Implements more of the runtime validations for host directives as compiler diagnostics so that they can be caught earlier. Also does some minor cleanup.
PR Close#47768
In AOT compilations, the `strictInjectionParameters` compiler option can
be enabled to report errors when an `@Injectable` annotated class has a
constructor with parameters that do not provide an injection token, e.g.
only a primitive type or interface.
Since Ivy it's become required that any class with Angular behavior
(e.g. the `ngOnDestroy` lifecycle hook) is decorated using an Angular
decorator, which meant that `@Injectable()` may need to have been added
to abstract base classes. Doing so would then report an error if
`strictInjectionParameters` is enabled, if the abstract class has an
incompatible constructor for DI purposes. This may be fine though, as
a subclass may call the constructor explicitly without relying on
Angular's DI mechanism.
Therefore, this commit excludes abstract classes from the
`strictInjectionParameters` check. This avoids an error from being
reported at compile time. If the constructor ends up being used by
Angular's DI system at runtime, then the factory function of the
abstract class will throw an error by means of the `ɵɵinvalidFactory`
instruction.
In addition to the runtime error, this commit also analyzes the inheritance
chain of an injectable without a constructor to verify that their inherited
constructor is valid.
BREAKING CHANGE: Invalid constructors for DI may now report compilation errors
When a class inherits its constructor from a base class, the compiler may now
report an error when that constructor cannot be used for DI purposes. This may
either be because the base class is missing an Angular decorator such as
`@Injectable()` or `@Directive()`, or because the constructor contains parameters
which do not have an associated token (such as primitive types like `string`).
These situations used to behave unexpectedly at runtime, where the class may be
constructed without any of its constructor parameters, so this is now reported
as an error during compilation.
Any new errors that may be reported because of this change can be resolved either
by decorating the base class from which the constructor is inherited, or by adding
an explicit constructor to the class for which the error is reported.
Closes#37914
PR Close#44615
`getPotentialImportsFor` returns an array of possible imports, including TypeScript module specifier and identifier name, for a requested trait in the context of a given component.
PR Close#47631
`getPotentialTemplateDirectives` returns possible directives which can be used in the provided context, whether already in scope or requiring an import.
This is necessary to implement auto-import support for standalone components in the language service.
PR Close#47561
Using raw objects as a lookup structure will inadvertently find methods defined on
`Object`, where strings are expected. This causes errors downstream when string
operations are applied on functions.
This commit switches over to use `Map`s in the DOM element schema registry to fix
this category of issues.
Fixes#46936
PR Close#47220
This option has no longer any effect as Ivy is the only rendering engine.
BREAKING CHANGE: Angular compiler option `enableIvy` has been removed as Ivy is the only rendering engine.
PR Close#47346
This is the compile-time implementation of the `hostDirectives` feature plus a little bit of runtime code to illustrate how the newly-generated code will plug into the runtime. It works by creating a call to the new `ɵɵHostDirectivesFeature` feature whenever a directive has a `hostDirectives` field. Afterwards `ɵɵHostDirectivesFeature` will patch a new function onto the directive definition that will be invoked during directive matching.
For example, if we take the following definition:
```ts
@Directive({
hostDirectives: [HostA, {directive: HostB, inputs: ['input: alias']}]
})
class MyDir {}
```
Will compile to:
```js
MyDir.ɵdir = ɵɵdefineComponent({
features: [ɵɵHostDirectivesFeature([HostA, {
directive: HostB,
inputs: {
input: "alias"
}
}])]
});
```
The template type checking is implemented during directive matching by adding the host directives applied on the host to the array of matched directives whenever the host is matched in a template.
Relates to #8785.
PR Close#46868
This helper accepts a class for an Angular trait, and returns the NgModule which owns that trait. This will be useful for the language service import project, which needs to edit import arrays on the module.
PR Close#47166
This helper accepts a class, and returns the primary Angular Decorator associated with that trait (e.g. the Component, Pipe, Directive, or NgModule decorator). This will be useful for the language service import project, which needs to edit import arrays inside the decorator.
PR Close#47180
improve the error message for non-standalone components which are not
exported from their module, and that are also imported directly as if
they were standalone
this change simply adds the suggestion to the developer to import the
ngModule instead
resolves#46004
PR Close#46114
The source-map package now requires the
`SourceMapConsumer`/`SourceMapGenerator` classes to be instantiated
asynchronously. This commit updates our tests to account for that.
PR Close#46888
When generating .d.ts metadata for NgModules, by default we emit type
references to their declarations, imports, and exports. However, this
information is not necessarily useful to consumers. References to private
directives (those that aren't exported by the NgModule) for example aren't
at all useful as they can only affect other components declared in the
NgModule. References to imports are of limited usefulness - they might be
helpful for an IDE to understand the DI structure of an application, but
aren't at all used by a downstream compiler.
Generating this metadata is not without cost. When an incremental build
system uses changes in inputs to determine when a rebuild is necessary, any
changes in .d.ts files might cause downstream targets to rebuild. If those
.d.ts changes are in the "private" side of the NgModule (imports or non-
exported directives/pipes), then these rebuilds are wholly unnecessary.
This commit introduces the `onlyPublishPublicTypingsForNgModules` flag for
the compiler. When this flag is set, the compiler will filter the emitted
references in NgModule .d.ts output and only reference those directives/
pipes that are exported from the NgModule (its public API surface). Omitting
the flag preserves the existing behavior of emitting all references, both
public and private.
This is especially useful for build systems such as Bazel.
PR Close#45894
This commit updates the error message to use correct info depending on whether a component is standalone or not. Previously we were always referring to @NgModules as a place to fix the issue, but not we also mention @Component when needed (for standalone components).
PR Close#46159
Angular generally supports cycles between components in the same NgModule.
We have a mechanism of moving the component scope declaration into the
NgModule file in this case. This ensures that Angular never itself
introduces an import which creates a cycle.
What happens if the cycle already exists in the user's program, though, is a
bit different. In these cases, the "correct" emit for Angular is to generate
the component scope (whether direct or remote) inside of a closure, to
prevent evaluating the scope's references until module evaluation is
complete and all cyclic imports have been resolved. We don't want to do this
for *all* scopes because the code size cost of emitting a function wrapper
is non-zero.
In this fix, we take the presence of a `forwardRef` in a component's
`imports` or in an NgModule `declarations` or `imports` as a sign that
component scopes emitted into those files need to be protected against
cyclic references. In a future commit, we may introduce a warning or error
if cyclic imports are not protected behind `forwardRef` in these cases, but
this will take some time to implement.
PR Close#46139
This commit improves the reported error when importing e.g. `RouterModule.forRoot()`
from within `Component.imports`. Such import is not supported, as standalone components
can only refer to other standalone entities or NgModules in their `imports` array;
`ModuleWithProviders` are not supported as `Component.imports` is meant to be used
for the compilation scope of the component, _not_ for configuring DI.
Closes#46003
PR Close#46009
update the error message presented during aot compilation when an unrecognized
tag/element is found in a standalone component so that it does not mention
the ngModule anymore
Note: the jit variant is present in PR #45920resolves#45818
PR Close#45919
This commit fixes a small issue in the logic around the calculation of
template scopes for standalone components. These scopes include a
`Reference` for each dependency of a standalone component, which is used to
generate references to that dependency in various contexts.
Previously, the `Reference` used for a dependency was the one generated from
its own metadata. For example, a referenced directive used the `Reference`
that was created when analyzing the directive declaration itself. This still
works, as the compiler is always able to emit a reference to any valid
`Reference`. However, it's not optimal.
The `Reference` which should be used instead is the one generated from
analyzing the standalone component's `imports` array, which has knowledge of
how the dependency is referenced from within the standalone component's file
itself. This allows the compiler to skip creating a new import for the
dependency when emitting the standalone component, and use the existing,
user-authored import instead. This saves on code size and avoids taxing the
bundler with unnecessary imports.
PR Close#46029
The Angular compiler performs cycle detection when generating imports within
component files. This was previously necessary as reifying dependencies
discovered via NgModules into the component output could add imports that
weren't present in the original component and potentially create cycles.
Doing this could cause order-of-execution issues with existing user imports,
so the compiler detects this case and falls back to an alternative way of
specifying component dependencies that doesn't risk creating cycles.
For standalone components, Angular does not need to add new imports to the
component file as the user has already explicitly referenced dependencies
in the `@Component.imports`. As a result, the cycle detection can be
skipped.
Correctly authoring a program with import cycles is always challenging. One
side of a cyclic import will always initially evaluate to `undefined`, and
this can result in errors in the component definition when this happens
within component `imports`.
Our compiler _could_ detect the cycle and choose to wrap the component
dependencies in an automatic closure instead, avoiding any issues with
`undefined` during an eager evaluation. However, this commit makes an active
choice not to do that as it only serves to mask the problems with cyclic
imports. Future refactorings may cause the "other half" of the cycle to
break. Users should instead be aware of the potential problems with cycles
and explicitly defer evaluations with `forwardRef` where needed. This
ensures that future implementations of Angular compilation which may not be
able to automatically detect import cycles and correct accordingly can still
compile such components.
PR Close#46029
This commit improves the error message for using `imports` on a component
that isn't set to `standalone: true`. Two concrete improvements are made:
* A related information message is added to the diagnostic which suggests
the fix of adding `standalone: true`.
* The component is marked as poisoned, preventing other errors which might
be caused by an incorrectly configured template scope from being generated
and thus masking the original problem.
Fixes#45850
PR Close#45851
In AOT compilations, the `strictInjectionParameters` compiler option can
be enabled to report errors when an `@Injectable` annotated class has a
constructor with parameters that do not provide an injection token, e.g.
only a primitive type or interface.
Since Ivy it's become required that any class with Angular behavior
(e.g. the `ngOnDestroy` lifecycle hook) is decorated using an Angular
decorator, which meant that `@Injectable()` may need to have been added
to abstract base classes. Doing so would then report an error if
`strictInjectionParameters` is enabled, if the abstract class has an
incompatible constructor for DI purposes. This may be fine though, as
a subclass may call the constructor explicitly without relying on
Angular's DI mechanism.
Therefore, this commit excludes abstract classes from the
`strictInjectionParameters` check. This avoids an error from being
reported at compile time. If the constructor ends up being used by
Angular's DI system at runtime, then the factory function of the
abstract class will throw an error by means of the `ɵɵinvalidFactory`
instruction.
In addition to the runtime error, this commit also analyzes the inheritance
chain of an injectable without a constructor to verify that their inherited
constructor is valid.
Closes#37914
PR Close#44615
Excludes styles that resolve to empty strings from the emitted metadata so that they don't result in empty `<style>` tags at runtime.
Fixes#31191.
PR Close#45459
This commit updates the logic to detect a situation when a standalone component is used in the NgModule-based bootstrap (`@NgModule.bootstrap`). Both AOT and JIT compilers are updated to handle this situation.
PR Close#45825
Before standalone, everything that could appear in an NgModule's `imports`
was relevant to DI, and needed to be emitted in the `imports` of the
generated `InjectorDef` definition. With the introduction of standalone
types, NgModule `imports` can now contain components, directives, and pipes
which are standalone. Only standalone components need to be included in
the `imports` of the generated injector definition - directives and pipes
have no effect on DI. Having them present doesn't cause any errors in the
runtime (they're filtered out by the injector itself) but it does prevent
tree-shaking.
With this commit, the generation of `InjectorDef` now filters the `imports`
to exclude directives and pipes as much as possible. It's not _always_
possible because an expression in `imports` may pull in both a directive and
a `ModuleWithProviders` reference, and we have no way of referencing just
the MWP part of that expression. Therefore this is an optimization, not a
rule of `InjectorDef` compilation.
PR Close#45701
Previously, the NgModule handler would resolve the `imports` field as one
unit, producing an array of `Reference`s. With this refactoring, if
`imports` is a literal array, each individual element will be resolved
independently. This will allow filtering in the future at the element level,
since there will be a separate `ts.Expression` for each individual element.
PR Close#45701
This commit bundles tests for standalone components that are possible after
previous implementation commits. Most new tests are compliance tests, but
a test is also included to validate that the template type-checking system
can work with standalone components as well.
PR Close#45672
This commit adds a type field to .d.ts metadata for directives, components,
and pipes which carries a boolean literal indicating whether the given type
is standalone or not. For backwards compatibility, this flag defaults to
`false`.
Tests are added to validate that standalone types coming from .d.ts files
can be correctly imported into new standalone components.
PR Close#45672
This commit propagates the `isStandalone` flag for a component, directive,
or pipe during partial compilation of a standalone declaration. This flag
allows the linker to properly process a standalone declaration that it
encounters.
PR Close#45672
Standalone component scopes were first implemented in the
`ComponentDecoratorHandler` itself, due to an assumption that "standalone"
allowed for a localized analysis of the component's dependencies. However,
this is not strictly true. Other compiler machinery also needs to understand
component scopes, including standalone component scopes. A good example is
the template type-checking engine, which uses a `ComponentScopeReader` to
build full metadata objects (that is, metadata that considers the entire
inheritance chain) for type-checking purposes. Therefore, the
`ComponentScopeReader` should be able to give the scope for a standalone
component.
To achieve this, a new `StandaloneComponentScopeReader` is implemented, and
the return type of `ComponentScopeReader.getScopeForComponent` is expanded
to express standalone scopes. This cleanly integrates the "standalone"
concept into the existing machinery.
PR Close#45672
This commit expands on the unified dependency tracking in the previous
commit and adds tracking of NgModule dependencies. These are not used for
standard components, but are emitted for standalone components to allow the
runtime to roll up providers from those NgModules into standalone injectors.
PR Close#45672
Previously, the compiler tracked directives and pipes in template scopes
separately. This commit refactors the scope system to unify them into a
single data structure, disambiguated by a `kind` field.
PR Close#45672
Previously, the compiler would represent template dependencies of a
component in its component definition through separate fields (`directives`,
`pipes`).
This commit refactors the compiler/runtime interface to use a single field
(`dependencies`). The runtime component definition object still has separate
`directiveDefs` and `pipeDefs`, which are calculated from the `dependencies`
when the definition is evaluated.
This change is also reflected in partially compiled declarations. To ensure
compatibility with partially compiled code already on NPM, the linker
will still honor the old form of declaration (with separate fields).
PR Close#45672
This commit carries the `standalone` flag forward from a directive/pipe
into its generated directive/pipe definition, allowing the runtime to
recognize standalone entities.
PR Close#44973
This commit implements the next step of Angular's "standalone" functionality,
by allowing directives/components/pipes declared as `standalone` to be imported
into NgModules. Errors are raised when such a type is not standalone but is
included in an NgModule's imports.
PR Close#44973
This commit improves the error messages generated by the compiler when NgModule
scope analysis finds structural issues within a compilation. In particular,
errors are now shown on a node within the metadata of the NgModule which
produced the error, as opposed to the node of the erroneous declaration/import/
export. For example, if an NgModule declares `declarations: [FooCmp]` and
`FooCmp` is not annotated as a directive, component, or pipe, the error is now
shown on the reference to `FooCmp` in the `declarations` array expression.
Previously, the error would have been shown on `FooCmp` itself, with a mention
in the error text of the NgModule name.
Additional error context in some cases has been moved to related information
attached to the diagnostic, which further improves the legibility of such
errors. Error text has also been adjusted to be more succinct, since more info
about the error is now delivered through context.
PR Close#44973
.substr() is deprecated so we replace it with functions which work similarily but aren't deprecated
Signed-off-by: Tobias Speicher <rootcommander@gmail.com>
PR Close#45397
In early versions of Angular, it was sometimes necessary to provide a
`moduleId` to `@Component` metadata, and the common pattern for doing this
was to set `moduleId: module.id`. This relied on the bundler to fill in a
value for `module.id`.
However, due to the superficial similarity between `Component.moduleId` and
`NgModule.id`, many users ended up setting `id: module.id` in their
NgModules. This is an anti-pattern that has a few negative effects,
including preventing the NgModule from tree-shaking properly.
This commit changes the compiler to ignore `id: module.id` in NgModules, and
instead provide a warning which suggests removing the line entirely.
PR Close#45024
Angular contains an NgModule registry, which allows a user to declare
NgModules with string ids and retrieve them via those ids, using the
`getNgModuleById` API.
Previously, we attempted to structure this registration in a clever fashion
to allow for tree-shaking of registered NgModules (that is, those with ids).
This sort of worked due to the accidental alignment of behaviors from the
different tree-shakers involved. However, this trick relies on the
generation of `.ngfactory` files and how they're specifically processed in
various bundling scenarios. We intend to remove `.ngfactory` files, hence
we can no longer rely on them in this way.
The correct solution here is to recognize that `@NgModule({id})` is
inherently declaring a global side-effect, and such classes should not
really be eligible for tree-shaking in the first place. This commit removes
all the old registration machinery, and standardizes on generating a side-
effectful call to `registerNgModuleType` for NgModules that have ids.
There is some risk here that NgModules with unnecessary `id`s may not
tree-shake as a result of this change, whereas they would have in previous
circumstances. The fix here should be to remove the `id` if it's not needed.
Specifying an `id` is a request that the NgModule be retained regardless of
any other references, in case it is later looked up by string id.
PR Close#45024
Jasmine logs a warning when there's a `describe` with no tests. These changes fix one such case in the compiler that happens when the tests are run against Windows.
PR Close#45285
This commit implements the first phase of standalone components in the Angular
compiler. This mainly includes the scoping rules for standalone components
(`@Component({imports})`).
Significant functionality from the design is _not_ implemented by this PR,
including:
* imports of standalone components into NgModules.
* the provider aspect of standalone components
Future commits will address these issues, as we proceed with the design of
this feature.
PR Close#44812
Previously, if a bad extended diagnostic category was given, it would fail with the expected error as well as an unexpected assertion error:
```
$ ng build -c development
✔ Browser application bundle generation complete.
./src/main.ts - Error: Module build failed (from ./node_modules/@ngtools/webpack/src/ivy/index.js):
Error: Unexpected call to 'assertNever()' with value:
test
at /home/douglasparker/Source/ng-new/node_modules/@ngtools/webpack/src/ivy/loader.js:77:18
at processTicksAndRejections (internal/process/task_queues.js:95:5)
./src/polyfills.ts - Error: Module build failed (from ./node_modules/@ngtools/webpack/src/ivy/index.js):
Error: Unexpected call to 'assertNever()' with value:
test
at /home/douglasparker/Source/ng-new/node_modules/@ngtools/webpack/src/ivy/loader.js:77:18
at processTicksAndRejections (internal/process/task_queues.js:95:5)
Error: error NG4004: Angular compiler option "extendedDiagnostics.checks['invalidBananaInBox']" has an unknown diagnostic category: "test".
Allowed diagnostic categories are:
warning
error
suppress
```
The assertion comes from `ExtendedTemplateCheckerImpl`, which expects a well-formed configuration, yet the compiler would construct it even when errors were found. This commit skips constructing and running extended diagnostics if the configuration had errors, which should avoid triggering these assertion errors.
I'm unfortunately not able to actually test this change. The test passes even before the fix because the `ngc` binary and end-to-end tests [don't request diagnostics unless the configuration is considered valid](ed21f5c753/packages/compiler-cli/src/perform_compile.ts (L292-L293)). See [Slack](https://angular-team.slack.com/archives/C4WHZQMRA/p1642641305003800) for more details.
PR Close#44778
Refs #42966.
Extended diagnostics provide additional analysis about Angular templates by emitting warnings for specific patterns known to be error prone or cause developer confusion. Currently, there are two such diagnostics which are enabled by default:
* `invalidBananaInBox` emits a warning if a user writes a two-way binding backwards like `([foo])="bar"`, when they actually wanted `[(foo)]="bar"`.
* `nullishCoalescingNotNullable` emits a warning if a binding attempts to perform nullish coalescing (`??`) on a type which does not include `null` or `undefined`, such as `{{ foo ?? 'bar' }}` where `foo` is defined as `string` instead of `string | null`.
These diagnostics are enabled as warnings by default, but this can be configured in the `tsconfig.json` like so:
```jsonc
{
"angularCompilerOptions": {
"extendedDiagnostics": {
// The categories to use for specific diagnostics.
"checks": {
// Maps check name to its category.
"invalidBananaInBox": "suppress"
},
// The category to use for any diagnostics not listed in `checks` above.
"defaultCategory": "error"
}
}
}
```
Allowed categories for a diagnostic are `warning` (default), `error`, or `suppress`. `warning` emits the diagnostic but allows the compilation to succeed, `error` *will* fail the compilation, while `suppress` will ignore the diagnostic altogether.
The initial release has two diagnostics, and we are hoping to expand this longer term to add more diagnostics and provide additional insight into Angular templates to detect and surface developer mistakes *before* hours of debugging are wasted.
PR Close#44712
Refs #42966.
This validates the `tsconfig.json` options for extended template diagnostics. It verifies:
* `strictTemplates` must be enabled if `extendedDiagnostics` have any explicit configuration.
* `extendedDiagnostics.defaultCategory` must be a valid `DiagnosticCategoryLabel`.
* `extendedDiagnostics.checks` keys must all be real diagnostics.
* `extendedDiagnostics.checks` values must all be valid `DiagnosticCategoryLabel`s.
These include new error codes, each of which prints out the exact property that was the issue and what the available options are to fix them.
It does disallow the config:
```json
{
"angularCompilerOptions": {
"strictTemplates": false,
"extendedDiagnostics": {
"defaultCategory": "suppress"
}
}
}
```
Such a configuration is technically valid and could be executed, but will be rejected by this verification logic. There isn't much reason to ever do this, since users could just drop `extendedDiagnostics` altogether and get the intended effect. This is unlikely to be a significant issue for users, so it is considered invalid for now to keep the implementation simple.
PR Close#44391
Refs #42966.
The `defaultCategory` option is used for any extended template diagnostics which do not have any specific category specified for them. It defaults to `warning`, since that is the most common behavior expected for users. This provides an easy way for users to promote all diagnostics to errors or suppress all diagnostics.
PR Close#44391
Refs #42966.
This updates `TemplateContext` to include a new `makeTemplateDiagnostic()` function which automatically uses the correct diagnostic category for that check. This makes sure that each diagnostic is emitted with the correct category. It also implicitly passes some known values like `component` and `code` to make the extended template diagnostics a little simpler. Diagnostics which are suppressed are never instantiated at all, which acts as a slight performance optimization since any emitted diagnostics would be ignored anyways.
Unfortunately, diagnostics still have access to `ctx.templateTypeChecker.makeTemplateDiagnostic()` to manually create diagnostics with a different category. Both banana in box and nullish coalescing checks include tests to make sure they respect a manually configured category. This convention should hopefully give a reasonable certainty that new diagnostics will use the correct reporting function, even if that is not strictly enforced.
PR Close#44391
