We currently face a lot of flakiness with our
Saucelabs CI jobs. These randomly exceed the 10min
CircleCI no-output limit because something throws
off `sauce-connect` in a long-lasting loop where
it tries to connect to some of their Saucelabs
servers. The initial assumption from the Saucelabs
team was that we might have some invalid firewall
rules, but this does not answer why this happens
_randomly_, so the latest update from the support
is that there have been some changes in the latest
version of `sauce-connect` version that **could**
cause this flakiness.
I've manually did multiple test runs and was only
able to reproduce the issues with v4.5.3 (latest
version), so it might be worth downgrading to
v4.5.1. This is also what the Saucelabs support
proposed us to do (though it's not guaranteed that
v4.5.1 is unaffected by the same issue)
PR Close#28659
Currently external static symbols which are referenced by AOT
compiler generated code, will be re-exported in the corresponding
`.ngfactory` files.
This way of handling the symbol resolution has been introduced in
favor of avoding dynamically generated module dependencies. This
behavior therefore avoids any strict dependency failures.
Read more about a particular scenario here: https://github.com/angular/angular/issues/25644#issuecomment-458354439
Now with `ngtsc`, this behavior has changed since `ngtsc` just
introduces these module dependencies in order to properly reference
the external symbol from its original location (also eliminating the need
for factories). Similarly we should provide a way to use the same
behavior with `ngc` because the downside of using the re-exported symbol
resolution is that user-code transformations (e.g. the `ngInjectableDef`
metadata which is added to the user source code), can resolve external
symbols to previous factory symbol re-exports. This is a critical issue
because it means that the actual JIT code references factory files in order
to access external symbols. This means that the generated output cannot
shipped to NPM without shipping the referenced factory files.
A specific example has been reported here: https://github.com/angular/angular/issues/25644#issue-353554070
PR Close#28594
The "tree/polymer_leaves" benchmark has no benchmark tests, nor do we install
Polymer anywhere. Polymer was previously installed through `bower`, but since
we removed bower, there is no easy way to run this benchmark with Polymer.
Considering that there are no benchmark tests, nor we have an easy way to
install/vendor Polymer, we should just remove this benchmark app.
This is also based on the assumption that we want to remove the
Polymer benchmarks anyway: see: 8a05199fb9be5ef91ec8094a3c6abc0a65eb6bcd
PR Close#28568
Currently it's not possible to run the benchmark e2e tests with Bazel on Windows
because the logic that imports the `perf_util`'s is not using a proper Bazel script
manifest path. A script manifest path should not start with `./angular/`, but rather
with the workspace name (which is in our case `angular/`)
PR Close#28568
Currently we install `firebase-tools` manually in the
integration tests run script. This is problematic
because it means that we cannot cache `firebase-tools`
properly and Yarn might time out downloading this
dependency. We can safely move this to the top level
`package.json` since Bazel now has a `.bazelignore` and
since we have a cache that works for PRs (with fallback
caching).
Note that the `.bazelignore` is relevant here because
`firebase-tools` has been mainly moved to the bash
script because it broke some Bazel calls.
See 4f0cae067656fa4563417542297c67030d911a36.
PR Close#28615
It seems that in some cases (especially on CI), global state is not
cleaned up properly causing a specific test to fail.
See #28045 and #28181 for more context.
This PR restores the global state for the affected test. This partly
defeat the purpose of the test, but is better than having flakes on CI.
Fixes#28614
PR Close#28617
Somehow the current list ommits quite a few important targets. Especially the cla and google3.
This changes adds all the statuses that must always be present and green for this agregate state expised as "ci/angular: merge status" to be green.
PR Close#28613
FileType objects are deprecated. They are not required for specifying valid file types for rule attributes, a list of strings can be used instead.
PR Close#28583
When a UrlTree of root url was returned by a guard as a redirection, the
navigation was not processed. The issue came from the error handler which
incorrectly marked the router as already navigated.
Fixes#27845
PR Close#28271
Previously, `ng-packages-installer` would replace the version ranges for
all dependencies that were peer dependencies of an Angular package with
the version range used in the Angular package. This effectively meant
that the pinned version (from `yarn.lock`) for that dependency was
ignored (even if the pinned version satisfied the new version range).
This commit reduces non-determinism in CI jobs using the locally built
Angular packages by always using pinned versions of dependencies for
Angular package peer dependencies if possible.
For example, assuming the following versions for the RxJS dependency:
- **aio/package.json**: `rxjs: ^6.3.0`
- **aio/yarn.lock**: `rxjs@^6.3.0: 6.3.3`
- **@angular/core#peerDependencies**: `rxjs: ^6.0.0`
...the following versions would be used with `ng-packages-installer`:
- Before this commit:
- **aio/package.json**: `rxjs: ^6.0.0`
- **node_modules/rxjs/**: `6.4.0` (latest version satisfying `^6.0.0`)
- After this commit:
- **aio/package.json**: `rxjs: ^6.3.0`
- **node_modules/rxjs/**: `6.3.3` (because it satisfies `^6.0.0`)
PR Close#28510
`ng-packages-installer` can be used to replace Angular packages with
locally built ones (from `dist/packages-dist/`) along with their peer
dependencies.
Previously, in order to achieve this, `yarn install` was called with the
`--no-lockfile` option, which resulted in installing the latest versions
of all dependencies (including transitive ones) permitted by the
corresponding version ranges in `package.json` files. As a result, newly
released versions would be picked, resulting in unexpected,
non-deterministic breakages in CI.
This commit calls `yarn install` with the `--pure-lockfile` option
instead. As a result, only the Angular packages (for which the locally
built ones are used) and their peer dependencies are unpinned; the
pinned versions from `yarn.lock` are used for all other (direct and
transitive) dependencies.
While this does not eliminate non-determinism across builds, it
significantly reduces it.
PR Close#28510
Since b43f8bc7d, RxJS does not need to be patched any more in the
top-level `node_modules/`, so we don't need to special-case RxJS in
`ng-package-installer` and use `node_modules/rxjs/`.
PR Close#28510
The docs don't mention that the app will never be stable if a `setInterval` is running somewhere, and that it will prevent the servcie worker to be registered too.
PR Close#28102
Note that this allows Angular to depend on the entirety of the ES2015 API, not just our restricted subset.
This change is needed because our copy of the subset was out-of-date, and prevents us using ES2015 target in dev mode.
This is a subset of #27738
PR Close#28570
Previously, in order for the `aio_monitoring` failure notifications to
work, the steps up to `define_env_vars` should have succeeded. This
meant that any failures in earlier steps would not send notifications.
This commit fixes it by making the notification step independent of the
`define_env_vars` step.
PR Close#28555
In most cases, it doesn't make a difference, because the cache does
already exist and is not saved. In the few cases where the dependencies
change (and the cache needs to be updated), it makes more sense to save
the cache in the `build-npm-packages` job, because most jobs depend on
it and thus will be able to take advantage of the updated cache right
away.
This seems to be an oversight in b26ac1c22.
PR Close#28555
Currently our version of Yarn is installed through
the "circleci/node" docker image. This is problematic
because in order to be able to update Yarn, we always
need to update the docker image to a version that
comes with the desired Yarn version. Sometimes there
is no docker image with the desired latest Yarn version,
and therefore we cannot easily update the Yarn version.
Additionally updating the docker image also means that
we need to update our version of NodeJS, as well as the
version of `openssl` might have changed (meaning that
our encrypted credential files may not be decodable with
the new version of `openssl`)
PR Close#28546
Note that this fixes `compiler-cli` tests within `compiler-cli/test`,
but there seem to be remaining `ngcc` tests within `compiler-cli/src`
which aren't working on Windows. This is out-of-scope for this commit.
PR Close#28550
Currently the "ngtsc` testing helpers resolve the `fake_core` NPM
package using the `TEST_SRCDIR` variable. This is problematic on Windows
where Bazel runfiles are not symlinked into the runfiles directory.
In order to properly resolve the NPM Bazel tree artifact, we use the
`resolveTreeNpmArtifact` runfile helper that properly resolves the artifact
properly on all platforms.
PR Close#28550
In order to support running "compiler-cli" tests that use the "test_support.ts"
utilities on Windows with Bazel, we need to imporve the logic that resolves NPM
packages and symlinks them into a temporary directory.
A more Bazel idiomatic and windows compatible way of resolving Bazel runfiles
is to use the "RUNFILES_MANIFEST" if present. This ensures that the NPM
packages can be also symlinked on Windows, and tests can execute properly
on Windows. Read more about why this is needed here:
* https://github.com/bazelbuild/bazel/issues/3726#issue-257062364
PR Close#28550
Since we recently removed the `test.sh` script, and now run
all tests with Bazel, we can remove the unused logic that makes
compiler-cli tests pass in non-Bazel.
This cleans up the tests, and also makes it easier to write tests
without worrying about two ways of the Angular package output
(Bazel `ng_package` rules vs. old `build.sh` logic of building)
PR Close#28550
Since we recently removed the `test.sh` script, and now run all
tests with Bazel, we can remove the unused logic that makes language-service
tests pass in non-Bazel.
This cleans up the tests, and also makes it easier to write tests
without worrying about two ways of the Angular package output
(Bazel `ng_package` rules vs. old `build.sh` logic of building)
PR Close#28550
There is no debouncing when we store the scroll Position.
Currently, we have a message in the console after a while:
>Throttling history state changes to prevent the browser from hanging
see: https://bugs.chromium.org/p/chromium/issues/detail?id=786211 for more informations
PR Close#28368
