Error stack in worker missing

[Fenzland]

Currently, every error thrown in worker has no trusted stacks, but only

at WorkerImpl.#poll ($deno$/web/workers.ts:140:17)

[bartlomieju]

@nayeemrmn could you look into this issue? It would require to serialize the error in Rust, forward it to JS (worker host) and then properly reconstruct it

[nayeemrmn]

Based on browser behaviour, the worker should print errors to the terminal on its own. The parent worker is only made aware of the error if the user sets an error event listener on the worker object. I'll see what I can do.

[bartlomieju]

Based on browser behaviour, the worker should print errors to the terminal on its own. The parent worker is only made aware of the error if the user sets an error event listener on the worker object. I'll see what I can do.

@nayeemrmn I think that's what happenning right now - if there's no error listener in the host then error is rethrown in host context. Thanks for looking into this 🙏

[nayeemrmn]

if there's no error listener in the host then error is rethrown in host context.

Okay, I think uncaught errors in Workers should print to the terminal themselves, and as a Deno-specific extension to satisfy our design goal, the main worker should terminate upon polling an unhandled error event like it does now. Since the Worker error would be printed above it, the error thrown by the main worker can stay vague.

error: Uncaught (in worker "worker-name") NotFound: No such file or directory (os error 2)
    at Object.jsonOpSync (core.js:247:13)
    at Object.openSync (deno:cli/rt/30_files.js:30:22)
    at file:///mnt/c/Users/Nayeem/projects/deno/temp-worker.js:3:8
error: Uncaught Error: Unhandled error in worker "worker-name".
    at Worker.#poll (deno:cli/rt/11_workers.js:157:19)

[bartlomieju]

@nayeemrmn I think that's not entirely correct - according to spec if there's no error handler attached to Worker then error should be propagated to the scope the worker was defined in - so send error event to the window

[nayeemrmn]

It seems the event should travel upwards, but I'm saying only the source Worker where the error occurred needs to be responsible for printing the genuine error stack (if it doesn't have a handler itself).

I'm proposing that if the source worker has its own error handler which prevents default, nothing is printed. If it doesn't, it prints the error stack to stdout and propagates an event error upwards. If the event error reaches the main worker without being handled, we would just let it error out in Worker.#poll() like it does now and cause termination.