1470293 - Long delay on trap in shell builds

Status: RESOLVED INVALID

(Core :: JavaScript Engine, defect, P5)

Description

[Matthew Gaudet (he/him) [:mgaudet]]

When the jsshell hits a trap (either masm.breakpoint(), or a segfault) on OS/X, I'm experiencing a large delay before the process actually exits: 

For example, I've been debugging something, and so I inject masm.breakpoint(). TO make sure I hit it, I run a test case, and and sure enough, masm.breakpoint() hits. 

There's then a ~10-30 second delay before it outputs "Trace/BPT trap: 5" as expected. 

No idea why this is happening. Seems to be OS/X only. Happens in both debug and opt builds. 

But if I just build a boring C program that segafults, I do not see this behaviour. 


Darwin Matthews-MacBook-Pro.local 17.6.0 Darwin Kernel Version 17.6.0: Tue May  8 15:22:16 PDT 2018; root:xnu-4570.61.1~1/RELEASE_X86_64 x86_64

Sampling the process while it's en-route to dying, I've seen a couple of interesting candidates: 

* MachExceptionHandlerThread
* js::MachExceptionHandler

Comment 1

[Nicolas B. Pierron [:nbp] {away}]

Is that a recent regression?

My guess would be that the system might catch it and wait for a debugger to be attached.
Do you see the same slow down while running the same program under a debugger? Does the trap stops the debugger immediately?

Comment 2

[Matthew Gaudet (he/him) [:mgaudet]]

(In reply to Nicolas B. Pierron [:nbp] {backlog: 39} from comment #1)
> Is that a recent regression?

It feels worse in the last couple months, but overall, I'd say no. I'm putting a todo item to build an old revision, and see what happens there.

> My guess would be that the system might catch it and wait for a debugger to
> be attached.
> Do you see the same slow down while running the same program under a
> debugger? Does the trap stops the debugger immediately?

Traps hit the debugger very quickly yes.

Comment 3

[Matthew Gaudet (he/him) [:mgaudet]]

Interesting! 

I think this might be a Visual Studio Code issue! 

I just tried matching the environment inside and outside VS Code, and found the same failures happen instantaneously inside Terminal.app that take 3-5 seconds to resolve inside VSCode! 

I'll report a bug there. Closing this as invalid for now.