When an effect is created in a component constructor, it might read signals
which are derived from component inputs. These signals may be unreliable or
(in the case of the proposed input signals) may throw if accessed before the
component is first change detected (which is what makes required inputs
available).
Depending on the scenario involved, the effect may or may not run before
this initialization takes place, which isn't a great developer experience.
In particular, effects created during CD (e.g. via control flow) work fine,
as do effects created in bootstrap thanks to the sync CD it performs. When
an effect is created through dynamic component creation outside of CD though
(such as on router navigations), it runs before the component is first CD'd,
causing the issue.
In fact, in the signal components RFC we described how effects would wait
until ngOnInit for their first execution for exactly this reason, but this
behavior was never implemented as it was thought our effect scheduling
design made it unnecessary. This is true of the regular execution of effects
but the above scenario shows that *creation* of the effect is still
vulnerable. Thus, this logic is needed.
This commit makes effects sensitive to their creation context, by injecting
`ChangeDetectorRef` optionally. An effect created with an injector that's
tied to a component will wait until that component is initialized before
initially being scheduled. TestBed effect flushing is also adjusted to
account for the additional interaction with change detection.
PR Close#52473
By default, `toSignal` transforms an `Observable` into a `Signal`, including
the error channel of the Observable. When an error is received, the signal
begins throwing the error.
`toSignal` is intended to serve the same purpose as the `async` pipe, but
the async pipe has a different behavior with errors: it rejects them
outright, throwing them back into RxJS. Rx then propagates the error into
the browser's uncaught error handling logic. In the case of Angular, the
error is then caught by zone.js and reported via the application's
`ErrorHandler`.
This commit introduces a new option for `toSignal` called `rejectErrors`.
With that flag set, `toSignal` copies the async pipe's behavior, allowing
for easier migrations.
Fixes#51949
PR Close#52474
These special providers are configured when `walkProviderTree` is called. Because of this, they do not maintain any equality between subsequent runs of `walkProviderTree`. This prevents us from being able to compare the provider objects for equality between runs.
This commit changes the behaviour of getInjectorProviders to ignore these providers. In the future we will consider another approach for differentiating these providers from ones provided by users rather than the framework.
PR Close#52458
Recreates the fix for content projection in control flow in the new template pipeline. I also had to make the following adjustments to the pipeline:
1. The `TemplateOp.tag` property was being used to generate the name of the template function, rather than the actual tag name being passed into `ɵɵtemplate`. Since the content projection fix requires the tag name to be passed in, I've introduced a new `functionNameSuffix` property instead.
2. `TemplateOp.block` was being used to determine whether to pass `TemplateOp.tag` into the `ɵɵtemplate` instruction. Now that we're always passing in the tag name after the refactor in point 1, we no longer need this flag.
In addition to the refactors above, I also made some minor cleanups where I saw the opportunity to do so.
PR Close#52414
With the directive-based control flow users were able to conditionally project content using the `*` syntax. E.g. `<div *ngIf="expr" projectMe></div>` will be projected into `<ng-content select="[projectMe]"/>`, because the attributes and tag name from the `div` are copied to the template via the template creation instruction. With `@if` and `@for` that is not the case, because the conditional is placed *around* elements, rather than *on* them. The result is that content projection won't work in the same way if a user converts from `*ngIf` to `@if`.
These changes aim to cover the most common case by doing the same copying when a control flow node has *one and only one* root element or template node.
This approach comes with some caveats:
1. As soon as any other node is added to the root, the copying behavior won't work anymore. A diagnostic will be added to flag cases like this and to explain how to work around it.
2. If `preserveWhitespaces` is enabled, it's very likely that indentation will break this workaround, because it'll include an additional text node as the first child. We can work around it here, but in a discussion it was decided not to, because the user explicitly opted into preserving the whitespace and we would have to drop it from the generated code. The diagnostic mentioned point #1 will flag such cases to users.
Fixes#52277.
PR Close#52414
Previously, we would modified `dep.flags` directly to convert injection flags to booleans. This caused a mutation bug where subsequent calls to `getDependenciesFromInjectable` would result in the flags object containing false for every injection flag.
Now, we stop modifying `dep.flags` directly and instead assign the converted flags to a new object.
PR Close#52450
The current regexp supposes that there is at least one space between the `=` and the aliased variable.
As it is possible to write `let myIndex=index`, this commit updates the regexp to handle such a case.
PR Close#52444
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
Fixes that our regex for parsing time values in defer blocks didn't allow for decimals. This isn't relevant for times in milliseconds, but it can be convenient to write something like `on timer(1.5s)`.
PR Close#52433
The `checkNoChanges` method does not belong in the API of production interface. `checkNoChanges` is
limited to testing and should not be used in any application code. Test
code should use `ComponentFixture` instead of `ChangeDetectorRef`.
Additionally, it is not desirable to have the `checkNoChanges` API
available in a context where `detectChanges` is not run first.
DEPRECATED: `ChangeDetectorRef.checkNoChanges` is deprecated.
Test code should use `ComponentFixture` instead of `ChangeDetectorRef`.
Application code should not call `ChangeDetectorRef.checkNoChanges` directly.
PR Close#52431
Adds some logic to skip over comments when resolving implicit `@defer` block triggers. This currently isn't a problem since we don't capture comments by default, but it may come up if we start capturing comments.
PR Close#52449
`RootViewRef<T>` extends `ViewRef<T>` and overrides 3 methods with behavior
that is identical to `ViewRef<T>`. This commit removes `RootViewRef<T>`
because it is not needed.
PR Close#52430
The `ViewRef<T>` interface extends `InternalViewRef` and is already not
part of the public API. There is no need for the extra `InternalViewRef`
interface. This confusing setup is likely leftover from the types
necessary to support both Ivy and ViewEngine.
PR Close#52430
When using `withFetch`, the `PRIMARY_HTTP_BACKEND` token is set.
The InMemory Backend services will also set that token.
This means that providers order will matter and the latest on the list will be the one instantiated
PR Close#52425
This commit adds a global epoch to the reactive graph, which can optimize
non-live reads.
When a non-live read occurs, a computed must poll its dependencies to check
if they've changed, and this operation is transitive and not cacheable.
Since non-live computeds don't receive dirty notifications, they're forced
to assume potential dirtiness on each and every read.
Using a global epoch, we can add an important optimization: if *no* signals
have been set globally since the last time it polled its dependencies, then
we *can* assume a clean state. This significantly improves performance of
large unwatched graphs when repeatedly reading values.
PR Close#52420
The `for` block has several parts which we know are required. This
commit improves the autocomplete snippet of the `for` block by adding
those required parts and providing placeholders.
PR Close#52405
The previous commits provided the scaffolding for `defer on`. In this commit, we build on that work, adding triggers for `immediate`, `timer`, `hover`, and `viewport`.
PR Close#52387
Previously, we supported a `HasConst` trait, allowing an op to be const collected automatically. However, that approach had the shortcoming that each op could only collect a single constant.
Instead, we now provide a `ConstCollectedExpr`, which collects constants at the expression level, allowing ops to have multiple collectible consts.
Then, we use this new abstraction to support the `defer on` conditions.
PR Close#52387
Previously, we had an "empty shell" implementation of defer conditions, and we used separate ops to represent secondary defer blocks.
Now, we have a real scaffolding for supporting the various defer conditions, and the secondary defer block information has been refactored onto the main defer op.
Additionally, to enable this, we refactor the way that using slot indices works. Instead of having a trait that causes users of slot indices to be linked to the allocated slot, we share a single `SlotHandle` object by reference. This allows an op to use slot information for more than one Xref at a time, and eliminates a layer of indirection.
Co-authored-by: Alex Rickabaugh <alxhub@users.noreply.github.com>
PR Close#52387
This commit updates the reactive template and host binding consumers to
only mark their declaration components for refresh, but not parents/ancestors.
This also updates the `AfterViewChecked` hook to run when a component is
refreshed during change detection but its host is not. It is reasonable
to expect that the `ngAfterViewChecked` lifecycle hook will run when a
signal updates and the component is refreshed. The hooks are typically
run when the host is refreshed so without this change, the update to
not mark ancestors dirty would have caused `ngAfterViewChecked` to not
run.
resolves#14628resolves#22646resolves#34347 - this is not the direct request of the issue but
generally forcing change detection to run is necessary only because a
value was updated that needs to be synced to the DOM. Values that use
signals will mark the component for check automatically so accessing the
`ChangeDetectorRef` of a child is not necessary. The other part of this
request was to avoid the need to "mark all views for checking since
it wouldn't affect anything but itself". This is directly addressed by
this commit - updating a signal that's read in the view's template
will not cause ancestors/"all views" to be refreshed.
PR Close#52302
Currently, the migration always use `$index` in the migrated trackBy function, whereas this variable might be aliased.
The compiler then errors with:
```
error TS2339: Property '$index' does not exist on type 'UsersComponent'.
110 @for (user of users; track byId($index, user); let i = $index) {
```
This commit updates the migration to use the aliased index if there is one.
PR Close#52423
The i18n placeholder resolution phase has accumulated too much logic,
making it difficult to understand. This commit refactors it into several
smaller phases to make it easier to manage.
I suspect this will undergo further refactoring in the near future as I
work through the ICU logic. In particular `ExtractedMessageOp` feels
like a bit of a grab bag of properties, and the i18n const collection
phase is also starting to get quite heavy. This refactor at least feels
like a good start.
PR Close#52390
Enables a handful of i18n tests that are currently skipped, but pass if
enabled. Some of them require alternate golden files because of
inconsequential differences in the cost array order.
PR Close#52390
This commit expands docs extraction for classes and interfaces to include inherited members. This relies on the type checker to get the _resolved_ members of the type so that the extractor doesn't need to reason about inheritance rules, which can get tricky (especially with regards to method overloads).
PR Close#52389
This commit adds decorators to the extracted API docs. It makes some
very hard-coded assumptions about the pattern used to declare decorators
that's extremely specific to what the framework does today.
PR Close#52389
Issue #50320 shows that in some cases, updating a signal that's a dependency
of a template during change detection of that template can have several
adverse effects. This can happen, for example, if the signal is set during
the lifecycle hook of a directive within the same template that reads the
signal.
This can cause a few things to happen:
* Straightforwardly, it can cause `ExpressionChanged` errors.
* Surprisingly, it can cause an assertion within the `ReactiveLViewConsumer`
to fail.
* Very surprisingly, it can cause change detection for an `OnPush` component
to stop working.
The root cause of these later behaviors is subtle, and is ultimately a
desync between the reactive graph and the view tree's notion of "dirty" for
a given view. This will be fixed with further work planned for change
detection to handle such updates directly. Until then, this commit improves
the DX through two changes:
1. The mechanism of "committing" `ReactiveLViewConsumer`s to a view is
changed to use the `consumerOnSignalRead` hook from the reactive graph.
This prevents the situation which required the assertion in the first
place.
2. A `console.warn` warning is added when a view is marked dirty via a
signal while it's still executing.
The warning informs users that they're pushing data against the direction of
change detection, risking `ExpressionChanged` or other issues. It's a
warning and not an error because the check is overly broad and captures
situations where the application would not actually break as a result, such
as if a `computed` marked the template dirty but still returned the same
value.
PR Close#52234
