When the `NgIf` directive is used in a template, its context variables
can be used to capture the bound value. This is typically used together
with a pipe or function call, where the resulting value is captured in a
context variable. There's two syntax forms available:
1. Binding to `NgIfContext.ngIf` using the `as` syntax:
```html
<span *ngIf="(user$ | async) as user">{{user.name}}</span>
```
2. Binding to `NgIfContext.$implicit` using the `let` syntax:
```html
<span *ngIf="user$ | async; let user">{{user.name}}</span>
```
Because of the semantics of `ngIf`, it is known that the captured
context variable is non-nullable, however the template type checker
would not consider them as such and still report errors when
`strictNullTypes` is enabled.
This commit updates `NgIf`'s context guard to make the types of the
context variables non-nullable, avoiding the issue.
Fixes#34572
PR Close#35125
For view and content queries, the Ivy compiler attempts to statically
evaluate the predicate token so that string predicates containing
comma-separated reference names can be split into an array of strings
during compilation. When the predicate is a dynamic value that cannot be
statically interpreted at compile time, the compiler would previously
produce an error. This behavior breaks a use-case where an `InjectionToken`
is being used as query predicate, as the usage of the `new` keyword
prevents such predicates from being statically evaluated.
This commit changes the behavior to no longer produce an error for
dynamic values. Instead, the expression is emitted as is into the
generated code, postponing the evaluation to happen at runtime.
Fixes#34267
Resolves FW-1828
PR Close#35307
It's possible to pass a directive as an input to itself. Consider:
```html
<some-cmp #ref [value]="ref">
```
Since the template type-checker attempts to infer a type for `<some-cmp>`
using the values of its inputs, this creates a circular reference where the
type of the `value` input is used in its own inference:
```typescript
var _t0 = SomeCmp.ngTypeCtor({value: _t0});
```
Obviously, this doesn't work. To resolve this, the template type-checker
used to generate a `null!` expression when a reference would otherwise be
circular:
```typescript
var _t0 = SomeCmp.ngTypeCtor({value: null!});
```
This effectively asks TypeScript to infer a value for this context, and
works well to resolve this simple cycle. However, if the template
instead tries to use the circular value in a larger expression:
```html
<some-cmp #ref [value]="ref.prop">
```
The checker would generate:
```typescript
var _t0 = SomeCmp.ngTypeCtor({value: (null!).prop});
```
In this case, TypeScript can't figure out any way `null!` could have a
`prop` key, and so it infers `never` as the type. `(never).prop` is thus a
type error.
This commit implements a better fallback pattern for circular references to
directive types like this. Instead of generating a `null!` in place for the
reference, a type is inferred by calling the type constructor again with
`null!` as its input. This infers the widest possible type for the directive
which is then used to break the cycle:
```typescript
var _t0 = SomeCmp.ngTypeCtor(null!);
var _t1 = SomeCmp.ngTypeCtor({value: _t0.prop});
```
This has the desired effect of validating that `.prop` is legal for the
directive type (the type of `#ref`) while also avoiding a cycle.
Fixes#35372Fixes#35603Fixes#35522
PR Close#35622
NG6002/NG6003 are errors produced when an NgModule being compiled has an
imported or exported type which does not have the proper metadata (that is,
it doesn't appear to be an @NgModule, or @Directive, etc. depending on
context).
Previously this error message was a bit sparse. However, Github issues show
that this is the most common error users receive when for whatever reason
ngcc wasn't able to handle one of their libraries, or they just didn't run
it. So this commit changes the error message to offer a bit more useful
context, instructing users differently depending on whether the class in
question is from their own project, from NPM, or from a monorepo-style local
dependency.
PR Close#35620
In #33705 we made it so that we generate pure functions for object/array literals in order to avoid having them be shared across elements/views. The problem this introduced is that further down the line the `ContantPool` uses the generated literal in order to figure out whether to share an existing factory or to create a new one. `ConstantPool` determines whether to share a factory by creating a key from the AST node and using it to look it up in the factory cache, however the key generation function didn't handle function invocations and replaced them with `null`. This means that the key for `{foo: pureFunction0(...)}` and `{foo: null}` are the same.
These changes rework the logic so that instead of generating a `null` key
for function invocations, we generate a variable called `<unknown>` which
shouldn't be able to collide with anything.
Fixes#35298.
PR Close#35481
Currently Ivy always generates the `$event` function argument, even if it isn't being used by the listener expressions. This can lead to unnecessary bytes being generated, because optimizers won't remove unused arguments by default. These changes add some logic to avoid adding the argument when it isn't required.
PR Close#35097
Prior to this commit, decorator handling logic in Ngtsc used `Error` to throw errors. This commit replaces most of these instances with `FatalDiagnosticError` class, which provider a better diagnostics error (including location of the problematic code).
PR Close#35244
In Ivy's template type checker, event bindings are checked in a closure
to allow for accurate type inference of the `$event` parameter. Because
of the closure, any narrowing effects of template guards will no longer
be in effect when checking the event binding, as TypeScript assumes that
the guard outside of the closure may no longer be true once the closure
is invoked. For more information on TypeScript's Control Flow Analysis,
please refer to https://github.com/microsoft/TypeScript/issues/9998.
In Angular templates, it is known that an event binding can only be
executed when the view it occurs in is currently rendered, hence the
corresponding template guard is known to hold during the invocation of
an event handler closure. As such, it is desirable that any narrowing
effects from template guards are still in effect within the event
handler closure.
This commit tweaks the generated Type-Check Block (TCB) to repeat all
template guards within an event handler closure. This achieves the
narrowing effect of the guards even within the closure.
Fixes#35073
PR Close#35193
A bug previously caused the template type-checking diagnostics produced by
TypeScript for template expressions to use -99-prefixed error codes. These
codes are converted to "NG" errors instead of "TS" errors during diagnostic
printing. This commit fixes the issue.
PR Close#35146
In #33551, a bug in `ngc --watch` mode was fixed so that a component is
recompiled when its template file is changed. Due to insufficient
normalization of files paths, this fix did not have the desired effect
on Windows.
Fixes#32869
PR Close#34015
We had some logic for generating and passing in the `elIndex` parameter into the `hostBindings` function, but it wasn't actually being used for anything. The only place left that had a reference to it was the `StylingBuilder` and it only stored it without referencing it again.
PR Close#34969
Component's decorator handler exposes `preanalyze` method to preload async resources (templates, stylesheets). The logic in preanalysis phase may throw `FatalDiagnosticError` errors that contain useful information regarding the origin of the problem. However these errors from preanalysis phase were not intercepted in TraitCompiler, resulting in just error message text be displayed. This commit updates the logic to handle FatalDiagnosticError and transform it before throwing, so that the result diagnostic errors contain the necessary info.
PR Close#34801
NOTE: This change must be reverted with previous deletes so that it code remains in build-able state.
This change deletes old styling code and replaces it with a simplified styling algorithm.
The mental model for the new algorithm is:
- Create a linked list of styling bindings in the order of priority. All styling bindings ere executed in compiled order and than a linked list of bindings is created in priority order.
- Flush the style bindings at the end of `advance()` instruction. This implies that there are two flush events. One at the end of template `advance` instruction in the template. Second one at the end of `hostBindings` `advance` instruction when processing host bindings (if any).
- Each binding instructions effectively updates the string to represent the string at that location. Because most of the bindings are additive, this is a cheap strategy in most cases. In rare cases the strategy requires removing tokens from the styling up to this point. (We expect that to be rare case)S Because, the bindings are presorted in the order of priority, it is safe to resume the processing of the concatenated string from the last change binding.
PR Close#34616
Compiler keeps track of number of slots (`vars`) which are needed for binding instructions. Normally each binding instructions allocates a single slot in the `LView` but styling instructions need to allocate two slots.
PR Close#34616
This change moves information from instructions to declarative position:
- `ɵɵallocHostVars(vars)` => `DirectiveDef.hostVars`
- `ɵɵelementHostAttrs(attrs)` => `DirectiveDef.hostAttrs`
When merging directives it is necessary to know about `hostVars` and `hostAttrs`. Before this change the information was stored in the `hostBindings` function. This was problematic, because in order to get to the information the `hostBindings` would have to be executed. In order for `hostBindings` to be executed the directives would have to be instantiated. This means that the directive instantiation would happen before we had knowledge about the `hostAttrs` and as a result the directive could observe in the constructor that not all of the `hostAttrs` have been applied. This further complicates the runtime as we have to apply `hostAttrs` in parts over many invocations.
`ɵɵallocHostVars` was unnecessarily complicated because it would have to update the `LView` (and Blueprint) while existing directives are already executing. By moving it out of `hostBindings` function we can access it statically and we can create correct `LView` (and Blueprint) in a single pass.
This change only changes how the instructions are generated, but does not change the runtime much. (We cheat by emulating the old behavior by calling `ɵɵallocHostVars` and `ɵɵelementHostAttrs`) Subsequent change will refactor the runtime to take advantage of the static information.
PR Close#34683
Previously, NgtscProgram lived in the main @angular/compiler-cli package
alongside the legacy View Engine compiler. As a result, the main package
depended on all of the ngtsc internal packages, and a significant portion of
ngtsc logic lived in NgtscProgram.
This commit refactors NgtscProgram and moves the main logic of compilation
into a new 'core' package. The new package defines a new API which enables
implementers of TypeScript compilers (compilers built using the TS API) to
support Angular transpilation as well. It involves a new NgCompiler type
which takes a ts.Program and performs Angular analysis and transformations,
as well as an NgCompilerHost which wraps an input ts.CompilerHost and adds
any extra Angular files.
Together, these two classes are used to implement a new NgtscProgram which
adapts the legacy api.Program interface used by the View Engine compiler
onto operations on the new types. The new NgtscProgram implementation is
significantly smaller and easier to reason about.
The new NgCompilerHost replaces the previous GeneratedShimsHostWrapper which
lived in the 'shims' package.
A new 'resource' package is added to support the HostResourceLoader which
previously lived in the outer compiler package.
As a result of the refactoring, the dependencies of the outer
@angular/compiler-cli package on ngtsc internal packages are significantly
trimmed.
This refactoring was driven by the desire to build a plugin interface to the
compiler so that tsc_wrapped (another consumer of the TS compiler APIs) can
perform Angular transpilation on user request.
PR Close#34887
This commit fixes a bug in the incremental rebuild engine of ngtsc, where if
a component was removed from its NgModule, it would not be properly
re-emitted.
The bug stemmed from the fact that whether to emit a file was a decision
based purely on the updated dependency graph, which captures the dependency
structure of the rebuild program. This graph has no edge from the component
to its former module (as it was removed, of course), so the compiler
erroneously decides not to emit the component.
The bug here is that the compiler does know, from the previous dependency
graph, that the component file has logically changed, since its previous
dependency (the module file) has changed. This information was not carried
forward into the set of files which need to be emitted, because it was
assumed that the updated dependency graph was a more accurate source of that
information.
With this commit, the set of files which need emit is pre-populated with the
set of logically changed files, to cover edge cases like this.
Fixes#34813
PR Close#34912
Previously, the template type-checker would always construct a generic
template context type with correct bounds, even when strictTemplates was
disabled. This meant that type-checking of expressions involving that type
was stricter than View Engine.
This commit introduces a 'strictContextGenerics' flag which behaves
similarly to other 'strictTemplates' flags, and switches the inference of
generic type parameters on the component context based on the value of this
flag.
PR Close#34649
FileToModuleHost aliasing supports compilation within environments that have
two properties:
1. A `FileToModuleHost` exists which defines canonical module names for any
given TS file.
2. Dependency restrictions exist which prevent the import of arbitrary files
even if such files are within the .d.ts transitive closure of a
compilation ("strictdeps").
In such an environment, generated imports can only go through import paths
which are already present in the user program. The aliasing system supports
the generation and consumption of such imports at runtime.
`FileToModuleHost` aliasing does not emit re-exports in .d.ts files. This
means that it's safe to rely on alias re-exports in generated .js code (they
are guaranteed to exist at runtime) but not in template type-checking code
(since TS will not be able to follow such imports). Therefore, non-aliased
imports should be used in template type-checking code.
This commit adds a `NoAliasing` flag to `ImportFlags` and sets it when
generating imports in template type-checking code. The testing environment
is also patched to support resolution of FileToModuleHost canonical paths
within the template type-checking program, enabling testing of this change.
PR Close#34649
It's possible to declare multiple inputs for a directive/component which all
map to the same property name. This is usually done in error, as only one of
any bindings to the property will "win".
In the template type-checker, an error was previously being raised as a
result of this ambiguity. Specifically, a type constructor was produced
which required a binding for each field, but only one of the fields had
a value via the binding. TypeScript would (rightfully) error on missing
values for the remaining fields. This ultimately was happening when the
code which generated the default values for "unset" inputs belonging to
directives or pipes used the final mapping from properties to fields as
a source for field names.
Instead, this commit uses the original list of fields to generate unset
input values, which correctly provides values for fields which shared a
property name but didn't receive the final binding.
PR Close#34649
This patch removes the need for the styleSanitizer() instruction in
favor of passing the sanitizer into directly into the styleProp
instruction.
This patch also increases the binding index size for all style/class bindings in preparation for #34418
PR Close#34480
Pipes in host binding expressions are not supported in View Engine and Ivy, but in some more complex cases (like `(value | pipe) === true`) compiler was not reporting errors. This commit extends Ivy logic to detect pipes in host binding expressions and throw in cases bindings are present. View Engine behavior remains the same.
PR Close#34655
The major one that affects the angular repo is the removal of the bootstrap attribute in nodejs_binary, nodejs_test and jasmine_node_test in favor of using templated_args --node_options=--require=/path/to/script. The side-effect of this is that the bootstrap script does not get the require.resolve patches with explicitly loading the targets _loader.js file.
PR Close#34736
Currently ngtsc looks for the first `ConstructorDeclaration` when figuring out what the parameters are so that it can generate the DI instructions. The problem is that if a constructor has overloads, it'll have several `ConstructorDeclaration` members with a different number of parameters. These changes tweak the logic so it looks for the constructor implementation.
PR Close#34590
The major one that affects the angular repo is the removal of the bootstrap attribute in nodejs_binary, nodejs_test and jasmine_node_test in favor of using templated_args --node_options=--require=/path/to/script. The side-effect of this is that the bootstrap script does not get the require.resolve patches with explicitly loading the targets _loader.js file.
PR Close#34589
The `getProjectAsAttrValue` in `node_selector_matcher` finds the
ProjectAs marker and then additionally checks that the marker appears in
an even index of the node attributes because "attribute names are stored
at even indexes". This is true for "regular" attribute bindings but
classes, styles, bindings, templates, and i18n do not necessarily follow
this rule because there can be an uneven number of them, causing the
next "special" attribute "name" to appear at an odd index. To address
this issue, ensure ngProjectAs is placed right after "regular"
attributes.
PR Close#34617
Prior to this commit, there were no `advance` instructions generated before `i18nExp` instructions and as a result, lifecycle hooks for components used inside i18n blocks were flushed too late. This commit adds the logic to generate `advance` instructions in front of `i18nExp` ones (similar to what we have in other places like interpolations, property bindings, etc), so that the necessary lifecycle hooks are flushed before expression value is captured.
PR Close#34436
Previously, it was required that both `fullTemplateTypeCheck` and
`strictTemplates` had to be enabled for strict mode to be enabled. This
is strange, as `strictTemplates` implies `fullTemplateTypeCheck`. This
commit makes setting the `fullTemplateTypeCheck` flag optional so that
strict mode can be enabled by just setting `strictTemplates`.
PR Close#34195
It is now an error if '"fullTemplateTypeCheck"' is disabled while
`"strictTemplates"` is enabled, as enabling the latter implies that the
former is also enabled.
PR Close#34195
The compiler has a translation mechanism to convert from an Angular
`Type` to a `ts.TypeNode`, as appropriate. Prior to this change, it
would translate certain Angular expressions into their value equivalent
in TypeScript, instead of the correct type equivalent. This was possible
as the `ExpressionVisitor` interface is not strictly typed, with `any`s
being used for return values.
For example, a literal object was translated into a
`ts.ObjectLiteralExpression`, containing `ts.PropertyAssignment` nodes
as its entries. This has worked without issues as their printed
representation is identical, however it was incorrect from a semantic
point of view. Instead, a `ts.TypeLiteralNode` is created with
`ts.PropertySignature` as its members, which corresponds with the type
declaration of an object literal.
PR Close#34021
Angular View Engine uses global knowledge to compile the following code:
```typescript
export class Base {
constructor(private vcr: ViewContainerRef) {}
}
@Directive({...})
export class Dir extends Base {
// constructor inherited from base
}
```
Here, `Dir` extends `Base` and inherits its constructor. To create a `Dir`
the arguments to this inherited constructor must be obtained via dependency
injection. View Engine is able to generate a correct factory for `Dir` to do
this because via metadata it knows the arguments of `Base`'s constructor,
even if `Base` is declared in a different library.
In Ivy, DI is entirely a runtime concept. Currently `Dir` is compiled with
an ngDirectiveDef field that delegates its factory to `getInheritedFactory`.
This looks for some kind of factory function on `Base`, which comes up
empty. This case looks identical to an inheritance chain with no
constructors, which works today in Ivy.
Both of these cases will now become an error in this commit. If a decorated
class inherits from an undecorated base class, a diagnostic is produced
informing the user of the need to either explicitly declare a constructor or
to decorate the base class.
PR Close#34460
Adds a compilation error if the consumer tries to pass in an undecorated class into the `providers` of an `NgModule`, or the `providers`/`viewProviders` arrays of a `Directive`/`Component`.
PR Close#34460
Previously, ngtsc would perform scope analysis (which directives/pipes are
available inside a component's template) and template type-checking of that
template as separate steps. If a component's scope was somehow invalid (e.g.
its NgModule imported something which wasn't another NgModule), the
component was treated as not having a scope. This meant that during template
type-checking, errors would be produced for any invalid expressions/usage of
other components that should have been in the scope.
This commit changes ngtsc to skip template type-checking of a component if
its scope is erroneous (as opposed to not present in the first place). Thus,
users aren't overwhelmed with diagnostic errors for the template and are
only informed of the root cause of the problem: an invalid NgModule scope.
Fixes#33849
PR Close#34460
This commit adds three previously missing validations to
NgModule.declarations:
1. It checks that declared classes are actually within the current
compilation.
2. It checks that declared classes are directives, components, or pipes.
3. It checks that classes are declared in at most one NgModule.
PR Close#34404
In Ivy it's illegal for a template to write to a template variable. So the
template:
```html
<ng-template let-somevar>
<button (click)="somevar = 3">Set var to 3</button>
</ng-template>
```
is erroneous and previously would fail to compile with an assertion error
from the `TemplateDefinitionBuilder`. This error wasn't particularly user-
friendly, though, as it lacked the context of which template or where the
error occurred.
In this commit, a new check in template type-checking is added which detects
such erroneous writes and produces a true diagnostic with the appropriate
context information.
Closes#33674
PR Close#34339
Previously, the compiler performed an incremental build by analyzing and
resolving all classes in the program (even unchanged ones) and then using
the dependency graph information to determine which .js files were stale and
needed to be re-emitted. This algorithm produced "correct" rebuilds, but the
cost of re-analyzing the entire program turned out to be higher than
anticipated, especially for component-heavy compilations.
To achieve performant rebuilds, it is necessary to reuse previous analysis
results if possible. Doing this safely requires knowing when prior work is
viable and when it is stale and needs to be re-done.
The new algorithm implemented by this commit is such:
1) Each incremental build starts with knowledge of the last known good
dependency graph and analysis results from the last successful build,
plus of course information about the set of files changed.
2) The previous dependency graph's information is used to determine the
set of source files which have "logically" changed. A source file is
considered logically changed if it or any of its dependencies have
physically changed (on disk) since the last successful compilation. Any
logically unchanged dependencies have their dependency information copied
over to the new dependency graph.
3) During the `TraitCompiler`'s loop to consider all source files in the
program, if a source file is logically unchanged then its previous
analyses are "adopted" (and their 'register' steps are run). If the file
is logically changed, then it is re-analyzed as usual.
4) Then, incremental build proceeds as before, with the new dependency graph
being used to determine the set of files which require re-emitting.
This analysis reuse avoids template parsing operations in many circumstances
and significantly reduces the time it takes ngtsc to rebuild a large
application.
Future work will increase performance even more, by tackling a variety of
other opportunities to reuse or avoid work.
PR Close#34288
The `ModuleWithProviders` type has an optional type parameter that
should be specified to indicate what NgModule class will be provided.
This enables the Ivy compiler to statically determine the NgModule type
from the declaration files. This type parameter will become required in
the future, however to aid in the migration the compiler will detect
code patterns where using `ModuleWithProviders` as return type is
appropriate, in which case it transforms the emitted .d.ts files to
include the generic type argument.
This should reduce the number of occurrences where `ModuleWithProviders`
is referenced without its generic type argument.
Resolves FW-389
PR Close#34235
The metadata collector for View Engine compilations emits error symbols
for static class members that have not been initialized, which prevents
a library from building successfully when `strictMetadataEmit` is
enabled, which is recommended for libraries to avoid issues in library
consumers. This is troublesome for libraries that are adopting static
members for the Ivy template type checker: these members don't need a
value assignment as only their type is of importance, however this
causes metadata errors. As such, a library used to be required to
initialize the special static members to workaround this error,
undesirably introducing a code-size overhead in terms of emitted
JavaScript code.
This commit modifies the collector logic to specifically ignore
the special static members for Ivy's template type checker, preventing
any errors from being recorded during the metadata collection.
PR Close#34296
For Ivy's template type checker it is possible to let a directive
specify static members to allow a wider type for some input:
```typescript
export class MatSelect {
@Input() disabled: boolean;
static ngAcceptInputType_disabled: boolean | string;
}
```
This allows a binding to the `MatSelect.disabled` input to be of type
boolean or string, whereas the `disabled` property itself is only of
type boolean.
Up until now, any static `ngAcceptInputType_*` property was not
inherited for subclasses of a directive class. This is cumbersome, as
the directive's inputs are inherited, so any acceptance member should as
well. To resolve this limitation, this commit extends the flattening of
directive metadata to include the acceptance members.
Fixes#33830
Resolves FW-1759
PR Close#34296
The compiler exports a `formatDiagnostics` function which consumers can use
to print both ts and ng diagnostics. However, this function was previously
using the "old" style TypeScript diagnostics, as opposed to the modern
diagnostic printer which uses terminal colors and prints additional context
information.
This commit updates `formatDiagnostics` to use the modern formatter, plus to
update Ivy's negative error codes to Angular 'NG' errors.
The Angular CLI needs a little more work to use this function for printing
TS diagnostics, but this commit alone should fix Bazel builds as ngc-wrapped
goes through `formatDiagnostics`.
PR Close#34234
Previously, ternary expressions were emitted as:
condExpr ? trueCase : falseCase
However, this causes problems when ternary operations are nested. In
particular, a template expression of the form:
a?.b ? c : d
would have compiled to:
a == null ? null : a.b ? c : d
The ternary operator is right-associative, so that expression is interpreted
as:
a == null ? null : (a.b ? c : d)
when in reality left-associativity is desired in this particular instance:
(a == null ? null : a.b) ? c : d
This commit adds a check in the expression translator to detect such
left-associative usages of ternaries and to enforce such associativity with
parentheses when necessary.
A test is also added for the template type-checking expression translator,
to ensure it correctly produces right-associative expressions for ternaries
in the user's template.
Fixes#34087
PR Close#34221
JoostK
Alex Rickabaugh
crisbeto
Miško Hevery
Andrew Kushnir
Matias Niemelä
Greg Magolan
atscott
Andrew Scott