Reworks the compiler to output the factories for directives, components and pipes under a new static field called `ngFactoryFn`, instead of the usual `factory` property in their respective defs. This should eventually allow us to inject any kind of decorated class (e.g. a pipe).
**Note:** these changes are the first part of the refactor and they don't include injectables. I decided to leave injectables for a follow-up PR, because there's some more cases we need to handle when it comes to their factories. Furthermore, directives, components and pipes make up most of the compiler output tests that need to be refactored and it'll make follow-up PRs easier to review if the tests are cleaned up now.
This is part of the larger refactor for FW-1468.
PR Close#31953
When ngcc is called for a specific entry-point, it has to determine
which dependencies to transitively process. To accomplish this, ngcc
traverses the full import graph of the entry-points it encounters, for
which it uses a dependency host to find all module imports. Since
imports look different in the various bundle formats ngcc supports, a
specific dependency host is used depending on the information provided
in an entry-points `package.json` file. If there's not enough
information in the `package.json` file for ngcc to be able to determine
which dependency host to use, ngcc would fail with an error.
If, however, the entry-point is not compiled by Angular, it is not
necessary to process any of its dependencies. None of them can have
been compiled by Angular so ngcc does not need to know about them.
Therefore, this commit changes the behavior to avoid recursing into
dependencies of entry-points that are not compiled by Angular.
In particular, this fixes an issue for packages that have dependencies
on the `date-fns` package. This package has various secondary
entry-points that have a `package.json` file only containing a `typings`
field, without providing additional fields for ngcc to know which
dependency host to use. By not needing a dependency host at all, the
error is avoided.
Fixes#32302
PR Close#32303
In ngc is was valid to set the "flatModuleOutFile" option to "null". This is sometimes
necessary if a tsconfig extends from another one but the "fatModuleOutFile" option
needs to be unset (note that "undefined" does not exist as value in JSON)
Now if ngtsc is used to compile the project, ngtsc will fail with an error because it
tries to do string manipulation on the "flatModuleOutFile". This happens because
ngtsc only skips flat module indices if the option is set to "undefined".
Since this is not compatible with what was supported in ngc and such exceptions
should be avoided, the flat module check is now aligned with ngc.
```
TypeError: Cannot read property 'replace' of null
at Object.normalizeSeparators (/home/circleci/project/node_modules/@angular/compiler-cli/src/ngtsc/util/src/path.js:35:21)
at new NgtscProgram (/home/circleci/project/node_modules/@angular/compiler-cli/src/ngtsc/program.js:126:52)
```
Additionally setting the `flatModuleOutFile` option to an empty string
currently results in unexpected behavior. No errors is thrown, but the
flat module index file will be `.ts` (no file name; just extension).
This is now also fixed by treating an empty string similarly to
`null`.
PR Close#32235
Previously, ngtsc attempted to use the .d.ts schema for HTML elements to
check bindings to DOM properties. However, the TypeScript lib.dom.d.ts
schema does not perfectly align with the Angular DomElementSchemaRegistry,
and these inconsistencies would cause issues in apps. There is also the
concern of supporting both CUSTOM_ELEMENTS_SCHEMA and NO_ERRORS_SCHEMA which
would have been very difficult to do in the existing system.
With this commit, the DomElementSchemaRegistry is employed in ngtsc to check
bindings to the DOM. Previous work on producing template diagnostics is used
to support generation of this different kind of error with the same high
quality of error message.
PR Close#32171
Historically, the Angular Compiler has produced both native TypeScript
diagnostics (called ts.Diagnostics) and its own internal Diagnostic format
(called an api.Diagnostic). This was done because TypeScript ts.Diagnostics
cannot be produced for files not in the ts.Program, and template type-
checking diagnostics are naturally produced for external .html template
files.
This design isn't optimal for several reasons:
1) Downstream tooling (such as the CLI) must support multiple formats of
diagnostics, adding to the maintenance burden.
2) ts.Diagnostics have gotten a lot better in recent releases, with support
for suggested changes, highlighting of the code in question, etc. None of
these changes have been of any benefit for api.Diagnostics, which have
continued to be reported in a very primitive fashion.
3) A future plugin model will not support anything but ts.Diagnostics, so
generating api.Diagnostics is a blocker for ngtsc-as-a-plugin.
4) The split complicates both the typings and the testing of ngtsc.
To fix this issue, this commit changes template type-checking to produce
ts.Diagnostics instead. Instead of reporting a special kind of diagnostic
for external template files, errors in a template are always reported in
a ts.Diagnostic that highlights the portion of the template which contains
the error. When this template text is distinct from the source .ts file
(for example, when the template is parsed from an external resource file),
additional contextual information links the error back to the originating
component.
A template error can thus be reported in 3 separate ways, depending on how
the template was configured:
1) For inline template strings which can be directly mapped to offsets in
the TS code, ts.Diagnostics point to real ranges in the source.
This is the case if an inline template is used with a string literal or a
"no-substitution" string. For example:
```typescript
@Component({..., template: `
<p>Bar: {{baz}}</p>
`})
export class TestCmp {
bar: string;
}
```
The above template contains an error (no 'baz' property of `TestCmp`). The
error produced by TS will look like:
```
<p>Bar: {{baz}}</p>
~~~
test.ts:2:11 - error TS2339: Property 'baz' does not exist on type 'TestCmp'. Did you mean 'bar'?
```
2) For template strings which cannot be directly mapped to offsets in the
TS code, a logical offset into the template string will be included in
the error message. For example:
```typescript
const SOME_TEMPLATE = '<p>Bar: {{baz}}</p>';
@Component({..., template: SOME_TEMPLATE})
export class TestCmp {
bar: string;
}
```
Because the template is a reference to another variable and is not an
inline string constant, the compiler will not be able to use "absolute"
positions when parsing the template. As a result, errors will report logical
offsets into the template string:
```
<p>Bar: {{baz}}</p>
~~~
test.ts (TestCmp template):2:15 - error TS2339: Property 'baz' does not exist on type 'TestCmp'.
test.ts:3:28
@Component({..., template: TEMPLATE})
~~~~~~~~
Error occurs in the template of component TestCmp.
```
This error message uses logical offsets into the template string, and also
gives a reference to the `TEMPLATE` expression from which the template was
parsed. This helps in locating the component which contains the error.
3) For external templates (templateUrl), the error message is delivered
within the HTML template file (testcmp.html) instead, and additional
information contextualizes the error on the templateUrl expression from
which the template file was determined:
```
<p>Bar: {{baz}}</p>
~~~
testcmp.html:2:15 - error TS2339: Property 'baz' does not exist on type 'TestCmp'.
test.ts:10:31
@Component({..., templateUrl: './testcmp.html'})
~~~~~~~~~~~~~~~~
Error occurs in the template of component TestCmp.
```
PR Close#31952
When a template contains a binding without a value, the template parser
creates an `EmptyExpr` node. This would previously be translated into
an `undefined` value, which would cause a crash downstream as `undefined`
is not included in the allowed type, so it was not handled properly.
This commit prevents the crash by returning an actual expression for empty
bindings.
Fixes#30076Fixes#30929
PR Close#31594
This commit switches the default value of the enableIvy flag to true.
Applications that run ngc will now by default receive an Ivy build!
This does not affect the way Bazel builds in the Angular repo work, since
those are still switched based on the value of the --define=compile flag.
Additionally, projects using @angular/bazel still use View Engine builds
by default.
Since most of the Angular repo tests are still written against View Engine
(particularly because we still publish VE packages to NPM), this switch
also requires lots of `enableIvy: false` flags in tsconfigs throughout the
repo.
Congrats to the team for reaching this milestone!
PR Close#32219
This option makes ngc behave as tsc, and was originally implemented before
ngtsc existed. It was designed so we could build JIT-only versions of
Angular packages to begin testing Ivy early, and is not used at all in our
current setup.
PR Close#32219
Previously, `ngcc` assumed that if a format property was defined in
`package.json` it would point to a valid format-path (i.e. a file that
is an entry-point for a specific format). This is generally the case,
except if a format property is set to a non-string value (such as
`package.json`) - either directly in the `package.json` (which is unusual)
or in ngcc.config.js (which is a valid usecase, when one wants a
format property to be ignored by `ngcc`).
For example, the following config file would cause `ngcc` to throw:
```
module.exports = {
packages: {
'test-package': {
entryPoints: {
'.': {
override: {
fesm2015: undefined,
},
},
},
},
},
};
```
This commit fixes it by ensuring that only format properties whose value
is a string are considered by `ngcc`.
For reference, this regression was introduced in #32052.
Fixes#32188
PR Close#32205
During the dependency analysis phase of ngcc, imports are resolved to
files on disk according to certain module resolution rules. Since module
specifiers are typically missing extensions, or can refer to index.js
barrel files within a directory, the module resolver attempts several
postfixes when searching for a module import on disk. Module specifiers
that already include an extension, however, would fail to be resolved as
ngcc's module resolver failed to check the location on disk without
adding any postfixes.
Closes#32097
PR Close#32181
During the recursive processing of dependencies, ngcc resolves the
requested file to an actual location on disk, by testing various
extensions. For recursive calls however, the path is known to have been
resolved in the module resolver. Therefore, it is safe to move the path
resolution to the initial caller into the recursive process.
Note that this is not expected to improve the performance of ngcc, as
the call to `resolveFileWithPostfixes` is known to succeed immediately,
as the provided path is known to exist without needing to add any
postfixes. Furthermore, the FileSystem caches whether files exist, so
the additional check that we used to do was cheap.
PR Close#32181
ngcc needs to solve a unique problem when compiling typings for an
entrypoint: it must resolve a declaration within a .js file to its
representation in a .d.ts file. Since such .d.ts files can be used in deep
imports without ever being referenced from the "root" .d.ts, it's not enough
to simply match exported types to the root .d.ts. ngcc must build an index
of all .d.ts files.
Previously, this operation had a bug: it scanned all .d.ts files in the
.d.ts program, not only those within the package. Thus, if a class in the
program happened to share a name with a class exported from a dependency's
.d.ts, ngcc might accidentally modify the wrong .d.ts file, causing a
variety of issues downstream.
To fix this issue, ngcc's .d.ts scanner now limits the .d.ts files it
indexes to only those declared in the current package.
PR Close#32129
One of the compiler's tasks is to enumerate the exports of a given ES
module. This can happen for example to resolve `foo.bar` where `foo` is a
namespace import:
```typescript
import * as foo from './foo';
@NgModule({
directives: [foo.DIRECTIVES],
})
```
In this case, the compiler must enumerate the exports of `foo.ts` in order
to evaluate the expression `foo.DIRECTIVES`.
When this operation occurs under ngcc, it must deal with the different
module formats and types of exports that occur. In commonjs code, a problem
arises when certain exports are downleveled.
```typescript
export const DIRECTIVES = [
FooDir,
BarDir,
];
```
can be downleveled to:
```javascript
exports.DIRECTIVES = [
FooDir,
BarDir,
```
Previously, ngtsc and ngcc expected that any export would have an associated
`ts.Declaration` node. `export class`, `export function`, etc. all retain
`ts.Declaration`s even when downleveled. But the `export const` construct
above does not. Therefore, ngcc would not detect `DIRECTIVES` as an export
of `foo.ts`, and the evaluation of `foo.DIRECTIVES` would therefore fail.
To solve this problem, the core concept of an exported `Declaration`
according to the `ReflectionHost` API is split into a `ConcreteDeclaration`
which has a `ts.Declaration`, and an `InlineDeclaration` which instead has
a `ts.Expression`. Differentiating between these allows ngcc to return an
`InlineDeclaration` for `DIRECTIVES` and correctly keep track of this
export.
PR Close#32129
In Angular today, the following pattern works:
```typescript
export class BaseDir {
constructor(@Inject(ViewContainerRef) protected vcr: ViewContainerRef) {}
}
@Directive({
selector: '[child]',
})
export class ChildDir extends BaseDir {
// constructor inherited from BaseDir
}
```
A decorated child class can inherit a constructor from an undecorated base
class, so long as the base class has metadata of its own (for JIT mode).
This pattern works regardless of metadata in AOT.
In Angular Ivy, this pattern does not work: without the @Directive
annotation identifying the base class as a directive, information about its
constructor parameters will not be captured by the Ivy compiler. This is a
result of Ivy's locality principle, which is the basis behind a number of
compilation optimizations.
As a solution, @Directive() without a selector will be interpreted as a
"directive base class" annotation. Such a directive cannot be declared in an
NgModule, but can be inherited from. To implement this, a few changes are
made to the ngc compiler:
* the error for a selector-less directive is now generated when an NgModule
declaring it is processed, not when the directive itself is processed.
* selector-less directives are not tracked along with other directives in
the compiler, preventing other errors (like their absence in an NgModule)
from being generated from them.
PR Close#31379
The language service relies on a "context" file that is used as the
canonical "containing file" when performing module resolution.
This file is unnecessary since the language service host's current
directory always default to the location of tsconfig.json for the
project, which would give the correct result.
This refactoring allows us to simplify the "typescript host" and also
removes the need for custom logic to find tsconfig.json.
PR Close#32015
Similar to interpolation, we do not want to completely remove whitespace
nodes that are siblings of an expansion.
For example, the following template
```html
<div>
<strong>items left<strong> {count, plural, =1 {item} other {items}}
</div>
```
was being collapsed to
```html
<div><strong>items left<strong>{count, plural, =1 {item} other {items}}</div>
```
which results in the text looking like
```
items left4
```
instead it should be collapsed to
```html
<div><strong>items left<strong> {count, plural, =1 {item} other {items}}</div>
```
which results in the text looking like
```
items left 4
```
---
**Analysis of the code and manual testing has shown that this does not cause
the generated ids to change, so there is no breaking change here.**
PR Close#31962
Previously if only a component template changed then we would know to
rebuild its component source file. But the compilation was incorrect if the
component was part of an NgModule, since we were not capturing the
compilation scope information that had a been acquired from the NgModule
and was not being regenerated since we were not needing to recompile
the NgModule.
Now we register compilation scope information for each component, via the
`ComponentScopeRegistry` interface, so that it is available for incremental
compilation.
The `ComponentDecoratorHandler` now reads the compilation scope from a
`ComponentScopeReader` interface which is implemented as a compound
reader composed of the original `LocalModuleScopeRegistry` and the
`IncrementalState`.
Fixes#31654
PR Close#31932
In Angular today, the following pattern works:
```typescript
export class BaseDir {
constructor(@Inject(ViewContainerRef) protected vcr: ViewContainerRef) {}
}
@Directive({
selector: '[child]',
})
export class ChildDir extends BaseDir {
// constructor inherited from BaseDir
}
```
A decorated child class can inherit a constructor from an undecorated base
class, so long as the base class has metadata of its own (for JIT mode).
This pattern works regardless of metadata in AOT.
In Angular Ivy, this pattern does not work: without the @Directive
annotation identifying the base class as a directive, information about its
constructor parameters will not be captured by the Ivy compiler. This is a
result of Ivy's locality principle, which is the basis behind a number of
compilation optimizations.
As a solution, @Directive() without a selector will be interpreted as a
"directive base class" annotation. Such a directive cannot be declared in an
NgModule, but can be inherited from. To implement this, a few changes are
made to the ngc compiler:
* the error for a selector-less directive is now generated when an NgModule
declaring it is processed, not when the directive itself is processed.
* selector-less directives are not tracked along with other directives in
the compiler, preventing other errors (like their absence in an NgModule)
from being generated from them.
PR Close#31379
Publishing of NGCC packages should not be allowed. It is easy for a user to publish an NGCC'd version of a library they have workspace libraries which are being used in a workspace application.
If a users builds a library and afterwards the application, the library will be transformed with NGCC and since NGCC taints the distributed files that should be published.
With this change we use the npm/yarn `prepublishOnly` hook to display and error and abort the process with a non zero error code when a user tries to publish an NGCC version of the package.
More info: https://docs.npmjs.com/misc/scripts
PR Close#32031
Previously, when run with `createNewEntryPointFormats: true`, `ngcc`
would only update `package.json` with the new entry-point for the first
format property that mapped to a format-path. Subsequent properties
mapping to the same format-path would be detected as processed and not
have their new entry-point format recorded in `package.json`.
This commit fixes this by ensuring `package.json` is updated for all
matching format properties, when writing an `EntryPointBundle`.
PR Close#32052
Remove the `formatProperty` property from the `EntryPointBundle`
interface, because the property is not directly related to that type.
It was only used in one place, when calling `fileWriter.writeBundle()`,
but we can pass `formatProperty` directrly to `writeBundle()`.
PR Close#32052
This refactoring more clearly separates the different phases of the work
performed by `ngcc`, setting the ground for being able to run each phase
independently in the future and improve performance via parallelization.
Inspired by/Based on @alxhub's prototype: alxhub/angular@cb631bdb1
PR Close#32052
This change basically moves some checks to happen up front and ensures
we don't try to process any more properties than we absolutely need.
(The properties would not be processed before either, but we would
consider them, before finding out that they have already been processed
or that they do not exist in the entry-point's `package.json`.)
This change should make no difference in the work done by `ngcc`, but it
transforms the code in a way that makes the actual work known earlier,
thus making it easier to parallelize the processing of each property in
the future.
PR Close#32052
Kristiyan Kostadinov
Kara Erickson
JoostK
Paul Gschwendtner
Alex Rickabaugh
Alan
atscott
Elvis Begovic
Keen Yee Liau
Pete Bacon Darwin
George Kalpakas