Based on recent discussions, these changes remove the Windows CI check because it has been too flaky for too long. Furthermore, we've concluded that the simulated file system in the compiler tests already catches the same set of bugs as running the tests on a real Windows system.
PR Close#52140
(cherry picked from commit 5a969e06b7)
PR Close#55527
This commit removes the `finish` listener from the Animation object once
the animation is finished, effectively resolving a memory leak. Previously,
the `finish` listener captured `this`, which prevented `this` from being garbage collected.
PR Close#51136
Prior to this change, the transform function would be referenced with a potentially
relative import into an external declaration file. Such imports are not portable
and should not be created in this context. This commit addresses the issue by threading
though the originally used module specifier by means of the `Reference` type.
Fixes#52324
PR Close#52437
This commit fixes an issue where using literal types in the arguments of an input coercion
function could result in emitting invalid output, due to an assumption that TypeScript makes
when emitting literal types. Specifically, it takes the literal's text from its containing
source file, but this breaks when the literal type node has been transplanted into a
different source file. This issue has surfaced in the type-check code generator and is
already being addressed there, so this commit moves the relevant `TypeEmitter` class
from the `typecheck` module to the `translator` module, such that it can be reused for
emitting types in the type translator.
Fixes#51672
PR Close#52437
The `RouterTestingHarness` should throw an error if the call to `navigateByUrl`
expects a component to be activated but the navigation failed.
fixes#52344
PR Close#52357
Previously, because the platform injector does not have a provider container, this API would fail. Now, we account for this case specifically by returning the found providers immediately, without trying to calculate their importpaths.
Also previously, in the case where a boostrapped standalone component did not import any feature modules, the environment injector connected to that bootstrapped component would be the root injector configured by `bootstrapApplication`. This injector is configured through a `providers` array instead of an `imports` array, and also does not have a provider container. Similarly to the platform case, we account this for this by returning the found providers immediately if there is no provider container for our standalone component.
PR Close#52365
Previously this case was missed by the default framework injector profiler. Now in ngDevMode this event emits correctly when a service is configured with `providedIn`. This includes the case where injection tokens are configured with a `providedIn`.
This commit also includes unit tests for this new case in the injector profiler.
PR Close#52365
Non typed forms allow to pass null to nested groups when calling `formGroup.reset()`, this commit prevent an undefined access.
fixes#20509
PR Close#48830
When the `ts.Project` creates the language service plugin (in this case,
the Angular Language Service), it sets the project's language service to
the new language service returned by the plugin create:
https://sourcegraph.com/github.com/microsoft/TypeScript@b12af0fa2bbd4b015e59adcfb49988cea7f919a1/-/blob/src/server/project.ts?L2035-2044
The project may be reloaded in response to various events, such as a
change to the tsconfig file, which then recreates the plugin. When this
happens, the language service that gets passed to the plugin `create`
function will not be the typescript language service, but rather the
previous instance of the new language service returned by the last call
to `create`.
This commit ensures that subsequent calls to `create` for the
`NgLanguageService` plugin for a project after the first call are able
to retrieve and hold on to the _TypeScript_ language service.
fixes https://github.com/angular/vscode-ng-language-service/issues/1923
PR Close#51912
#51885 patched a call site that threw an error but there were 2 others call that needed to be wrapped in the same way by a try/catch.
`initializeFully` is part of the calls in `responseWith(handleFetch)`.
Same #51885, throwing `SwCriticalError`allows the driver to fallback to `safeFetch` and ensure `responseWith` doesn't fail.
Fixes#50378
PR Close#51960
When adding a new view flag, you currently need to adjust the last number of the last
3 flags. All of these share the same number so the shifting ones can just use
the base-10 IndexWithinInitPhaseShift.
PR Close#51839
This PR moves the Observable subscription of toSignal outside of the
reactive context. As the result the toSignal calls are allowed in the
computed, effect and all other reactive consumers.
This is based on the reasoning that we already allow signals creation
in a reactive context. Plus a similar change was done to the async pipe
in the https://github.com/angular/angular/pull/50522Fixes#51027
PR Close#51892
This commit updates hydration runtime code to avoid creating an empty array when we can avoid it. Instead, we just check whether the field is `null` directly (without using nullish coalescing).
PR Close#51917
On Safari, the cache might fail on methods like `match` with an `Internal error`. Critical errors allows to fallback to `safeFetch()` in the `Driver`.
fixes: #50378
PR Close#51885
We control most flags via Starlark and therefore limit configuration
options via `tsconfig` to a minimum. We do not intend to support the
enabled block types option via Starlark, so this commit allows for
the option to be picked up.
(This is useful for benchmarking the new control flow blocks).
PR Close#51862
This commit refactors the loadChildren callback execution of the router config loader
into a function that can be used with SSR for the purposes of route extraction.
PR Close#51818
Currently when ESBuild bundles an application importing from
`@angular/core`, the signals library will be discovered during
export analysis. ESBuild will come across the constants for the reactive
signal graph- and end up considering some of these as side-effects given
the pattern of using a spread assignment for extending from e.g.
`REACTIVE_NODE` (a similar issue may occur if we e.g. extend from the
computed reactive node).
See more details on the issue: https://github.com/evanw/esbuild/issues/3392
Even though, ESBuild preserves these constants now, and all of its
dependencies— Terser will consider these as side-effect free and
eliminate these constants. This may require multiple passes though, and
might not be sufficient, depending on the chain of reactive node
extensions. E.g. in the signals branch we noticed some constants
unnecessarily being preserved.
PR Close#51809
PR Close#51813
Another try at deflaking the tests on Windows. I'm trying a couple of fixes here:
1. I noticed that it's usually the indexer tests that fail during flaky runs. These tests also happen to be the only ones that don't pass in the `files` argument of `NgtscTestEnvironment.setup`. When `files` isn't passed in, we don't hit the file path that sets up the `MockFileSystem`. With these changes I make it so that we always initialize the mock file system.
2. The missing file system error usually comes from the `absoluteFrom` call that initializes the optional `workingDir` argument. My theory is that because it's a default value for an argument, it gets called too early before everything is initialized. These changes move the `absoluteFrom` call further down until it's needed.
PR Close#51804
Terser does not treat `Symbol` as side-effect free- so if we end up with
a symbol export being loaded, it will result in the symbol being
retained.
We noticed this in the signals prototyping where symbols exported
from `computed` ended up appearing in symbol bundling tests.
PR Close#51776
This commit adds explicit type annotations to the reactive node prototype objects,
such that the prototypes are type-checked against the interface they are supposed
to (partially) implement. This also allows IDEs to better track usages of reactive
node properties, improving code navigation.
PR Close#51722
PR Close#51772
When a producer is no longer used, the consumer has to update its internal data structure
that keeps track of all producers. There used to be an issue where only half of the stale
producers would actually be removed from this data structure, as the intended upper bound
of the number of producers to remove would decrease with each removed producer, therefore
not reaching all producers that should be removed from the data structure.
This commit fixes the issue by truncating the arrays directly, without going through
individual `pop` operations. An assertion that would catch the inconsistent state in
the internal data structures of the signal graph has been introduced.
PR Close#51722
PR Close#51772
Improves the error handling story for after*Render by delegating errors to an ErrorHandler, so that one failure does not break every callback.
PR Close#51711
In preparation for adding support for phases to after*Render, which will increase the implementation size, this commit splits out the optional logic so that it can be tree-shaken and dynamically loaded.
PR Close#51711
This flag allows message event listeners to prevent callbacks from executing within the NgZone if they contain a special `__ignore_ng_zone__` flag.
This functionality is built with Angular DevTools in mind, where it prevents an infinite change detection loop in inspecting applications that have message event listeners:
CD -> Inspected app emits componentTreeDirty event to DevTools -> DevTools emits event to get new component Tree from Inspected app -> Inspected app message event listener fires -> CD
PR Close#51339
This commit switches the signals library from a bidirectional symmetric
dependency graph using weak references, to a bidirectional _asymmetric_
graph which uses strong references. This is made possible with a reference
counting algorithm which only tracks producer -> consumer references for
effect-like "live" consumers, preventing memory leaks.
The new algorithm should be simpler and faster than the previous
implementation as weak references are fairly slow to create and traverse.
A tradeoff is that non-live consumers must now poll their producers when
read, as they cannot rely on dirty notifications.
As part of this refactoring, the `ReactiveNode` class is replaced with an
interface instead, and methods are moved to standalone functions. This is
paired with instantiating individual signals/computeds via `Object.create`
against a prototype node which contains static or initial values. This
technique, in conjunction with the rest, greatly improves the performance
of node creation.
PR Close#51226
- Emphasized the importance of using the same InjectionToken instance for both provider and injection call.
- Added examples to illustrate correct usages to prevent NullInjectorError.
PR Close#51386
Kristiyan Kostadinov