This CL builds upon https://crrev.com/c/3284887 (and partly reverts it).
Class literals are a bit iffy when it comes to source position and
debugging. Mainly the debugger assumes the following invariant:
When we are paused inside a class scope, then we expect the class's
BlockContext to be pushed already. On the other hand, when we are
paused outside a class scope in a function, we don't expect to find
the class's BlockContext.
The problem is that there are cases where we can either pause
"inside" or "outside" the class scope. E.g.:
* `var x = class {};` will break on `class` which is inside
the class scope, so we expect the BlockContext to be pushed
* `new class x {};` will break on `new` which is outside the
class scope, so we expect the BlockContext to not be pushed
yet.
The issue with the fix in https://crrev.com/c/3284887 is that it
adjusted the break position for the bytecode of class literals to
ALWAYS be after the BlockContext is pushed. This breaks the
second example above. We need to tighten the fix a bit and only
defer the break position if the "current source position" is
inside the class's scope. This way we always guarantee that the
BlockContext is pushed or not, depending if the source position
that corresponds to the break position is inside or outside the
class's scope.
Note 1: The CL updates a lot of the bytecode expectations. This
is because the class literals are often the first statement in
the snippet so we don't need to defer the break position.
Note 2: We add a mirrored debugger test to the inspector test so
the fuzzer can have some more fun.
Fixed: chromim:1350842
Change-Id: I9b5a409f77be80db674217a685a3fc9f8a0a71cf
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3827871
Reviewed-by: Shu-yu Guo <syg@chromium.org>
Reviewed-by: Kim-Anh Tran <kimanh@chromium.org>
Commit-Queue: Simon Zünd <szuend@chromium.org>
Reviewed-by: Marja Hölttä <marja@chromium.org>
Cr-Commit-Position: refs/heads/main@{#82473}
This CL removes the bailout when trying to collect the scope info
for the class member initializer function. While this might not have
worked previously, now we only need to tweak the scope search
slightly to fix this. Class member initializer functions never
have their own context but instead us the class context. That means
that most of the logic in debug-scopes.cc doesn't really matter and we
only need to initialize the ScopeIterator properly with the class
context and the member initializer JSFunction.
Note that this still does not fully fix bug 1350842. That is because
we still run into a DCHECk when paused at a `new class { ... }`
statement. We'll fix that in a separate CL.
R=bmeurer@chromium.org
Bug: chromium:1350842
Change-Id: Id128b10676a5aa8a77309735e755e485f2c14446
Fixed: chromium:1246889
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/3825881
Commit-Queue: Simon Zünd <szuend@chromium.org>
Reviewed-by: Benedikt Meurer <bmeurer@chromium.org>
Cr-Commit-Position: refs/heads/main@{#82387}
