It was always the intent to have `afterRender` hooks allow updating
state, as long as those state updates specifically mark views for
(re)check. This commit delivers that behavior. Developers can now use
`afterRender` hooks to perform further updates within the same change
detection cycle rather than needing to defer this work to another round
(i.e. `queueMicrotask(() => <<updateState>>)`). This is an important
change to support migrating from the `queueMicrotask`-style deferred
updates, which are not entirely compatible with zoneless or event `NgZone` run
coalescing.
When using a microtask to update state after a render, it
doesn't work with coalescing because the render may not have happened
yet (coalescing and zoneless use `requestAnimationFrame` while the
default for `NgZone` is effectively a microtask-based change detection
scheduler). Second, when using a `microtask` _during_ change detection to defer
updates to the next cycle, this can cause updates to be split across
multiple frames with run coalescing and with zoneless (again, because of
`requestAnimationFrame`/`macrotask` change detection scheduling).
PR Close#54074
While `Array.at` is technically supported in all browsers we officially
support, the change was needlessly breaking without any real benefit.
PR Close#54021
This commit splits the query implementation and instructions
into a separate files. This is a pattern frequently used by
other functional areas of the framework and is a preparation for
introducing queries-as-signals where we are going to see more
instructions delegating to the same core functionality.
PR Close#53922
This commit moves the implementation of the change detection scheduler
used for testing to angular/core along with a (private export) provider function.
Note: Naming of the provider function is absolutely not final (and not
public API). I would prefer one that did not mention "zones"
but the easiest thing for now is to have a "Zone" and "Zoneless" naming
scheme.
PR Close#53579
This commit ensures that change detection runs when an `LView` is
removed. Change detection is required because DOM nodes aren't actually
removed until the animation engine flushes and this doesn't happen until
the end of `detectChangesInternal` (`rendererFactory.end`).
PR Close#53812
This commit introduces a new enum for capturing additional metadata
about inputs. Called `InputFlags`. These will be built up at compile
time and then propagated into the runtime logic, in a way that does
not require additional lookup dictionaries data structures, or
additional memory allocations for "common inputs" that do not have any flags.
The flags will incorporate information on whether an input is signal
based. This can then be used to avoid megamorphic accesses when such
input is set- as we'd not need to check the input field value. This also
avoids cases where an input signal may be used as initial value for an
input (as we'd not incorrectly detect the input as a signal input then).
The new metadata emit will be useful for incorporating additional
metadata for inputs, such as whether they are required etc (although
required inputs are a build-time only construct right now- but this is a
good illustration of why input flags can be useful). An alternative
could have been to have an additional boolean entry for signal inputs,
but allocating a number with more flexible input flags seems more future
proof and more reasonable andreadable.
More information on the megamorphic access when updating an input
signal
https://docs.google.com/document/d/1FpnFruviKb6BFTQfMAP2AMEqEB0FI7z-3mT_qm7lzX8/edit.
PR Close#53571
Currently when a base class defines an input with a transform, derived
classes re-defining the input via `@Input`, or `inputs: [<..>]`, end up
inherting the transform due to a bug in the inherit definitions feature.
This commit fixes this. We verified in the Google codebase that this is
an unlikely occurrence and it's trivial to fix on user side by removing
the re-declaration/override, or explictly adding the necessary
transform.
Conceptually, the behavior was quite inconsistent as everything else of
inputs was overridden as expected. i.e. alias, required state etc. The
exception were input transforms. This commit fixes this.
PR Close#53571
At this point, we have the following pieces in place:
* the input signature is implemented
* the compiler properly parses and recognizes signal inputs
* the compiler supports type-checking of signal inputs
* input signal metadata is passed to partial output
This commit adds a naive runtime solution to distinguishing between
signal inputs and decorator inputs when the `property` instruction
invokes. This is not ideal and non-performant as we introduce additional
megamorphic reads for every property instruction invocation, or if we'd
use `instanceof`, introducing a hard dependency on `InputSignal` and
risking potentially slower detection.
This code exists purely for testing, to enable playing with input
signals in the playground. In a future commit, we will pass around the
input signal metadata at runtime and can perform highly optimized checks
to distinguish between signal or non-signal inputs- when assigning
values.
More information: https://docs.google.com/document/d/1FpnFruviKb6BFTQfMAP2AMEqEB0FI7z-3mT_qm7lzX8/edit#heading=h.oloxympe902x
PR Close#53571
This commit introduces the runtime `InputSignal` implementation.
Input initializers using `input` or `input.required` will result in
an instance of `InputSignal` to be created.
An input signal extends the signal primtive, with a couple of small
differences:
- it's a readonly signal. There is no public `set` or `update`.
- equality is non-configurable. As per CD semantics, the value is
guaranteed to be different when the `property` instruction attempts
to update an input signal.
- we support a `transform` function, that allows transforming input
values. The transform is called whenever the input is set. An
alternative could have been to follow computed-semantics and call the
transform upon accessing, if dirty.
In the future, we might change this to extend the computed reactive
node, so that we can support computed inputs that do not rely on
continious bound value assignments. See signal based components RFC.
PR Close#53571
In order to provide a reasonable experience for Angular without Zones,
we need a mechanism to run change detection when we receive a change
notification. There are several existing APIs today that serve as the
change notification: `ChangeDetectorRef.markForCheck`, signal updates,
event listeners (since they mark the view dirty), and attaching a view to
either the `ApplicationRef` or `ChangeDetectorRef`. These operations
are now paired with a notification to the change detection scheduler.
The concrete implementation for this scheduler is still being designed.
However, this gives us a starting point to partner with teams to
experiment with what that might look like.
PR Close#53499
This commit updates the `ApplicationRef.isStable` implementation to use
a single `Observable` to manage the state. This simplifies the mental
model quite a bit and removes the need for rx operators like
`distinctUntilChanged` and `combineLatest`.
PR Close#53576
When an application does not use zones, it does not need a default value
for the zone stableness token. This will allow zoneless applications to
tree-shake a lot of rxjs operators out of `ApplicationRef`.
Note that at the moment, `provideZoneChangeDetection` is included in all
applications as well as the `TestBed` environment. It is not currently
possible to remove the zone stable code as a result. This will be
possible only when we make zones an opt-in rather than opt-out.
PR Close#53505
The InitialRenderPendingTasks currently attempts to only contribute to
ApplicationRef stableness one time to support SSR. This isn't actually
how the switchMap works in reality. This commit updates
the isStable observable to be more clear that it's always a combination
of the zone stableness and pending tasks.
In addition, this commit renames the service to just be PendingTasks
because it doesn't directly relate to rendering. While the purpose is
to track things that might cause rendering to happen, we don't know if the
tasks will affect rendering at all.
PR Close#53534
Core bundles were retaining the `Version` class and `VERSION` constant, because we stamp out the current version in the DOM. This shouldn't be necessary, because any usage of `0.0.0-PLACEHOLDER` will be replaced with the current version at build time. These changes remove the reference so it can be tree shaken away.
PR Close#53598
The version of rxjs used to build the repository has been updated to v7.
This required only minimal changes to the code. Most of which were type
related only due to more strict types in v7. The behavior in those cases
was left intact. The most common type related change was to handle the
possibility of `undefined` with `toPromise` which was always possible with
v6 but the types did not reflect the runtime behavior. The one change that
was not type related was to provide a parameter value to the `defaultIfEmpty`
operator. It no longer defaults to a value of `null` if no default is provided.
To provide the same behavior the value of `null` is now passed to the operator.
PR Close#53500
When a view has the `Dirty` flag and is reattached, we should ensure that it is
reached and refreshed during the next change detection run from above.
In addition, when a view is created and attached, we should ensure that it is reached
and refreshed during change detection. This can happen if the view is
created and attached outside a change run or when it is created and
attached after its insertion view was already checked. In both cases, we
should ensure that the view is reached and refreshed during either the
current change detection or the next one (if change detection is not
already running).
We can achieve this by creating all views with the `Dirty` flag set.
However, this does happen to be a breaking change in some scenarios.
The one identified internally was actually depending on change detection
_not_ running immediately because it relied on an input value that was
set using `ngModel`. Because `ngModel` sets its value in a `Promise`, it
is not available until the _next_ change detection cycle. Ensuring
created views run in the current change change detection will result in
different behavior in this case.
Making option the default is the solution to #52928. That will have to
wait for a major version.
PR Close#53022
This commit introduces the initial type-checking for signal inputs.
To enable type-checking od signal inputs, there are a couple of tricks
needed. It's not trivial as it would look like at first glance.
Initial attempts could have been to generate additional statements in
type-checking blocks for signal inputs to simply call a method like
`InputSignal#applyNewValue`. This would seem natural, as it would match
what will happen at runtime, but this would break the language-service
auto completion in a highly subtle way. Consider the case where multiple
directives match the same input. Consider the directives have some
overlap in accepted input values, but they also have distinct diverging
values, like:
```ts
class DirA {
value = input<'apple'|'shared'>();
}
class DirB {
value = input<'orange'|'shared'>();
}
```
In such cases, auto completion for the binding expression should suggest
the following values: `apple`, `shared`, `orange` and `undefined`.
The language service achieves this by getting completions in the
type-check block where the user expression would live. This BREAKS if
we'd have multiple places where the expression from the user is used.
Two different places, or more, surface additional problems with
diagnostic collection. Previously diagnostics would surface the union
type of allowed values, but with multiple places, we'd have to work with
potentially 1+ diagnostics. This is non-ideal.
Another important consideration is test coverage. It might sound
problematic to consider the existing test infrastructure as relevant,
but in practice, we have thousands of diagnostic type check block tests
that would greatly benefit if the general emit structure would still
match conceptually. This is another bonus argument on why changing the
way inputs are applied is probably an option we should consider as a
last resort.
Ultimately, there is a good solution where we unwrap directive signal
inputs, based on metadata, and access a brand type field on the
`InputSignal`. This ensures auto-completion continues to work as is, and
also the structure of type check blocks doesn't change conceptually. In
future commits we also need to handle type-inference for generic signal
inputs.
Note: Another alternative considered, in terms of using metadata or not.
We could have type helpers to unwrap signal inputs using type helpers
like: `T extends InputSignal<any, WriteT> ? WriteT : T`. This would
allow us to drop the input signal metadata dependency, but in reality,
this has a few issues:
- users might have `@Input`'s passing around `InputSignal`'s. This is
unlikely, but shows that the solution would not be fully correct.
- we need the metadata regardless, as we plan on accessing it at runtime
as well, to distinguish between signal inputs and normal inputs when
applying new values. This was not clear when this option was
considered initially.
PR Close#53521
This commit introduces a function for declaring inputs in
components. The function is called `input`. It comes in two flavors:
- `input` for optional inputs with initial values
- `input.required` for required inputs
Inputs are declared as class members, like with `@Input`- except that
the class field will no longer hold the input value directly. Angular
takes control over the input field and exposes the input value as a
signal. The runtime implementation will follow in future commits.
This commit simply introduces:
- initial compiler detection to recognize such inputs in classes
- the initial signature of `input` and `input.required`.
Note: the defer size test is flawed and there is no minification- hence
this commit also needs to incorporate the new dependency graph changes.
PR Close#53521
The behavior of `ApplicationRef.isStable` changed in 16.1 due to
https://github.com/angular/angular/commit/28c68f709cdc930e12bac51a266e7bf790656034.
This change added a `share` to the `isStable` observable, which prevents
additional subscribers from getting a value until a new one emits. One
solution to the problem would be `shareReplay(1)`. However, that would
increase the bundle size since we do not use `shareReplay` elsewhere.
Instead, we don't even really need to share the observable.
The `Observable` available in `ApplicationRef.isStable` before the above commit
was the zone stable observable, without a `share`. The new behavior adds
only an additional observable to the stream, `hasPendingTasks` (a `BehaviorSubject`).
The observables in this stream are not expensive to subscribe to. The
only one with side effects is the `isStable` (because it subscribes to
onStable), but that one already has the `share` operator on it.
Omitting the `share` in `ApplicationRef` also means that applications on `zoneless` will not
have to pay the cost of the operator when we make zones optional because
the zone stable observable is the only place we use it.
PR Close#53541
This separates application and platform code into even more files. This now removes
the ciruclar dependency between scheduling and application ref.
PR Close#53371
Currently, the link to an error guide is only included into an error message in dev mode. This change makes the `Find more at https://angular.io/errors/NG0XYZ` appear in the error message even in prod mode. Note: the rest of the error message is still tree-shaken away in prod mode (as it happens today).
PR Close#53324
Currently the way we extract the pathname of a URL is by creating an anchor node, assigning the URL to its `href` and reading the `pathname`. This is inefficient and it triggers an internal security check that doesn't allow the `href` attribute to be set which ends up blocking https://github.com/angular/components/pull/28155.
These changes switch to using the browser's built-in URL parsing instead.
PR Close#53097
Related to #52928 but `updateAncestorTraversalFlagsOnAttach` is called
on view insertion and _should_ have made that work for views dirty from
signals but it wasn't updated to read the `dirty` flag when we changed
it from sharing the `RefreshView` flag.
For #52928, we've traditionally worked under the assumption that this is working
as expected. The created view is `CheckAlways`. There is a question of whether we
should automatically mark things for check when the attached view has
the `Dirty` flag and/or has the `FirstLViewPass` flag set (or other
flags that indicate it definitely needs to be prefreshed).
PR Close#53001
`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
While `performance.mark` is available on all supported browsers and node.js version this API is not available in JSDOM which is used by Jest and Cloudflare worker.
This commit, updates the usage to a safer variant.
PR Close#52505
Producers represent values which can deliver change notifications.
When a producer value is changed, a change notification is propagated through the graph,
notifying live consumers which depend on the producer of the potential update.
Note here that this is a _potential_ update.
A producer may not have actually "changed" based on its equality function. With
this commit, before refreshing a view that is only marked for refresh
because its consumer is dirty, we poll producers for change to see if
they really have. If not, we can skip the refresh. The example test in this commit
shows that a `computed` which depends on a `signal` that is updated but
produces a value that is the same as before will _not_ cause the
component's template to refresh.
fixes#51797
PR Close#52476
`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
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
Angular recently gained a local compilation mode (see commit
345dd6d81a). This is intended to be used
with the TypeScript compiler option isolatedModules, which bans imports
of const enums.
This changes all const enums tagged with @publicApi to regular enums.
Fixes#46240
PR Close#51670
Mateusz Daniluk