[vm/ffi] Requesting isolate shutdown in a callback

[dcharkes]

We should check that the FFI does not swallow isolate shutdown requests.

It should probably return the error value to C, let C unwind its stack, and continue unwinding the dart stack upon returning to Dart.

[dcharkes]

@rmacnak-google @liamappelbe based on the test that we added, is anyone working on this? Or did we simply add a test for documentation?

[rmacnak-google]

I'm not working on a fix, I just wanted to added the test while it was fresh in my mind.

[liamappelbe]

I'm not working on it at the moment.

[mkustermann]

Propagating unwind error will need checking on return from non-leaf ffi calls. In some sense this is similar to b113fffb65a002accf53724e5d57110ad215b5f7 which has added checking whether return values are exceptions and need to be propagated.

@dcharkes maybe you can add this issue to your list?

[mkustermann]

The ffi/callback_unwind_error_test flaked again today.

The work to make flutter hot-restart work better will require shutting the entire isolate group down on flutter hot-restart, which will send kill messages to isolates, which in return will require solving this issue.

[eseidel]

Ran into this in our fork, went to fix it, and then tried it in stable dart and realized it fails there too. (Testing on Mac ARM64 dart 3.7.0)