⚡️ Deallocate operations and requests for cancellation

[AlexV525]

Resolve #2111

New Pull Request Checklist

• I have read the Documentation
• I have searched for a similar pull request in the project and found none
• I have updated this branch with the latest main branch to avoid conflicts (via merge from master or rebase)
• I have added the required tests to prove the fix/feature I'm adding
• I have updated the documentation (if necessary)
• I have run the tests without failures
• I have updated the CHANGELOG.md in the corresponding package

[AlexV525]

Apparently badssl.com is unstable these days. We might also explore deploying it locally in the future.

[CaiJingLong]

    test('cancels multiple requests', () async {
      final client = MockHttpClient();
      CancelToken? token = CancelToken();
      final tokenRef = WeakReference(token);
      const reason = 'cancel';
      final dio = Dio()
        ..httpClientAdapter = IOHttpClientAdapter(
          createHttpClient: () => client,
        );

      final requests = <MockHttpClientRequest>[];
      when(client.openUrl(any, any)).thenAnswer((_) async {
        final request = MockHttpClientRequest();
        requests.add(request);
        when(request.close()).thenAnswer((_) async {
          final response = MockHttpClientResponse();
          when(response.headers).thenReturn(MockHttpHeaders());
          when(response.statusCode).thenReturn(200);
          await Future.delayed(const Duration(milliseconds: 200));
          return response;
        });
        return request;
      });

      final futures = [
        dio.get('https://pub.dev', cancelToken: token),
        dio.get('https://pub.dev', cancelToken: token),
      ];

      for (final future in futures) {
        expectLater(
          future,
          throwsDioException(
            DioExceptionType.cancel,
            matcher: isA<DioException>().having(
              (e) => e.error,
              'error',
              reason,
            ),
          ),
        );
      }

      await Future.delayed(const Duration(milliseconds: 100));
      token.cancel(reason);

      expect(requests, hasLength(2));

      try {
        await Future.wait(futures);
      } catch (_) {
        // ignore, just waiting here till all futures are completed.
      }

      for (final request in requests) {
        verify(request.abort()).called(1);
      }

      token = null;

      Object? obj = Object();
      final objRef = WeakReference(obj);
      print(tokenRef.target);
      print(objRef.target);

      obj = null;

      List.generate(
        1 * 1024 * 1024,
        (index) => Object(),
      );

      print(tokenRef.target);
      print(objRef.target);

      expect(objRef.target, isNull);
      expect(tokenRef.target, isNull);
    });

I rewrite the method of cancel_token_test.dart.

I got next log:

Instance of 'CancelToken'
Instance of 'Object'
Instance of 'CancelToken'
null
Expected: null
  Actual: <Instance of 'CancelToken'>

package:matcher                    expect
test/cancel_token_test.dart 109:7  main.<fn>.<fn>

So, I think the cancel token isn't be deallocate in GC.

[CaiJingLong]

BTW, maybe 1*1024*1024 is not big enough, you can use bigger.

[AlexV525]

@kuhnroyal I've also configured the Melos script not to produce coverage outputs when running with the minimum SDK so the min test will not fail because of [Sentinel kind: Expired, valueAsString: <expired>] from getIsolateGroup(). That issue has been fixed by Dart 2.19.0.

[kuhnroyal]

Do you have experience with https://pub.dev/packages/leak_tracker?

[AlexV525]

Do you have experience with https://pub.dev/packages/leak_tracker?

I glanced at those packages, which seemed to not fit the current case and require a significant amount of work. Also not sure if they are compatible with older Dart. Spoken with @CaiJingLong and a stable way to test GC (also in Dart's tests) is generating big objects so here we have 1024 * 1024. This is what I've got so far.

[CaiJingLong]

There is a method in dart's testcase:

https://github.com/dart-lang/sdk/blob/59add4f01ef4741e10f64db9c2c8655cfe738ccb/tests/corelib/finalizer_test.dart#L86-L101

This method may be more in line with the rules of dart memory allocation.

void produceGarbage() {
  const approximateWordSize = 4;


  List<dynamic> sink = [];
  for (int i = 0; i < 500; ++i) {
    final filler = i % 2 == 0 ? 1 : sink;
    if (i % 250 == 1) {
      // 2 x 25 MB in old space.
      sink = List.filled(25 * 1024 * 1024 ~/ approximateWordSize, filler);
    } else {
      // 498 x 50 KB in new space
      sink = List.filled(50 * 1024 ~/ approximateWordSize, filler);
    }
  }
  print(sink.hashCode); // Ensure there's real use of the allocation.
}

[AlexV525]

This method may be more in line with the rules of dart memory allocation.

Cool, now we can run that test on the minimal SDK requirement too.

[github-actions]

Code Coverage Report: Only Changed Files listed

Package	Base Coverage	New Coverage	Difference
dio/lib/src/adapters/io_adapter.dart	🟢 88.89%	🟢 88.75%	🔴 -0.14%
dio/lib/src/dio_mixin.dart	🟢 93.86%	🟢 93.99%	🟢 0.13%
plugins/http2_adapter/lib/src/http2_adapter.dart	🟠 71.72%	🟠 71.92%	🟢 0.2%
Overall Coverage	🟢 82.35%	🟢 82.44%	🟢 0.09%

Minimum allowed coverage is 0%, this run produced 82.44%