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

Promise.reject() crashes repl when using --unhandled-rejections=strict #35232

Closed
dfabulich opened this issue Sep 16, 2020 · 19 comments
Closed

Promise.reject() crashes repl when using --unhandled-rejections=strict #35232

dfabulich opened this issue Sep 16, 2020 · 19 comments
Labels
confirmed-bug Issues with confirmed bugs. repl Issues and PRs related to the REPL subsystem.

Comments

@dfabulich
Copy link
Contributor

  • Version: 14.11.0
  • Platform: macOS Catalina 10.15.6

I have also reproduced this in Node 12.18.4.

What steps will reproduce the bug?

Launch the repl with this command:

node --unhandled-rejections=strict

(--unhandled-rejections=throw has the same issue.)

On the repl command line, type Promise.reject().

What is the expected behavior?

When an uncaught error is thrown in the repl, the repl should print "Uncaught error" without terminating the process. For example:

> throw new Error()
Uncaught Error
> void setTimeout(_=>{throw new Error()}, 0)
undefined
> Uncaught Error

What do you see instead?

The process crashes.

> Promise.reject()
Promise { <rejected> undefined }
> internal/process/promises.js:213
        triggerUncaughtException(err, true /* fromPromise */);
        ^

[UnhandledPromiseRejection: This error originated either by throwing inside of an async function without a catch block, or by rejecting a promise which was not handled with .catch(). The promise rejected with the reason "undefined".] {
  code: 'ERR_UNHANDLED_REJECTION'
}

Additional information

This bug will become more important in Node 15 when --unhandled-rejections=throw becomes the default, per PR #33021. At that point, the bug will repro when launching node using the default settings with no flags.

@BridgeAR BridgeAR added confirmed-bug Issues with confirmed bugs. repl Issues and PRs related to the REPL subsystem. labels Sep 16, 2020
@dfabulich
Copy link
Contributor Author

FWIW, this bug is not a regression. It happens in Node 12 as well.

I'm curious to take a crack at this, but I have no idea where to begin. Can anyone suggest a starting point? (How do repl tests work…?)

@mmarchini
Copy link
Contributor

mmarchini commented Sep 16, 2020

Should repl ignore the flag as use warn /none instead? That's more similar to what happens with sync errors

@BridgeAR
Copy link
Member

Setting it to none would silence such issues in the REPL and the main issue is the programmatic REPL instances. If these are started and we just set the flag to none, rejections in the "regular code" are also silenced. Otherwise it's a good solution (if set to warn).

@dfabulich
Copy link
Contributor Author

That would fix the crash, but it would be a little noisy.

$ node --unhandled-rejections=warn
Welcome to Node.js v14.11.0.
Type ".help" for more information.
> Promise.reject()
Promise { <rejected> undefined }
> (node:66834) UnhandledPromiseRejectionWarning: undefined
(Use `node --trace-warnings ...` to show where the warning was created)
(node:66834) UnhandledPromiseRejectionWarning: Unhandled promise rejection. This error originated either by throwing inside of an async function without a catch block, or by rejecting a promise which was not handled with .catch(). To terminate the node process on unhandled promise rejection, use the CLI flag `--unhandled-rejections=strict` (see https://nodejs.org/api/cli.html#cli_unhandled_rejections_mode). (rejection id: 1)

I feel like I'd want to make the repl set an unhandledRejections hook to log a warning that wasn't so noisy.

Is there a good place I could add code like that?

@devsnek
Copy link
Member

devsnek commented Sep 16, 2020

with throw mode, it should bubble up as an uncaught exception, the same way throwing does.

@mmarchini
Copy link
Contributor

@devsnek yeah I'm a bit surprised with the behavior here honestly. I won't have time to look into it during the week but might have some time during the weekend.

@BridgeAR
Copy link
Member

The reason why rejections behave differently is that we use domains to track uncaught exceptions in the REPL.

@mmarchini
Copy link
Contributor

mmarchini commented Sep 21, 2020

Didn't had time to look into it over the weekend, and I'm not sure if I'll have time to look into it during the week, so if anyone else wants to take a look go ahead. As a workaround we should probably let it to warn, and then keep this issue open to investigate making it work without changing the default.

@Aplet123
Copy link

With the new defaults added in v15, this issue is especially pertinent.

@benjamingr
Copy link
Member

This is quite the blast from the past, I haven't touched domains in several years ^^ I will take a look.

@benjamingr
Copy link
Member

benjamingr commented Nov 11, 2020

So the issue appears to be that process.domain is not set to the domain of the repl inside unhandledRejection so a fix like:

process.on('unhandledRejection', (e, p) => {
    p.catch((e) => {
      if (process.domain === self._domain) {
        self._domain.emit('error', e);
      }
    });
  });

Wouldn't work. I think we'd have to parse the stack trace which sucks but I don't really see an alternative, that wouldn't work with Promise.reject('foo') which has no stack.

Also we should probablyt refactor the REPL to not use domains 🤷

@benjamingr
Copy link
Member

benjamingr commented Nov 11, 2020

Ok so the issue is:

  • When we set the promise to pendingUnhandledRejections its domain is set correctly (i.e. promise.catch(e => console.log(process.domain)) is the REPL domain.
  • However, we defer a microtick - so when we run the microticks and process unhandled rejections the domain context is lost.

Rather than fix domains: I think the REPL needs to set a (separate) setPromiseRejectCallback and catch the promise "earlier" when the domain is still configured rather than using our "unhandled rejection" microtick heuristic.

Edit: Actually I'll attempt a more general fix first

@benjamingr

This comment has been minimized.

@benjamingr
Copy link
Member

Talking to Anna, she suggested emitting the ereor on the domain if unhandledRejection happens inside a domain. I think it's probably the better idea so I'll amend to that.

@benjamingr
Copy link
Member

This should be fixed in the next version.

@BridgeAR BridgeAR unpinned this issue Nov 20, 2020
@ajporrasm
Copy link

ajporrasm commented Feb 3, 2021

is this fixed? I'm still seeing this on windows 10 with node v15.6.0

behavior is the same with and without flag
node --unhandled-rejections=strict index.js
and
node index.js

node:internal/process/promises:208
        triggerUncaughtException(err, true /* fromPromise */);
        ^

Error: read ECONNRESET
    at TLSWrap.onStreamRead (node:internal/stream_base_commons:211:20) {
  errno: -4077,
  code: 'ECONNRESET',
  syscall: 'read'
}

@Aplet123
Copy link

Aplet123 commented Feb 3, 2021

Running a file isn't an issue since it's supposed to error and stop (that's the whole point of the flag). What might be considered an issue (although I'm not sure if this is very important) is that --unhandled-rejections=strict still exits out of the REPL on a promise rejection.

@sgentile
Copy link

This should be fixed in the next version.

I'm getting this in current lts 16.14.0 what would be the 'next version' your referring too ?

@sgentile
Copy link

(i do not get this in v14.17.0)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
confirmed-bug Issues with confirmed bugs. repl Issues and PRs related to the REPL subsystem.
Projects
None yet
Development

Successfully merging a pull request may close this issue.

8 participants