff]]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=5c77276d344c407ce94baf1f86cc2bfd90714409

Comment 15

[Jason Orendorff [:jorendorff]]

Attachment: Bug 1477063 - Part 6: Implement DebuggerFramePtr.

This is an AbstractFramePtr lookalike that can also handle suspended
generators, answering many of the same questions the debugger might ask of any
on-stack frame.

In this patch the type is implemented but not used.

TODO: Implement js::GetThisValueForDebuggerMaybeOptimizedOutGenerator.

Depends on D7019

Comment 16

[Jason Orendorff [:jorendorff]]

Attachment: Bug 1477063 - Part 7: Change methods to use DebuggerFramePtr.

With this change, suspended Debugger.Frame objects are now `.live`.

Depends on D10386

Comment 17

[Logan Smyth [:loganfsmyth]]

I did this in https://bugzilla.mozilla.org/show_bug.cgi?id=1592431