async* inconsistency in DDC

[sigmundch]

We know we have several inconsistencies so far, here is a case originally reported in the flutter repo: flutter/flutter#140199

import 'dart:async';

Future main() async {
  Stream<int> step1() async* {
    yield 0;
    await Future.delayed(const Duration(milliseconds: 100));
    yield 1;
    throw Exception('error');
  }

  Stream<int> step2() async* {
    await for (final v in step1()) {
      yield v;
    }
  }

  try {
    await for (final v in step2()) {
      print('v: $v');
    }
  } catch (e) {
    print('e: $e');
  }
  print('finish');
}

Output in VM and dart2js:

v: 0
v: 1
e: Exception: error
finish

Output in DDC:

v: 0
v: 1

[nietsmmar]

Is there any progress on this? I find it quite difficult to debug my applications with this wrong behavior in dev build.
Thanks :)

[nietsmmar]

@sigmundch are there any updates on this?

[sigmundch]

Thanks for following up, unfortunately no, we haven't been able to prioritize this 😞

This may not be very helpful or you may already be aware of this, but just in case: some developers find useful to do some additional debugging with the --profile mode. That mode runs the dart2js compiler (same one use for release builds) in a mode that keeps around the original names of your classes and members. That compiler will have the correct async* semantics. However, it can only be used with the default browser devtools and not via the Dart Devtools. Anyways, sometimes it can provide a bit of additional info that it's otherwise hard to get in other ways today.

[nshahan]

@natebiggs
I believe this has been corrected with your change. WDYT?

[biggs0125]

I just verified locally that DDC, with the new async changes, now shows an output aligned with the VM and dart2js. These changes should roll into Flutter soon but likely won't be in a stable release for a few months. Going to close out this issue though as the fix has been landed in the Dart SDK.

[sigmundch]

Just wondering, is it safe to assume the current test coverage also covers this case, or is it worth adding it as a regression test case?