NgModules in Ivy have a definition which contains various different bits
of metadata about the module. In particular, this metadata falls into two
categories:
* metadata required to use the module at runtime (for bootstrapping, etc)
in AOT-only applications.
* metadata required to depend on the module from a JIT-compiled app.
The latter metadata consists of the module's declarations, imports, and
exports. To support JIT usage, this metadata must be included in the
generated code, especially if that code is shipped to NPM. However, because
this metadata preserves the entire NgModule graph (references to all
directives and components in the app), it needs to be removed during
optimization for AOT-only builds.
Previously, this was done with a clever design:
1. The extra metadata was added by a function called `setNgModuleScope`.
A call to this function was generated after each NgModule.
2. This function call was marked as "pure" with a comment and used
`noSideEffects` internally, which causes optimizers to remove it.
The effect was that in dev mode or test mode (which use JIT), no optimizer
runs and the full NgModule metadata was available at runtime. But in
production (presumably AOT) builds, the optimizer runs and removes the JIT-
specific metadata.
However, there are cases where apps that want to use JIT in production, and
still make an optimized build. In this case, the JIT-specific metadata would
be erroneously removed. This commit solves that problem by adding an
`ngJitMode` global variable which guards all `setNgModuleScope` calls. An
optimizer can be configured to statically define this global to be `false`
for AOT-only builds, causing the extra metadata to be stripped.
A configuration for Terser used by the CLI is provided in `tooling.ts` which
sets `ngJitMode` to `false` when building AOT apps.
PR Close#33671
The Ivy template type-checker is capable of inferring the type of a
structural directive (such as NgForOf<T>). Previously, this was done with
fullTemplateTypeCheck: true, even if strictTemplates was false. View Engine
previously did not do this inference, and so this causes breakages if the
type of the template context is not what the user expected.
In particular, consider the template:
```html
<div *ngFor="let user of users as all">
{{user.index}} out of {{all.length}}
</div>
```
As long as `users` is an array, this seems reasonable, because it appears
that `all` is an alias for the `users` array. However, this is misleading.
In reality, `NgForOf` is rendered with a template context that contains
both a `$implicit` value (for the loop variable `user`) as well as a
`ngForOf` value, which is the actual value assigned to `all`. The type of
`NgForOf`'s template context is `NgForContext<T>`, which declares `ngForOf`'s
type to be `NgIterable<T>`, which does not have a `length` property (due to
its incorporation of the `Iterable` type).
This commit stops the template type-checker from inferring template context
types unless strictTemplates is set (and strictInputTypes is not disabled).
Fixes#33527.
PR Close#33537
Previously, the ngtsc compiler attempted to reuse analysis work from the
previous program during an incremental build. To do this, it had to prove
that the work was safe to reuse - that no changes made to the new program
would invalidate the previous analysis.
The implementation of this had a significant design flaw: if the previous
program had errors, the previous analysis would be missing significant
information, and the dependency graph extracted from it would not be
sufficient to determine which files should be re-analyzed to fill in the
gaps. This often meant that the build output after an error was resolved
would be wholly incorrect.
This commit switches ngtsc to take a simpler approach to incremental
rebuilds. Instead of attempting to reuse prior analysis work, the entire
program is re-analyzed with each compilation. This is actually not as
expensive as one might imagine - analysis is a fairly small part of overall
compilation time.
Based on the dependency graph extracted during this analysis, the compiler
then can make accurate decisions on whether to emit specific files. A new
suite of tests is added to validate behavior in the presence of source code
level errors.
This new approach is dramatically simpler than the previous algorithm, and
should always produce correct results for a semantically correct program.s
Fixes#32388Fixes#32214
PR Close#33862
Originally, QueryList implemented Iterable and provided a Symbol.iterator
on its prototype. This caused issues with tree-shaking, so QueryList was
refactored and the Symbol.iterator added in its constructor instead. As
part of this change, QueryList no longer implemented Iterable directly.
Unfortunately, this meant that QueryList was no longer assignable to
Iterable or, consequently, NgIterable. NgIterable is used for NgFor's input,
so this meant that QueryList was not usable (in a type sense) for NgFor
iteration. View Engine's template type checking would not catch this, but
Ivy's did.
As a fix, this commit adds the declaration (but not the implementation) of
the Symbol.iterator function back to QueryList. This has no runtime effect,
so it doesn't affect tree-shaking of QueryList, but it ensures that
QueryList is assignable to NgIterable and thus usable with NgFor.
Fixes#29842
PR Close#33536
Previously, the compiler assumed that all TS files logically within a
project existed under one or more "root directories". If the TS compiler
option `rootDir` or `rootDirs` was set, they would dictate the root
directories in use, otherwise the current directory was used.
Unfortunately this assumption was unfounded - it's common for projects
without explicit `rootDirs` to import from files outside the current
working directory. In such cases the `LogicalProjectStrategy` would attempt
to generate imports into those files, and fail. This would lead to no
`ReferenceEmitStrategy` being able to generate an import, and end in a
compiler assertion failure.
This commit introduces a new strategy to use when there are no `rootDirs`
explicitly present, the `RelativePathStrategy`. It uses simpler, filesystem-
relative paths to generate imports, even to files above the current working
directory.
Fixes#33659Fixes#33562
PR Close#33828
This commit adds the ability to change directories using the compiler's
internal filesystem abstraction. This is a prerequisite for writing tests
which are sensitive to the current working directory.
In addition to supporting the `chdir()` operation, this commit also fixes
`getDefaultLibLocation()` for mock filesystems to not assume `node_modules`
is in the current directory, but to resolve it similarly to how Node does
by progressively looking higher in the directory tree.
PR Close#33828
Adds support for chaining of `styleProp`, `classProp` and `stylePropInterpolateX` instructions whenever possible which should help generate less code. Note that one complication here is for `stylePropInterpolateX` instructions where we have to break into multiple chains if there are other styling instructions inbetween the interpolations which helps maintain the execution order.
PR Close#33837
Since i18n messages are mapped to `$localize` tagged template strings,
the "raw" version must be properly escaped. Otherwise TS will throw an
error such as:
```
Error: Debug Failure. False expression: Expected argument 'text' to be the normalized (i.e. 'cooked') version of argument 'rawText'.
```
This commit ensures that we properly escape these raw strings before creating
TS AST nodes from them.
PR Close#33820
The `:` char is used as a metadata marker in `$localize` messages.
If this char appears in the metadata it must be escaped, as `\:`.
Previously, although the `:` char was being escaped, the TS AST
being generated was not correct and so it was being output double
escaped, which meant that it appeared in the rendered message.
As of TS 3.6.2 the "raw" string can be specified when creating tagged
template AST nodes, so it is possible to correct this.
PR Close#33820
The following files are consumed only by the language service and do not
have to be in compiler-cli:
1. expression_diagnostics.ts
2. expression_type.ts
3. typescript_symbols.ts
4. symbols.ts
PR Close#33809
Previously, due to a bug a `Context` with `isStatement: false` could be
returned in places where a `Context` with `isStatement: true` was
requested. As a result, some statements would be unnecessarily wrapped
in parenthesis.
This commit fixes the bug in `Context#withStatementMode` to always
return a `Context` with the correct `isStatement` value. Note that this
does not have any impact on the generated code other than avoiding some
superfluous parenthesis on certain statements.
PR Close#33514
Currently if a consumer does something like the following, the object literal will be shared across the two elements and any instances of the component template. The same applies to array literals:
```
<div [someDirective]="{}"></div>
<div [someDirective]="{}"></div>
```
These changes make it so that we generate a pure function even if an object is constant so that each instance gets its own object.
Note that the original design for this fix included moving the pure function factories into the `consts` array. In the process of doing so I realized that pure function are also used inside of directive host bindings which means that we don't have access to the `consts`.
These changes also:
* Fix an issue that meant that the `pureFunction0` instruction could only be run during creation mode.
* Make the `getConstant` utility slightly more convenient to use. This isn't strictly required for these changes to work, but I had made it as a part of a larger refactor that I ended up reverting.
PR Close#33705
During incremental compilations, ngtsc needs to know which metadata
from a previous compilation can be reused, versus which metadata has to
be recomputed as some dependency was updated. Changes to
directives/components should cause the NgModule in which they are
declared to be recompiled, as the NgModule's compilation is dependent
on its directives/components.
When a dependent source file of a directive/component is updated,
however, a more subtle dependency should also cause to NgModule's source
file to be invalidated. During the reconciliation of state from a
previous compilation into the new program, the component's source file
is invalidated because one of its dependency has changed, ergo the
NgModule needs to be invalidated as well. Up until now, this implicit
dependency was not imposed on the NgModule. Additionally, any change to
a dependent file may influence the module scope to change, so all
components within the module must be invalidated as well.
This commit fixes the bug by introducing additional file dependencies,
as to ensure a proper rebuild of the module scope and its components.
Fixes#32416
PR Close#33522
When the Angular compiler is operated through the ngc binary in watch
mode, changing a template in an external file would not cause the
component to be recompiled if Ivy is enabled.
There was a problem with how a cached compiler host was present that was
unaware of the changed resources, therefore failing to trigger a
recompilation of a component whenever its template changes. This commit
fixes the issue by ensuring that information about modified resources is
correctly available to the cached compiler host.
Fixes#32869
PR Close#33551
Chains multiple listener instructions on a particular element into a single call which results in less generated code. Also handles listeners on templates, host listeners and synthetic host listeners.
PR Close#33720
When the Angular compiler is operated through the ngc binary in watch
mode, changing a template in an external file would not cause the
component to be recompiled if Ivy is enabled.
There was a problem with how a cached compiler host was present that was
unaware of the changed resources, therefore failing to trigger a
recompilation of a component whenever its template changes. This commit
fixes the issue by ensuring that information about modified resources is
correctly available to the cached compiler host.
Fixes#32869
PR Close#33551
This commit moves nested i18n section detection to an earlier stage where we convert HTML AST to Ivy AST. This also gives a chance to produce better diagnistic message for nested i18n sections, that also includes a file name and location.
PR Close#33583
These exports are no longer used by the CLI since 7.1.0. Since major versions of the CLI are now locked to major versions of the framework, a CLI user will not be able to use FW 9.0+ on an outdated version (<7.1.0) of the CLI that uses these old APIs.
PR Close#33242
Previously the compiler would crash if a pipe was encountered which did not
match any pipe in the scope of a template.
This commit introduces a new diagnostic error for unknown pipes instead.
PR Close#33454
Previously the template binder would crash when encountering an unknown
localref (# reference) such as `<div #ref="foo">` when no directive has
`exportAs: "foo"`.
With this commit, the compiler instead generates a template diagnostic error
informing the user about the invalid reference.
PR Close#33454
Previously declarations that were imported via a namespace import
were given the same `bestGuessOwningModule` as the context
where they were imported to. This causes problems with resolving
`ModuleWithProviders` that have a type that has been imported in
this way, causing errors like:
```
ERROR in Symbol UIRouterModule declared in
.../@uirouter/angular/uiRouterNgModule.d.ts
is not exported from
.../@uirouter/angular/uirouter-angular.d.ts
(import into .../src/app/child.module.ts)
```
This commit modifies the `TypescriptReflectionHost.getDirectImportOfIdentifier()`
method so that it also understands how to attach the correct `viaModule` to
the identifier of the namespace import.
Resolves#32166
PR Close#33495
Now that we've replaced `ngBaseDef` with an abstract directive definition, there are a lot more cases where we generate a directive definition without a selector. These changes make it so that we don't generate the `selectors` array if it's going to be empty.
PR Close#33431
Removes `ngBaseDef` from the compiler and any runtime code that was still referring to it. In the cases where we'd previously generate a base def we now generate a definition for an abstract directive.
PR Close#33264
For abstract directives, i.e. directives without a selector, it may
happen that their constructor is called explicitly from a subclass,
hence its parameters are not required to be valid for Angular's DI
purposes. Prior to this commit however, having an abstract directive
with a constructor that has parameters that are not eligible for
Angular's DI would produce a compilation error.
A similar scenario may occur for `@Injectable`s, where an explicit
`use*` definition allows for the constructor to be irrelevant. For
example, the situation where `useFactory` is specified allows for the
constructor to be called explicitly with any value, so its constructor
parameters are not required to be valid. For `@Injectable`s this is
handled by generating a DI factory function that throws.
This commit implements the same solution for abstract directives, such
that a compilation error is avoided while still producing an error at
runtime if the type is instantiated implicitly by Angular's DI
mechanism.
Fixes#32981
PR Close#32987
This patch ensures that the `[style]` and `[class]` based bindings
are directly applied to an element's style and className attributes.
This patch optimizes the algorithm so that it...
- Doesn't construct an update an instance of `StylingMapArray` for
`[style]` and `[class]` bindings
- Doesn't apply `[style]` and `[class]` based entries using
`classList` and `style` (direct attributes are used instead)
- Doesn't split or iterate over all string-based tokens in a
string value obtained from a `[class]` binding.
This patch speeds up the following cases:
- `<div [class]>` and `<div class="..." [class]>`
- `<div [style]>` and `<div style="..." [style]>`
The overall speec increase is by over 5x.
PR Close#33336
The template type checking abilities of the Ivy compiler are far more
advanced than the level of template type checking that was previously
done for Angular templates. Up until now, a single compiler option
called "fullTemplateTypeCheck" was available to configure the level
of template type checking. However, now that more advanced type checking
is being done, new errors may surface that were previously not reported,
in which case it may not be feasible to fix all new errors at once.
Having only a single option to disable a large number of template type
checking capabilities does not allow for incrementally addressing newly
reported types of errors. As a solution, this commit introduces some new
compiler options to be able to enable/disable certain kinds of template
type checks on a fine-grained basis.
PR Close#33365
During the creation of an Angular program in the compiler, a check is
done to verify whether the version of TypeScript is considered
supported, producing an error if it is not. This check was missing in
the Ivy compiler, so users may have ended up running an unsupported
TypeScript version inadvertently.
Resolves FW-1643
PR Close#33377
Recently it was made possible to have a directive without selector,
which are referred to as abstract directives. Such directives should not
be registered in an NgModule, but can still contain decorators for
inputs, outputs, queries, etc. The information from these decorators and
the `@Directive()` decorator itself needs to be registered with the
central `MetadataRegistry` so that other areas of the compiler can
request information about a given directive, an example of which is the
template type checker that needs to know about the inputs and outputs of
directives.
Prior to this change, however, abstract directives would only register
themselves with the `MetadataRegistry` as being an abstract directive,
without all of its other metadata like inputs and outputs. This meant
that the template type checker was unable to resolve the inputs and
outputs of these abstract directives, therefore failing to check them
correctly. The typical error would be that some property does not exist
on a DOM element, whereas said property should have been bound to the
abstract directive's input.
This commit fixes the problem by always registering the metadata of a
directive or component with the `MetadataRegistry`. Tests have been
added to ensure abstract directives are handled correctly in the
template type checker, together with tests to verify the form of
abstract directives in declaration files.
Fixes#30080
PR Close#33131
Often the types of an `@Input`'s field don't fully reflect the types of
assignable values. This can happen when an input has a getter/setter pair
where the getter always returns a narrow type, and the setter coerces a
wider value down to the narrow type.
For example, you could imagine an input of the form:
```typescript
@Input() get value(): string {
return this._value;
}
set value(v: {toString(): string}) {
this._value = v.toString();
}
```
Here, the getter always returns a `string`, but the setter accepts any value
that can be `toString()`'d, and coerces it to a string.
Unfortunately TypeScript does not actually support this syntax, and so
Angular users are forced to type their setters as narrowly as the getters,
even though at runtime the coercion works just fine.
To support these kinds of patterns (e.g. as used by Material), this commit
adds a compiler feature called "input coercion". When a binding is made to
the 'value' input of a directive like MatInput, the compiler will look for a
static field with the name ngAcceptInputType_value. If such a field is found
the type-checking expression for the input will use the static field's type
instead of the type for the @Input field,allowing for the expression of a
type conversion between the binding expression and the value being written
to the input's field.
To solve the case above, for example, MatInput might write:
```typescript
class MatInput {
// rest of the directive...
static ngAcceptInputType_value: {toString(): string};
}
```
FW-1475 #resolve
PR Close#33243
Libraries can expose directive/component base classes that will be
used by consumer applications. Using such a base class from another
compilation unit works fine with "ngtsc", but when using "ngc", the
compiler will thrown an error saying that the base class is not
part of a NgModule. e.g.
```
Cannot determine the module for class X in Y! Add X to the NgModule to fix it.
```
This seems to be because the logic for distinguishing directives from
abstract directives is scoped to the current compilation unit within
ngc. This causes abstract directives from other compilation units to
be considered as actual directives (causing the exception).
PR Close#33347
When computing i18n messages for templates there are two passes.
This is because messages must be computed before any whitespace
is removed. Then on a second pass, the messages must be recreated
but reusing the message ids from the first pass.
Previously ICUs were losing their legacy ids that had been computed
via the first pass. This commit fixes that by keeping track of the
message from the first pass (`previousMessage`) for ICU placeholder
nodes.
// FW-1637
PR Close#33318
This commit refactors the aliasing system to support multiple different
AliasingHost implementations, which control specific aliasing behavior
in ngtsc (see the README.md).
A new host is introduced, the `PrivateExportAliasingHost`. This solves a
longstanding problem in ngtsc regarding support for "monorepo" style private
libraries. These are libraries which are compiled separately from the main
application, and depended upon through TypeScript path mappings. Such
libraries are frequently not in the Angular Package Format and do not have
entrypoints, but rather make use of deep import style module specifiers.
This can cause issues with ngtsc's ability to import a directive given the
module specifier of its NgModule.
For example, if the application uses a directive `Foo` from such a library
`foo`, the user might write:
```typescript
import {FooModule} from 'foo/module';
```
In this case, foo/module.d.ts is path-mapped into the program. Ordinarily
the compiler would see this as an absolute module specifier, and assume that
the `Foo` directive can be imported from the same specifier. For such non-
APF libraries, this assumption fails. Really `Foo` should be imported from
the file which declares it, but there are two problems with this:
1. The compiler would have to reverse the path mapping in order to determine
a path-mapped path to the file (maybe foo/dir.d.ts).
2. There is no guarantee that the file containing the directive is path-
mapped in the program at all.
The compiler would effectively have to "guess" 'foo/dir' as a module
specifier, which may or may not be accurate depending on how the library and
path mapping are set up.
It's strongly desirable that the compiler not break its current invariant
that the module specifier given by the user for the NgModule is always the
module specifier from which directives/pipes are imported. Thus, for any
given NgModule from a particular module specifier, it must always be
possible to import any directives/pipes from the same specifier, no matter
how it's packaged.
To make this possible, when compiling a file containing an NgModule, ngtsc
will automatically add re-exports for any directives/pipes not yet exported
by the user, with a name of the form: ɵngExportɵModuleNameɵDirectiveName
This has several effects:
1. It guarantees anyone depending on the NgModule will be able to import its
directives/pipes from the same specifier.
2. It maintains a stable name for the exported symbol that is safe to depend
on from code on NPM. Effectively, this private exported name will be a
part of the package's .d.ts API, and cannot be changed in a non-breaking
fashion.
Fixes#29361
FW-1610 #resolve
PR Close#33177
Often the types of an `@Input`'s field don't fully reflect the types of
assignable values. This can happen when an input has a getter/setter pair
where the getter always returns a narrow type, and the setter coerces a
wider value down to the narrow type.
For example, you could imagine an input of the form:
```typescript
@Input() get value(): string {
return this._value;
}
set value(v: {toString(): string}) {
this._value = v.toString();
}
```
Here, the getter always returns a `string`, but the setter accepts any value
that can be `toString()`'d, and coerces it to a string.
Unfortunately TypeScript does not actually support this syntax, and so
Angular users are forced to type their setters as narrowly as the getters,
even though at runtime the coercion works just fine.
To support these kinds of patterns (e.g. as used by Material), this commit
adds a compiler feature called "input coercion". When a binding is made to
the 'value' input of a directive like MatInput, the compiler will look for a
static function with the name ngCoerceInput_value. If such a function is
found, the type-checking expression for the input will be wrapped in a call
to the function, allowing for the expression of a type conversion between
the binding expression and the value being written to the input's field.
To solve the case above, for example, MatInput might write:
```typescript
class MatInput {
// rest of the directive...
static ngCoerceInput_value(value: {toString(): string}): string {
return null!;
}
}
```
FW-1475 #resolve
PR Close#33243
As a hack to get the Ivy compiler ngtsc off the ground, the existing
'allowEmptyCodegenFiles' option was used to control generation of ngfactory
and ngsummary shims during compilation. This option was selected since it's
enabled in google3 but never enabled in external projects.
As ngtsc is now mature and the role shims play in compilation is now better
understood across the ecosystem, this commit introduces two new compiler
options to control shim generation:
* generateNgFactoryShims controls the generation of .ngfactory shims.
* generateNgSummaryShims controls the generation of .ngsummary shims.
The 'allowEmptyCodegenFiles' option is still honored if either of the above
flags are not set explicitly.
PR Close#33256
Currently if a `ModuleWithProviders` is missng its generic type, we throw a cryptic error like:
```
error TS-991010: Value at position 3 in the NgModule.imports of TodosModule is not a reference: [object Object]
```
These changes add a better error to make it easier to debug.
PR Close#33187
Until now, the template type checker has not checked any of the event
bindings that could be present on an element, for example
```
<my-cmp
(changed)="handleChange($event)"
(click)="handleClick($event)"></my-cmp>
```
has two event bindings: the `change` event corresponding with an
`@Output()` on the `my-cmp` component and the `click` DOM event.
This commit adds functionality to the template type checker in order to
type check both kind of event bindings. This means that the correctness
of the bindings expressions, as well as the type of the `$event`
variable will now be taken into account during template type checking.
Resolves FW-1598
PR Close#33125
Prior to this commit, metadata defined on ICU container element was not inherited by the ICU if the whole message is a single ICU (for example: `<ng-container i18n="meaning|description@@id">{count, select, ...}</ng-container>). This commit updates the logic to use parent container i18n meta information for the cases when a message consists of a single ICU.
Fixes#33171
PR Close#33191
Injectable defs are not considered public API, so the property
that contains them should be prefixed with Angular's marker
for "private" ('ɵ') to discourage apps from relying on def
APIs directly.
This commit adds the prefix and shortens the name from
ngInjectableDef to "prov" (for "provider", since injector defs
are known as "inj"). This is because property names cannot
be minified by Uglify without turning on property mangling
(which most apps have turned off) and are thus size-sensitive.
PR Close#33151
Injector defs are not considered public API, so the property
that contains them should be prefixed with Angular's marker
for "private" ('ɵ') to discourage apps from relying on def
APIs directly.
This commit adds the prefix and shortens the name from
ngInjectorDef to inj. This is because property names
cannot be minified by Uglify without turning on property
mangling (which most apps have turned off) and are thus
size-sensitive.
PR Close#33151
The `legacyMessageIdFormat` is taken from the `i18nInFormat` property but we were only considering
`xmb`, `xlf` and `xlf2` values.
The CLI also supports `xliff` and `xliff2` values for the
`i18nInFormat`.
This commit adds support for those aliases.
PR Close#33160
Module defs are not considered public API, so the property
that contains them should be prefixed with Angular's marker
for "private" ('ɵ') to discourage apps from relying on def
APIs directly.
This commit adds the prefix and shortens the name from
ngModuleDef to mod. This is because property names
cannot be minified by Uglify without turning on property
mangling (which most apps have turned off) and are thus
size-sensitive.
PR Close#33142
Pipe defs are not considered public API, so the property
that contains them should be prefixed with Angular's marker
for "private" ('ɵ') to discourage apps from relying on def
APIs directly.
This commit adds the prefix and shortens the name from
ngPipeDef to pipe. This is because property names
cannot be minified by Uglify without turning on property
mangling (which most apps have turned off) and are thus
size-sensitive.
PR Close#33142
Factory defs are not considered public API, so the property
that contains them should be prefixed with Angular's marker
for "private" ('ɵ') to discourage apps from relying on def
APIs directly.
This commit adds the prefix and shortens the name from
ngFactoryDef to fac. This is because property names
cannot be minified by Uglify without turning on property
mangling (which most apps have turned off) and are thus
size-sensitive.
Note that the other "defs" (ngPipeDef, etc) will be
prefixed and shortened in follow-up PRs, in an attempt to
limit how large and conflict-y this change is.
PR Close#33116
Alex Rickabaugh
Kristiyan Kostadinov
Pete Bacon Darwin
Misko Hevery
Keen Yee Liau
George Kalpakas
JoostK
Andrew Scott
Andrew Kushnir
Charles Lyding
Matias Niemelä
Paul Gschwendtner
Igor Minar
Kara Erickson