Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add Vm trace functionality and add trace to boa_wasm #3227

Open
wants to merge 11 commits into
base: main
Choose a base branch
from

Conversation

nekevss
Copy link
Member

@nekevss nekevss commented Aug 19, 2023

This is the branch that was used for the web-debugger concept.

This branch is technically ready to be merged, but thought I'd post it as a draft for now to get feedback as it has some API changes along with quite a few changes to the Vm trace overall.

It changes the following:

  • Search and handle the compiled bytecode trace prior to instruction execution and trace.
  • Creates a BoaJs object and an evaluate_with_debug_hooks function in boa_wasm
  • Creates a VmTrace (not the biggest fan of the name here if anyone has other ideas) for handling tracing and allowing users to define handle trace themselves via "registering" a function (defaults to println).

For reference, the standard trace output would now print like the below.

----------------------Compiled Output: '<main>'-----------------------
Location  Count    Handler    Opcode                     Operands

000000    0000      none      GetName                    0000: 'greet'
000005    0001      none      PushUndefined              
000006    0002      none      Swap                       
000007    0003      none      PushLiteral                0
000012    0004      none      Call                       1
000017    0005      none      SetReturnValue             
000018    0006      none      Return                     

Literals:
    0000: <string> "World"

Bindings:
    0000: greet

Functions:
    0000: name: 'greet' (length: 1)

Handlers:
    <empty>


-----------------------Compiled Output: 'greet'-----------------------
Location  Count    Handler    Opcode                     Operands

000000    0000      none      PutLexicalValue            0000: 'targetName'
000005    0001      none      RestParameterPop           
000006    0002      none      PushLiteral                0
000011    0003      none      GetName                    0000: 'targetName'
000016    0004      none      Add                        
000017    0005      none      PushLiteral                1
000022    0006      none      Add                        
000023    0007      none      SetReturnValue             
000024    0008      none      Return                     
000025    0009      none      Return                     

Literals:
    0000: <string> "Hello, "
    0001: <string> "!"

Bindings:
    0000: targetName

Functions:
    <empty>

Handlers:
    <empty>

------------------------------------ Call Frame -- <main> ------------------------------------
Time          Opcode                     Operands                   Stack

14╬╝s          GetName                    0000: 'greet'              [ [function] ]
0╬╝s           PushUndefined                                         [ undefined, [function] ]
0╬╝s           Swap                                                  [ [function], undefined ]
1╬╝s           PushLiteral                0                          [ "World", [function], undefined ]
------------------------------------ Call Frame -- greet -------------------------------------
0╬╝s           PutLexicalValue            0000: 'targetName'         [ ]
1╬╝s           RestParameterPop                                      [ ]
0╬╝s           PushLiteral                0                          [ "Hello, " ]
1╬╝s           GetName                    0000: 'targetName'         [ "World", "Hello, " ]
5╬╝s           Add                                                   [ "Hello, World" ]
0╬╝s           PushLiteral                1                          [ "!", "Hello, World" ]
1╬╝s           Add                                                   [ "Hello, World!" ]
0╬╝s           SetReturnValue                                        [ ]
0╬╝s           Return                                                [ ]
-------------------------- Call Frame -- <Exiting greet via Return> --------------------------
140╬╝s         Call                       1                          [ "Hello, World!" ]
0╬╝s           SetReturnValue                                        [ ]
0╬╝s           Return                                                [ ]
------------------------- Call Frame -- <Exiting <main> via Return> --------------------------
"Hello, World!"

Let me know what you think 😄

@github-actions
Copy link

github-actions bot commented Aug 19, 2023

Test262 conformance changes

Test result main count PR count difference
Total 50,731 50,731 0
Passed 42,973 42,973 0
Ignored 1,395 1,395 0
Failed 6,363 6,363 0
Panics 18 18 0
Conformance 84.71% 84.71% 0.00%

@jedel1043
Copy link
Member

jedel1043 commented Aug 19, 2023

Seeing this, I think it would be worth to create a Tracer trait instead of passing closures to handle tracing. Would also make it easier to send the output to files, loggers or other interesting usages.

@nekevss
Copy link
Member Author

nekevss commented Aug 19, 2023

Would also make it easier to send the output to files, loggers or other interesting usages.

Kind of what I was going for.

I was initially thinking of how to plug the trace output into a tui out of curiosity which led me to the wasm. I don't think it should be too crazy to switch it over to being trait based. 😄

@codecov
Copy link

codecov bot commented Aug 20, 2023

Codecov Report

Attention: Patch coverage is 10.10638% with 169 lines in your changes are missing coverage. Please review.

Project coverage is 47.21%. Comparing base (6ddc2b4) to head (606f48e).
Report is 141 commits behind head on main.

❗ Current head 606f48e differs from pull request most recent head 94045e1. Consider uploading reports for the commit 94045e1 to get more accurate results

Files Patch % Lines
core/engine/src/vm/trace.rs 13.92% 68 Missing ⚠️
ffi/wasm/src/lib.rs 0.00% 64 Missing ⚠️
cli/src/debug/function.rs 0.00% 15 Missing ⚠️
core/engine/src/vm/code_block.rs 0.00% 10 Missing ⚠️
core/engine/src/vm/mod.rs 57.14% 6 Missing ⚠️
core/engine/src/context/mod.rs 0.00% 4 Missing ⚠️
cli/src/main.rs 0.00% 2 Missing ⚠️
Additional details and impacted files
@@            Coverage Diff             @@
##             main    #3227      +/-   ##
==========================================
- Coverage   47.24%   47.21%   -0.03%     
==========================================
  Files         476      476              
  Lines       46892    46084     -808     
==========================================
- Hits        22154    21760     -394     
+ Misses      24738    24324     -414     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

@nekevss
Copy link
Member Author

nekevss commented Aug 20, 2023

Switched the custom closures bits over to a trait.

It may be possible to go even further with the trait than currently implemented and selectively emit Vm state rather than pre-formatted messages. But I'm not exactly sure how much access we should provided, so I kept the current output as is for now.

@nekevss nekevss marked this pull request as ready for review September 23, 2023 17:04
@nekevss nekevss requested a review from a team September 23, 2023 17:04
@jedel1043 jedel1043 added API waiting-on-review Waiting on reviews from the maintainers labels Nov 29, 2023
@jedel1043
Copy link
Member

@HalidOdat and @jedel1043 will review this to see if we want to expose the instructions or if this is good enough.

Copy link
Member

@jedel1043 jedel1043 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Really nice API! This just needs a rebase, since it's still using the old std::time::Instant. I also have some suggestions and a small question.

boa_engine/src/vm/code_block.rs Outdated Show resolved Hide resolved
boa_engine/src/vm/mod.rs Outdated Show resolved Hide resolved
boa_engine/src/vm/trace.rs Outdated Show resolved Hide resolved
Comment on lines 164 to 176
if self.is_full_trace() {
self.trace_compiled_bytecode(vm, interner);
self.call_frame_header(vm);
} else if self.is_partial_trace() && vm.frame().code_block().traceable() {
if !vm.frame().code_block().frame_traced() {
self.trace_current_bytecode(vm, interner);
vm.frame().code_block().set_frame_traced(true);
}
self.call_frame_header(vm);
self.activate();
} else {
self.call_frame_header(vm);
}
Copy link
Member

@jedel1043 jedel1043 Dec 4, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hmm, I think we can reduce the complexity of this by adding a Trace::should_trace(code: &Gc<CodeBlock>) -> ShouldTrace method with an enum defined by:

enum ShouldTrace {
    No,
    OnlyInstructions,
    Full
}

Tracers can carry a traceable list of Gc<Codeblock> to determine if a code should be traced, and also a first_trace list to determine which codeblocks haven't been traced so that it displays their bytecode first.

Copy link
Member Author

@nekevss nekevss Dec 7, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Still need to work on this portion. The rest is rebased, so it shouldn't take too long 😄

EDIT: Actually I was looking at the above, and I'm not really sure why that was so convoluted. I cleaned up the logic a bit, so it may be better.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Added an enum (went with TraceAction over ShouldTrace), but now some of that logic is loaded into should_trace. Let me know what you think.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Really nice! Though, I think I didn't explain myself good enough. The goal of adding that enum was so that we can move the "should be traced partially or fully" logic from VmTracer to the tracers itself, by adding a Tracer::should_trace(&self, code: Gc<CodeBlock>) -> TraceAction method. Then, implementors would be the ones taking care of choosing when to just print the instructions vs the compiled bytecode.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should we be exposing that much of the Vm internals to Tracer? Part of the current implementation is to only expose preformatted strings in a bit more of a "controlled environment" so to speak, which included handling the logic aspect.

Copy link
Member

@jedel1043 jedel1043 Dec 9, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm pretty sure we already expose CodeBlock publicly, and users cannot do much with it aside from getting its address and name, so I think it should be fine.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm playing around with the update, and it's nicer in some places, but feels a lot more convoluted in others. I'm also not entirely sure on how to calling multiple $boa.function.traceable() might work as calling the function would remove the previous Tracer, which may cause problems on generators and async functions.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'd say that we should try to find an architecture that allows us to always use the same Tracer, mostly for simplicity.

Copy link
Member

@HalidOdat HalidOdat left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sorry for the late review.

Besides jedel1043's comments, this looks good me! :)

Copy link
Member

@HalidOdat HalidOdat left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Did another review, just some small nitpicks that don't block merging. The rest looks good to me! :)

core/engine/src/vm/trace.rs Outdated Show resolved Hide resolved
core/engine/src/vm/code_block.rs Outdated Show resolved Hide resolved
@nekevss
Copy link
Member Author

nekevss commented Jan 20, 2024

Made some changes to have VmTrace have a Vec<Box<dyn Tracer>> and moved should_trace to Tracer. Let me know what you think!

@HalidOdat
Copy link
Member

Looks good to me! Just needs a rebase :)

@@ -76,7 +83,7 @@ pub struct Vm {
pub(crate) realm: Realm,

#[cfg(feature = "trace")]
pub(crate) trace: bool,
pub(crate) trace: VmTrace,
Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A thought that I've had over the last day or so is whether it would be worthwhile redesigning so that the Trace is a trait on Vm.

struct Vm<T: ExecutionTrace=()> {
    ...
    trace: T
}

The only issue with this that I haven't totally thought through is that I'm not sure we'd be able to feature flag the fields (all though we could probably still feature flag the method calls)

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think the current design is fine. What would be the benefit of having it as a trait?

Copy link
Member Author

@nekevss nekevss Apr 19, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Mostly I'm not sure if I'm totally in love with the Vec<Box<dyn Tracer>> on VmTrace. It works and improves the current functionality, but I'm mostly just thinking whether there may be a bit of a better design 😕 That being said, I'm not sure a theoretically better design out weighs the functionality here.

Edit: Using a trait would allow the trace to be defined at compile time essentially.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Imo the Vec<Box<dyn Tracer>> is totally fine for this use case.

Using a trait would allow the trace to be defined at compile time essentially.

I'm not sure if that is a benefit. The upside would be reducing dynamic dispatch, but if you enable tracing, you probably do not care about that last bit of performance.
Being able to change the tracing at runtime also seems like a nice feature right?

Copy link
Member Author

@nekevss nekevss Apr 19, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It is a nice feature! And you're right, if someone enables tracing then they are probably not concerned about performance. But if it's switched to a trait that doesn't mean that there couldn't be an implementation of the trait that can be updated at runtime. And if we were to update this to something without dynamic dispatching, then we even provide an impl of the trait that could be updated at runtime.

The primary difference would be that it would be a user deciding to use dynamic dispatching over Boa locking the user into using it.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I would say the current implementation is good engough. If there are shortcomings we could change it. I would prefer that to adding a generic to the Vm struct. But if you think it is significantly better, go for it :)

Copy link
Member

@jedel1043 jedel1043 May 7, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I do think we could offer "combinators" for tracers. Instead of having Vec<Box<dyn Tracer>>, why not something like Box<dyn Tracer>, where we impl<T: Tracer, U: Tracer> Tracer for (T, U)?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

😕 Maybe.

There's probably a better option with how Tracer is defined too. I was playing around with a new design and have something like:

pub trait Tracer: std::fmt::Debug {
    fn trace_instruction(&self, ctx: &TraceDecisionCtx<'_>) -> InstructionTraceAction;
    fn trace_bytecode(&self, ctx: &TraceDecisionCtx<'_>) -> BytecodeTraceAction;
    fn emit(&self, event: &TraceEvent<'_>);
}

Where TraceEvent is something like:

pub enum TraceEvent<'a> {
    NewCallFrame(&'a NewFrameEvent<'a>),
    ExitCallFrame(&'a ExitFrameEvent<'a>),
    Instruction(&'a InstructionTrace),
    ByteCode(&'a BytecodeTraceEvent),
}

But that design still has VmTrace using a Vec.

Copy link
Member

@raskad raskad left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Just found some leftover comments / typos.

core/engine/src/context/mod.rs Outdated Show resolved Hide resolved
core/engine/src/vm/code_block.rs Outdated Show resolved Hide resolved
@@ -76,7 +83,7 @@ pub struct Vm {
pub(crate) realm: Realm,

#[cfg(feature = "trace")]
pub(crate) trace: bool,
pub(crate) trace: VmTrace,
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think the current design is fine. What would be the benefit of having it as a trait?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
API waiting-on-review Waiting on reviews from the maintainers
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

4 participants