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

Details of QUnit.begin error are lost #1446

Closed
asakusuma opened this issue Jun 10, 2020 · 4 comments · May be fixed by #1391
Closed

Details of QUnit.begin error are lost #1446

asakusuma opened this issue Jun 10, 2020 · 4 comments · May be fixed by #1391
Assignees
Labels
Component: CLI help welcome Type: Bug Something isn't working right.

Comments

@asakusuma
Copy link

asakusuma commented Jun 10, 2020

  • QUnit version: 2.10.0
  • What environment are you running QUnit in? (e.g., browser, Node): node 12.16.0
  • How are you running QUnit? (e.g., script, testem, Grunt): yarn script
  • Env: Mac OS 10.15.5

If QUnit.begin returns a rejected promise, I would expect the rejection error to be shown to the user, but it is not. You get a Error: Process exited before tests finished running error with no error info.

Reproduction repo: https://github.com/asakusuma/qunit-begin-issue

@trentmwillis trentmwillis added Component: CLI Type: Bug Something isn't working right. labels Jun 10, 2020
@trentmwillis
Copy link
Member

trentmwillis commented Jun 10, 2020

Definitely sounds like a bug. Based on the report, my guess is it has something to do with the handlers in this code:

qunit/src/cli/run.js

Lines 70 to 91 in 474a708

// Listen for unhandled rejections, and call QUnit.onUnhandledRejection
process.on( "unhandledRejection", function( reason ) {
QUnit.onUnhandledRejection( reason );
} );
process.on( "uncaughtException", function( error ) {
QUnit.onError( error );
} );
process.on( "exit", function() {
if ( running ) {
console.error( "Error: Process exited before tests finished running" );
const currentTest = QUnit.config.current;
if ( currentTest && currentTest.semaphore ) {
const name = currentTest.testName;
console.error( "Last test to run (" + name + ") has an async hold. " +
"Ensure all assert.async() callbacks are invoked and Promises resolve. " +
"You should also set a standard timeout via QUnit.config.testTimeout." );
}
}
} );

@smcclure15
Copy link
Member

#1595 has given me some better insight into this kind of exception management. Building on that PR makes this look very solvable - I'll wait for that to land and then iterate on this, given that there's a little overlap with how QUnit.onError might change.

@Krinkle Krinkle changed the title QUnit.begin doesn't show rejection error or stack Details of QUnit.begin error are lost May 8, 2021
@Krinkle
Copy link
Member

Krinkle commented May 8, 2021

I'm able to reproduce this still, and it seems like this is quite similar to #1377 indeed. And the same applies to QUnit.done() as well. We need to re-think how we handle these very early and very late errors.

Pen repro: https://codepen.io/Krinkle/pen/PomYvwN

Krinkle added a commit to Krinkle/qunit that referenced this issue Jun 28, 2021
== Background ==

Previously, QUnit.onError and QUnit.onUnhandledRejection could report
global errors by synthesizing a new test, even after a run has ended.

This is problematic when an errors ocurrs after all modules (and their
hooks) have finished, and the overall test run has ended.

The most immediate problem is that hooks having finished already,
means it is illegal for a new test to start since "after" has already
run. To protect against such illegal calls, the hooks object is
emptied internally, and this new test causes an internal error:

```
TypeError: Cannot read property 'length' of undefined
```

This is not underlying problem though, but rather our internal
safeguard working as intended. The higher-level problem is that there
is no appropiate way to report a late error as a test since the run
has already ended. The `QUnit.done()` callbacks have run, and
the `runEnd` event has been emitted.

== Approach ==

Instead of trying to report (late) errors as a test, only print them
to `console.warn()`, which goes to stderr in Node.js. For the CLI, also
remember that uncaught errors were found and use that to make sure we
don't change exitCode back to zero (e.g. in case we have an uncaught
error after the last test but before our `runEnd` callback is called).

== Changes ==

* Generalise `QUnit.onUnhandledRejection` and re-use it for
  `window.onerror` (browser), and uncaught exceptions (CLI).

* Fix broken use of `QUnit.onError` in `process.on( "uncaughtException" )`.
  This was passing the wrong parameters. Use the new onUncaughtException
  method instead.

* Clarify that `QUnit.onError` is only for `window.onerror`. For now,
  keep its strange non-standard signature as-is (with the custom object
  parameter), but document this and its return value.

* Remove the unused "..args" from `QUnit.onError`. This was only ever
  passed from one of our unit tests to give one extra argument (a
  string of "actual"), which then ended up passed as "actual" parameter
  to `pushFailure()`. We never used this in the actual onError binding,
  so remove this odd variadic construct for now.

* Change `ProcessingQueue#done`, which is in charge of reporting
  the "No tests were run" error, to no longer rely on the way that
  `QUnit.onError` previously queued a late test.

  The first part of this function may run twice (same as before, once
  after an empty test run, and one more time after the synthetic
  test has finished and the queue is empty again). Change this so that
  we no longer assign `finished = true` in that first part. This means
  we will still support queueing of this one late test. But, since the
  quueue is empty, we do need to call `advance()` manually as otherwise
  it'd never get processed.

  Previously, `finished = true` was assigned first, which meant that
  `QUnit.onError` was adding a test under that condition. But this
  worked anyway because `Test#queue` internally had manual advancing
  exactly for this use case, which is also where we now emit a
  deprecation warning (to become an error in QUnit 3). Note that using
  this for anything other than the "No tests run" error was already
  unreliable since generally runEnd would have been emitted already.
  The "No tests run" test was exactly done from the one sweet spot
  where it was (and remains) safe because that threw an error and thus
  prevented runEnd from being emitted.

Fixes qunitjs#1377.
Ref qunitjs#1322.
Ref qunitjs#1446.
Krinkle added a commit to Krinkle/qunit that referenced this issue Jul 3, 2021
== Background ==

Previously, QUnit.onError and QUnit.onUnhandledRejection could report
global errors by synthesizing a new test, even after a run has ended.

This is problematic when an errors ocurrs after all modules (and their
hooks) have finished, and the overall test run has ended.

The most immediate problem is that hooks having finished already,
means it is illegal for a new test to start since "after" has already
run. To protect against such illegal calls, the hooks object is
emptied internally, and this new test causes an internal error:

```
TypeError: Cannot read property 'length' of undefined
```

This is not underlying problem though, but rather our internal
safeguard working as intended. The higher-level problem is that there
is no appropiate way to report a late error as a test since the run
has already ended. The `QUnit.done()` callbacks have run, and
the `runEnd` event has been emitted.

== Approach ==

Instead of trying to report (late) errors as a test, only print them
to `console.warn()`, which goes to stderr in Node.js. For the CLI, also
remember that uncaught errors were found and use that to make sure we
don't change exitCode back to zero (e.g. in case we have an uncaught
error after the last test but before our `runEnd` callback is called).

== Changes ==

* Generalise `QUnit.onUnhandledRejection` and re-use it for
  `window.onerror` (browser), and uncaught exceptions (CLI).

* Fix broken use of `QUnit.onError` in `process.on( "uncaughtException" )`.
  This was passing the wrong parameters. Use the new onUncaughtException
  method instead.

* Clarify that `QUnit.onError` is only for `window.onerror`. For now,
  keep its strange non-standard signature as-is (with the custom object
  parameter), but document this and its return value.

* Remove the unused "..args" from `QUnit.onError`. This was only ever
  passed from one of our unit tests to give one extra argument (a
  string of "actual"), which then ended up passed as "actual" parameter
  to `pushFailure()`. We never used this in the actual onError binding,
  so remove this odd variadic construct for now.

* Change `ProcessingQueue#done`, which is in charge of reporting
  the "No tests were run" error, to no longer rely on the way that
  `QUnit.onError` previously queued a late test.

  The first part of this function may run twice (same as before, once
  after an empty test run, and one more time after the synthetic
  test has finished and the queue is empty again). Change this so that
  we no longer assign `finished = true` in that first part. This means
  we will still support queueing of this one late test. But, since the
  quueue is empty, we do need to call `advance()` manually as otherwise
  it'd never get processed.

  Previously, `finished = true` was assigned first, which meant that
  `QUnit.onError` was adding a test under that condition. But this
  worked anyway because `Test#queue` internally had manual advancing
  exactly for this use case, which is also where we now emit a
  deprecation warning (to become an error in QUnit 3). Note that using
  this for anything other than the "No tests run" error was already
  unreliable since generally runEnd would have been emitted already.
  The "No tests run" test was exactly done from the one sweet spot
  where it was (and remains) safe because that threw an error and thus
  prevented runEnd from being emitted.

Fixes qunitjs#1377.
Ref qunitjs#1322.
Ref qunitjs#1446.
Krinkle added a commit that referenced this issue Jul 3, 2021
== Background ==

Previously, QUnit.onError and QUnit.onUnhandledRejection could report
global errors by synthesizing a new test, even after a run has ended.

This is problematic when an errors ocurrs after all modules (and their
hooks) have finished, and the overall test run has ended.

The most immediate problem is that hooks having finished already,
means it is illegal for a new test to start since "after" has already
run. To protect against such illegal calls, the hooks object is
emptied internally, and this new test causes an internal error:

```
TypeError: Cannot read property 'length' of undefined
```

This is not underlying problem though, but rather our internal
safeguard working as intended. The higher-level problem is that there
is no appropiate way to report a late error as a test since the run
has already ended. The `QUnit.done()` callbacks have run, and
the `runEnd` event has been emitted.

== Approach ==

Instead of trying to report (late) errors as a test, only print them
to `console.warn()`, which goes to stderr in Node.js. For the CLI, also
remember that uncaught errors were found and use that to make sure we
don't change exitCode back to zero (e.g. in case we have an uncaught
error after the last test but before our `runEnd` callback is called).

== Changes ==

* Generalise `QUnit.onUnhandledRejection` and re-use it for
  `window.onerror` (browser), and uncaught exceptions (CLI).

* Fix broken use of `QUnit.onError` in `process.on( "uncaughtException" )`.
  This was passing the wrong parameters. Use the new onUncaughtException
  method instead.

* Clarify that `QUnit.onError` is only for `window.onerror`. For now,
  keep its strange non-standard signature as-is (with the custom object
  parameter), but document this and its return value.

* Remove the unused "..args" from `QUnit.onError`. This was only ever
  passed from one of our unit tests to give one extra argument (a
  string of "actual"), which then ended up passed as "actual" parameter
  to `pushFailure()`. We never used this in the actual onError binding,
  so remove this odd variadic construct for now.

* Change `ProcessingQueue#done`, which is in charge of reporting
  the "No tests were run" error, to no longer rely on the way that
  `QUnit.onError` previously queued a late test.

  The first part of this function may run twice (same as before, once
  after an empty test run, and one more time after the synthetic
  test has finished and the queue is empty again). Change this so that
  we no longer assign `finished = true` in that first part. This means
  we will still support queueing of this one late test. But, since the
  quueue is empty, we do need to call `advance()` manually as otherwise
  it'd never get processed.

  Previously, `finished = true` was assigned first, which meant that
  `QUnit.onError` was adding a test under that condition. But this
  worked anyway because `Test#queue` internally had manual advancing
  exactly for this use case, which is also where we now emit a
  deprecation warning (to become an error in QUnit 3). Note that using
  this for anything other than the "No tests run" error was already
  unreliable since generally runEnd would have been emitted already.
  The "No tests run" test was exactly done from the one sweet spot
  where it was (and remains) safe because that threw an error and thus
  prevented runEnd from being emitted.

Fixes #1377.
Ref #1322.
Ref #1446.
@Krinkle
Copy link
Member

Krinkle commented Jul 4, 2021

Circling back to this after the refactoring in #1629.

I could still reproduce the issue of swalled stacktraces for uncaught errors and rejections from QUnit.begin(), however all this needs is a .catch() handler for the queue advance step of the early hooks.

... and, that's exactly what @step2yeung already proposed with #1391.

@Krinkle Krinkle self-assigned this Jul 4, 2021
Krinkle added a commit that referenced this issue Jul 5, 2021
Capture the status quo before changing it.

Minor changes:

* Switch remaining notEquals/indexOf uses to the preferred
  `assert.true( str.includes() )` idiom.

* Fix duplicate printing of error message due to V8's `Error#stack`,
  as used by onUncaughtException.
  Ref #1629.

* Start normalizing stderror in tests like we do with stdout.

* Account for qunit.js stack frames from native Promise in V8,
  which doesn't include a function name or paranthesis.

Ref #1446.
Ref #1633.
Krinkle added a commit that referenced this issue Jul 5, 2021
Capture the status quo before changing it.

Minor changes:

* Switch remaining notEquals/indexOf uses to the preferred
  `assert.true( str.includes() )` idiom.

* Fix duplicate printing of error message due to V8's `Error#stack`,
  as used by onUncaughtException.
  Ref #1629.

* Start normalizing stderror in tests like we do with stdout.

* Account for qunit.js stack frames from native Promise in V8,
  which doesn't include a function name or paranthesis.

Ref #1446.
Ref #1633.
Krinkle added a commit that referenced this issue Jul 5, 2021
Capture the status quo before changing it.

Minor changes:

* Switch remaining notEquals/indexOf uses to the preferred
  `assert.true( str.includes() )` idiom.

* Fix duplicate printing of error message due to V8's `Error#stack`,
  as used by onUncaughtException.
  Ref #1629.

* Start normalizing stderror in tests like we do with stdout.

* Account for qunit.js stack frames from native Promise in V8,
  which doesn't include a function name or paranthesis.

Ref #1446.
Ref #1633.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Component: CLI help welcome Type: Bug Something isn't working right.
Development

Successfully merging a pull request may close this issue.

4 participants