New failures (compared to Chrome) when enabling Dart2Wasm testing on Firefox

[mkustermann]

The tests

co19/Language/Functions/element_type_A02_t03 RuntimeError (expected Pass)
co19/Language/Functions/element_type_A02_t04 RuntimeError (expected Pass)
co19/Language/Functions/element_type_A02_t06 RuntimeError (expected Pass)
co19/Language/Libraries_and_Scripts/Scripts/top_level_main_t01 RuntimeError (expected Pass)
co19/Language/Libraries_and_Scripts/Scripts/top_level_main_t06 RuntimeError (expected Pass)
language/async/return_throw_test RuntimeError (expected Pass)
language/regress/regress21795_test RuntimeError (expected Pass)
language/regress/regress23244_test RuntimeError (expected Pass)
language/stack_trace/full1_test RuntimeError (expected Pass)
language/stack_trace/full2_test RuntimeError (expected Pass)
language/stack_trace/rethrow_error_test/none RuntimeError (expected Pass)
language/stack_trace/rethrow_error_test/withtraceparameter RuntimeError (expected Pass)
language/stack_trace/rethrow_nonerror_test RuntimeError (expected Pass)
language/stack_trace/stack_trace_test RuntimeError (expected Pass)

are failing on configurations

dart2wasm-linux-firefox

The typed data tests are seemingly flaky:

co19/LibTest/typed_data/Int8List/isNotEmpty_A01_t02
co19/LibTest/typed_data/Int8List/operator_equality_A01_t01
co19/LibTest/typed_data/Uint16List/Uint16List_A01_t01
co19/LibTest/typed_data/Uint16List/elementSizeInBytes_A01_t02
co19/LibTest/typed_data/Uint16List/operator_equality_A01_t01
co19/LibTest/typed_data/Uint8ClampedList/isEmpty_A01_t02
co19/LibTest/typed_data/Uint8ClampedList/operator_equality_A01_t01
co19/LibTest/typed_data/Uint8List/add_A01_t01
co19/LibTest/typed_data/Uint8List/clear_A01_t01

[mkustermann]

Probably minor things.

/cc @osa1

[osa1]

The ones about the stack traces are because of different toString outputs for JS stack traces.

When throwing an exception we get the stack trace with new Error().stack.toString(), which for the same exception generates this in Firefox:

["_125@http://0.0.0.0:8000/test.mjs:161:29",
 "StackTrace.current@http://0.0.0.0:8000/test.wasm:wasm-function[108]:0x11acb",
 "a@http://0.0.0.0:8000/test.wasm:wasm-function[713]:0x205d4",
 ... ]

and this in Chrome:

['Error',
 '    at _125 (http://0.0.0.0:8000/test.mjs:161:29)',
 '    at StackTrace.current (http://0.0.0.0:8000/test.wasm:wasm-function[108]:0x11acb)',
 '    at a (http://0.0.0.0:8000/test.wasm:wasm-function[713]:0x205d4)',
 ... ]

The first line "Error" doesn't exist in Firefox, so when we drop the first 3 lines to show the stack trace of the throw (instead of the JS code that generates the stack trace) we end up dropping the throws frame.

I'm not sure how to fix yet, but it seems a bit strange that Chrome generates the "Error" in string representation of a stack trace. I wonder if this is something we can change in Chrome.

[osa1]

co19/Language/Functions/element_type_A02_t03 (and probably other element_type tests) are actually failing with Chrome and d8 as well, but I think there's a problem somewhere in the test runner, it doesn't realize that the test is failing. In Chrome, after printing the test failure, in the console there doesn't seem to be uncaught exceptions reported. In Firefox it shows an uncaught exception, which is probably why it's correctly marking the test as failing.

[osa1]

Fix for stack traces: https://dart-review.googlesource.com/c/sdk/+/338000

[osa1]

co19/Language/Functions/element_type_A02_t03 (and probably other element_type tests) are actually failing with Chrome and d8 as well, but I think there's a problem somewhere in the test runner, it doesn't realize that the test is failing.

The problem is that the exception that reports a test failure is sometimes thrown not from main but from a queueMicrotask callback. This happens when main is an async function, and probably in some other cases as well.

Related issue: #52561

[sgrekhov]

See also #54159

[osa1]

After fixes in co19 and dart2wasm stack traces on Firefox, currently these tests from the original list are still failing:

co19/Language/Functions/element_type_A02_t03
co19/Language/Functions/element_type_A02_t04
co19/Language/Functions/element_type_A02_t06
co19/Language/Libraries_and_Scripts/Scripts/top_level_main_t01
co19/Language/Libraries_and_Scripts/Scripts/top_level_main_t06
language/async/return_throw_test
language/regress/regress23244_test

However these tests are now failing with d8 and Chrome as well, so Firefox specific issues seem to be resolved.