Closes #764

If there were no passing tests at all exit with a failing exit code.
This could happen if all tests are skipped, or file is run through
the test runner that does not define any tests. This will reduce the
chances of unintentionally configuring CI such that it doesn't actually
check anything.

Closes #865

Move a `pumpEventQueue` to fix a failing test.

The test was passing because of the extra event loop cycle in
`close()`. The intent of the test is to check behavior when a test
hasn't completed yet. By closing immediately the test was actually
looking at behavior of when the tests hadn't even been declared yet, and
the extra `await` was hiding that.

Reorder the closing of the `_suiteController` and
`_onSuiteAddedController`. Values flow first through the
`_suitController` and then to the `_onSuitAddedController`. I saw at
least one case where a value flowed through late and cause an error
attempting to add a value on a closed controller.

The `resolver` argument was only used from external packages which can
be migrated off this utility.

The `checked` argument was only used to allow `spawnHybrid` APIs to run
code without asserts, but we want all test code to run with asserts,
including server side utilities.

Closes #1015

Our recommendation is that code which wants to `await` the result of an
expectation should use `expectLater`. Update the docs and README to
reflect this.

Closes #1007
Closes #1008

We already forward errors across the `StreamChannel` so it isn't
particularly useful to also have them printed to the output on the VM
hosting the runner.

We know that `listen` is only ever called in an Isolate that was spawned
from the runner and the `zone` is the root zone. The uncaught error
handler on the root zone will print the error and then kill the isolate,
we jump straight to killing the isolate instead. This Isolate is only
ever spawned without an `onError` handler that would override the
default.

Change the behavior of the concurrency setting so that loading and running
  don't have separate pools.
  * The loading and running of a test are now done with the same resource, and
    the concurrency setting uniformly affects each. With `-j1` only a single
    test will ever be loaded at a time.
  * Previously the loading pool was 2x larger than the actual concurrency
    setting which could cause flaky tests due to tests being loaded while
    other tests were running, even with `-j1`.

Partial resolution for https://github.com/dart-lang/test/issues/1159.

Retries loading test suites (in loader.dart). Retry configuration is currently based on the suite metadata - and is counted separately from the test retries (does not count as a retry for tests).

Support coverage collection for the Chrome platform. Coverage information is output to `.chrome.json` in a format suitable for consumption by `package:coverage`.

Closes https://github.com/dart-lang/test/issues/36

The breaking changes were handled in #1125

* Flexible Coverage API

* Differentiate between seed=0 and not set

* Add note to readme about shuffling tests

Usages were removed in https://github.com/dart-lang/test/pull/1119

Since there have since been version constraint bumps it is now safe to
remove entirely.

Closes #1141

Bump min SDK in all packages for safety. Restore use of for-loop
elements in a top level variable since they work in this SDK.

Towards #1141

Work around the bug in the SDK by switching back to `..addAll`. After we
publish this version that works we can bump the min SDK.

Prepare to publish.

Remove the `new` from Dart code in strings, docs, comments, and the
README. Leave the keyword if it's used in a doc comment references to
link specifically to the unnamed constructor instead of a class.

Simplify some implementation comments to avoid repeating information.

In some user facing Dart code, switch to single quotes for consistency.

The field is no longer used by pub and triggers a warning on publish.

- Bump to feature version change in `test`.
- Drop `-dev` from pubspecs.
- Pin to latest `test_core` from `test`.

Add support for writing any reporter output to a file in addition to the
reporter on `stdout`. The `json` reporter is the primary use case.
Adds a `--file-reporter` CLI argument, and `file_reporters` yaml config
option.

- Add a `MultiplexingReporter` to allow wiring up multiple reporters to
  listen at once.
- Plumb through the configuration mapping a reporter to a file path.
- Instantiate each file reporter with a sink from the file instead of
  the normal `stdout`.

Remove TODO comments waiting for things that will never change or that
we don't intend to take action on.

- https://github.com/dart-lang/sdk/issues/22977 is closed as not
  planned.
- Dart 2 does not support importing `dart:io` in cross platform code.
- Setting the exit code is not supported on web.
- We plan on keeping `executable.dart` under lib to allow reuse from
  other packages internally.
- We don't plan on trying to support `spawnHybridUri` outside of the
  test runner.
- `executable.main` is now declared to return a `Future`.
- We don't intend to take a dependency on `json_schema`.

The stack trace changed in `1.24` and the tests were updated to allow
either format. Since we don't support that SDK anymore we can simplify
back to the single format that will be used.

Flutter pins package dependencies, including all transitive dependencies.

Because test_api depends on pedantic, Flutter pins that dependency.

However, pedantic is only depended on for the `unawaited` function, which is a one-line piece of code.

This will enable resolution of https://github.com/flutter/flutter/issues/48246 without requiring Flutter to unpin its dependencies, and with a very small overhead on test_api to not re-import pedantic :)  This helps downstream consumers who want to depend on the package for the actual analysis options files.

Also related: https://github.com/dart-lang/pedantic/issues/46

We never published `1.9.5` so don't need any entry for it. The changes
don't need to be mentioned since they aren't the only thing and they
aren't user facing.

This allows for reusing one template file across all tests in use cases
where external scripts or html elements are required for all tests. The
possibility to still use local html files per test file is retained.

Fixes https://github.com/dart-lang/test/issues/39

Prepare for a potential breaking change in `package:boolean_selector`.
Currently the `evaluate` method takes an argument which is either an
`Iterable<String>` or a `bool Function(String)` which matches the
`.toSet()contains` tearoff on that Iterable. We might tighten that to
always take a `bool Function(String)` since the tearoff is easy to
handle at the calling side.

This has the same intended purpose as `unawaited` so it can be easily
replaced. Technically even the `unawaited` is not necessary here since
none of the surrounding functions are `async`, but it is fine to use it
for documentation purposed.

Leave the utility in place for now since it will take some version bump
gymnastics to roll out.

- always_declare_return_types
- annotate_overrides
- await_only_futures
- omit_local_variable_types
- prefer_collection_literals
- prefer_conditional_assignment
- prefer_final_fields
- prefer_for_elements_to_map_fromiterable
- prefer_if_null_operators
- prefer_spread_collections
- unnecessary_this
- use_function_type_syntax_for_parameters

Bump min SDK to 2.3.0 everywhere to allow using spreads in collection
literals.

Skip an already failing test

This lint may soon be enforced in `package:pedantic`.

If a test times out at the default timeout, point to the docs for
changing timeouts in the error message.

* Extend synthetic test timeout

* Add test-randomize-ordering-seed arg