Refactor environment, exception handling and jumping in VM

[HalidOdat]

Depends on #3053

Fixes #2424

This PR refactors environment handling and jumping with break, continues, etc in the VM. Currently the opcodes are very complex, they dynamically search for the environments, for this we also need to keep a lot of state in the CallFrame, this has a performance penalty as well as being hard to implement optimizers (since it has to mirror the state, so it does not change behaviour).

Since this the bytecompiler has full knowledge of where it needs to jump, how many environments to pop, etc. This PR aims to move that opcode logic to bytecompiler, generating opcodes that are "dumber"/simple, hence easier to optimize ( using #3037 ).

This still needs a lot of work!

It changes the following:

• Removes opcodes: FinallyStart, FinallyEnd, Break, Continue, LoopStart, LoopEnd, LabelledStart, LabelledEnd, TryStart, TryEnd, IteratorLoopStart,GeneratorResumeReturn, GeneratorSetReturn, GeneratorJumpOnResumeKind
• Add opcodes: ReThrow, Exception, JumpTable, JumpIfNotResumeKind, IteratorNextWithoutPop, IteratorValueWithoutPop
• Replaces Breaks and Continues with simple unconditional Jump.
• Remove EnvEntry type and env entry stack
• Remove abrupt_completion record from CallFrame
• Remove generator_resume_kind from CallFrame
• Remove yield field from CallFrame, added CompletionType::Yield

TODO:

• Figure out fp on exception handlers
• Fix bytecode flowgraph creation
• Add more documentation.

[github-actions]

Test262 conformance changes

Test result	main count	PR count	difference
Total	95,260	95,260	0
Passed	74,910	74,918	+8
Ignored	19,230	19,230	0
Failed	1,120	1,112	-8
Panics	0	0	0
Conformance	78.64%	78.65%	+0.01%

Fixed tests (8):

test/language/statements/for-of/iterator-close-via-return.js [strict mode] (previously Failed)
test/language/statements/for-of/iterator-close-via-return.js (previously Failed)
test/language/statements/for-of/generator-close-via-return.js [strict mode] (previously Failed)
test/language/statements/for-of/generator-close-via-return.js (previously Failed)
test/language/statements/class/subclass/derived-class-return-override-for-of-arrow.js [strict mode] (previously Failed)
test/language/statements/class/subclass/derived-class-return-override-for-of-arrow.js (previously Failed)
test/language/statements/class/subclass/derived-class-return-override-for-of.js [strict mode] (previously Failed)
test/language/statements/class/subclass/derived-class-return-override-for-of.js (previously Failed)

[codecov]

Codecov Report

Merging #3059 (f88af46) into main (33e8c51) will decrease coverage by 0.09%.
Report is 6 commits behind head on main.
The diff coverage is 72.19%.

@@            Coverage Diff             @@
##             main    #3059      +/-   ##
==========================================
- Coverage   50.50%   50.41%   -0.09%     
==========================================
  Files         443      436       -7     
  Lines       42712    42340     -372     
==========================================
- Hits        21572    21347     -225     
+ Misses      21140    20993     -147     

Files Changed	Coverage Δ
boa_engine/src/bytecompiler/expression/mod.rs	55.88% <0.00%> (-0.84%)	⬇️
boa_engine/src/bytecompiler/statement/labelled.rs	68.42% <ø> (-5.50%)	⬇️
boa_engine/src/bytecompiler/statement/switch.rs	100.00% <ø> (ø)
boa_engine/src/module/source.rs	0.00% <0.00%> (ø)
boa_engine/src/vm/flowgraph/mod.rs	0.00% <0.00%> (ø)
boa_engine/src/vm/opcode/mod.rs	69.23% <ø> (ø)
boa_engine/src/vm/opcode/pop/mod.rs	100.00% <ø> (ø)
boa_engine/src/vm/opcode/push/environment.rs	92.30% <ø> (-0.38%)	⬇️
boa_engine/src/vm/code_block.rs	57.62% <28.84%> (-2.00%)	⬇️
boa_engine/src/bytecompiler/class.rs	15.56% <33.33%> (-0.10%)	⬇️
... and 26 more

... and 26 files with indirect coverage changes

[HalidOdat]

Ran the quickjs benchmarks:

Main

PROGRESS Richards
RESULT Richards 30.1
PROGRESS DeltaBlue
RESULT DeltaBlue 35.5
PROGRESS Encrypt
PROGRESS Decrypt
RESULT Crypto 46.1
PROGRESS RayTrace
RESULT RayTrace 124
PROGRESS Earley
PROGRESS Boyer
RESULT EarleyBoyer 112
ERROR RegExp TypeError: not a constructor
undefined
PROGRESS RegExp
PROGRESS Splay
RESULT Splay 120
PROGRESS NavierStokes
RESULT NavierStokes 11.8
SCORE 51.5
Uncaught Error: Benchmark had 1 errors

PR

PROGRESS Richards
RESULT Richards 31.1
PROGRESS DeltaBlue
RESULT DeltaBlue 36.7
PROGRESS Encrypt
PROGRESS Decrypt
RESULT Crypto 46.9
PROGRESS RayTrace
RESULT RayTrace 126
PROGRESS Earley
PROGRESS Boyer
RESULT EarleyBoyer 114
ERROR RegExp TypeError: not a constructor
undefined
PROGRESS RegExp
PROGRESS Splay
RESULT Splay 127
PROGRESS NavierStokes
RESULT NavierStokes 11.7
SCORE 52.7
Uncaught Error: Benchmark had 1 errors

There is performance increase while greatly simplifying the execution :)

There are still some to-dos left (updated in the PR comment) that I have to do, in the mean time making this ready for review :)

[raskad]

I really like this change. The logic is much simpler than before. Amazing work in this refactor! As soon as all todos are done, I will give it a second review :)

[nekevss]

Hey! This is all looking pretty amazing! Thought I'd provide some early feedback on a couple things.

Sort of an aside: I'm curious what you were thinking about regarding Yield and GeneratorResumeKind given the changes. The one's in this PR are fine, but I keep thinking that they could probably be cleaned up by adding something like an execution context stack to the vm. Maybe that's the wrong approach though.

[HalidOdat]

This PR is now complete! Ready for review/merge :)

[nekevss]

Great work on this! Just one question I wanted to double check 😄

[nekevss]

One more small misspelling that I noticed in docs. 😄 But overall this looks great!

[raskad]

Amazing refactor!