In View engine it is possible to instantiate a service that that has no
`@Injectable` decorator as long as it satisfies one of:
1) It has no dependencies and so a constructor with no parameters.
This is already supported in Ivy.
2) It has no constructor of its own and sub-classes a service which has
dependencies but has its own `@Injectable` decorator. This second
scenario was broken in Ivy.
In Ivy, previous to this commit, if a class to be instantiated did not have
its own `@Injectable` decorator and did not provide a constructor of
its own, then it would be created using `new` with no arguments -
i.e. falling back to the first scenario.
After this commit Ivy correctly uses the `ngInjectableDef` inherited
from the super-class to provide the `factory` for instantiating the
sub-class.
FW-1314
PR Close#30388
- Moves template ref tests from render3 unit tests to acceptance tests.
- Marks tests testing ivy specific changes as `onlyInIvy`.
- Deletes old render3 unit tests that are no longer necessary
PR Close#30443
PR #30393 corrected behavior where Object.keys sometimes returns an `undefined` value. However, the types didn't reflect this in the code. That fix actually missed one value that could return `undefined`. This PR corrects this by casting the types to what they can be in IE 11. This ensures the code behaves as it should when this edge case comes up.
PR Close#30464
In some cases where multiple navigations happen to the same URL, the router will not process a given URL. In those cases, we fall into logic that resets state for the next navigation. One piece of this resetting is to set the `browserUrlTree` to the most recent `urlAfterRedirects`i.
However, there was bug in this logic because in some cases the `urlAfterRedirects` is a stale value. This happens any time a URL won't be processed, and the previous URL will also not be processed. This creates unpredictable behavior, not the least of which ends up being a broken `back` button.
This PR kicks off new navigations with the current value the router assumes is in the browser. All the logic around how to handle future navigations is based on this value compared to the current transition, so it's important to kick off all new navigations with the current value so in the edge case described above we don't end up with an old value being set into `browserUrlTree`.
Fixes#30340
Related to #30160
PR Close#30344
Slightly improves the messages for the static-query migration in order
to make the terminal output less verbose but more helpful. Unfortunately
we are limited in what we can print due to the devkit not providing much
utilities for printing good messages from a migration schematic.
PR Close#30458
Currently if something fails in the selected strategy (e.g. AOT failures),
the migration currently accidentally falls back to the test strategy. This
is not helpful as we want to give developers the possibility to re-run
the migration after fixing potential AOT failures.
PR Close#30458
Apparently the devkit logger is not able to properly print
out error objects, so we need to convert them to a string
before in order to make the error visible to the user.
This is not testable without an e2e test that validates the CLI
terminal output.
PR Close#30458
Preserve compatibility with rollup_bundle rule.
Add missing npm dependencies, which are now enforced by the strict_deps plugin in tsc_wrapped
PR Close#30370
Moves most of the r3 change detection tests into `acceptance`. Notes:
* A handful of tests weren't migrated, because they were testing an API that isn't exposed publicly yet.
* The `should throw if bindings in children of current view have changed` and `should NOT throw if bindings in ancestors of current view have changed` tests were removed, because there's not nice way of hitting the same code path with `TestBed` and doing the same assertion as with the raw instructions. I'm open to ideas on how we could do them.
* There were a few tests that assert that the `innerHTML` looks in a particular way. I've switched them to use `textContent`, because Ivy and ViewEngine produce slightly different DOM. The tests were only checking whether the text has changed anyway.
PR Close#30425
Moves all manual render3 property binding tests to
TestBed acceptance tests. Unfortunately three property
binding tests could not be migrated as these rely on
manual Ivy template code that is not supported within
TestBed. These can be revisited as discussed in the
framework team meeting.
PR Close#30426
Moves all manual render3 tests which are located within the
`renderer_factory_spec.ts` file to acceptance tests. A few tests
that use Ivy-specific logic which is not replicable with `TestBed`
remain in the render3 folder (e.g. using `renderTemplate`)
Additionally migrated tests that assert the lifecycles of the
renderer_factory are set to *ivy only* as the lifecycle seems
to be different in Ivy. Tracked with: FW-1320
PR Close#30435
This PR fixes an issue where IE 11 can return `undefined` in with an `Object.keys` call. Solution is to add a runtime check on the value. Based on the types being passed, this shouldn't be necessary, but is needed only for IE 11. Unit test doesn't work for this PR because it can't be replicated easily.
PR Close#30393
At the moment the module resolver will end up in an infinite loop in Windows because we are assuming that the root directory is always `/` however in windows this can be any drive letter example `c:/` or `d:/` etc...
With this change we also resolve the drive letter in windows, when using `AbsoluteFsPath.from` for consistence so under `/foo` will be converted to `c:/foo` this is also needed because of relative paths with different drive letters.
PR Close#30297
Currently in Ivy `NgModule` registration happens when the class is declared, however this is inconsistent with ViewEngine and requires extra generated code. These changes remove the generated code for `registerModuleFactory`, pass the id through to the `ngModuleDef` and do the module registration inside `NgModuleFactory.create`.
This PR resolves FW-1285.
PR Close#30244
```
//packages/compiler-cli/test:ngc
//packages/compiler/test:test
```
This also address `node_modules` to the ignored paths for ngc compiler as otherwise the `ready` is never fired
Partially addresses #29785
PR Close#30146
- Moves tests related to the pureFunction instructions from render3 to acceptance tests
- Leaves behind one test for in-template javascript that is not supported syntax yet
PR Close#30406
Moves over the tests from `pipe_spec` into `acceptance`. Note that the two `WrappedValue` tests haven't been moved over, because impure pipes always throw "changed after checked" errors in `TestBed`. This seems to be consistent with ViewEngine.
PR Close#30389
Now that the dependent files and compilation scopes are being tracked in
the incremental state, we can skip analysing and emitting source files if
none of their dependent files have changed since the last compile.
The computation of what files (and their dependencies) are unchanged is
computed during reconciliation.
This commit also removes the previous emission skipping logic, since this
approach covers those cases already.
PR Close#30238
To support skipping analysis of a file containing a component
we need to know that none of the declarations that might affect
its ngtsc compilation have not changed. The files that we need to
check are those that contain classes from the `CompilationScope`
of the component. These classes are already tracked in the
`LocalModuleScopeRegistry`.
This commit modifies the `IvyCompilation` class to record the
files that are in each declared class's `CompilationScope` via
a new method, `recordNgModuleScopeDependencies()`, that is called
after all the handlers have been "resolved".
Further, if analysis is skipped for a declared class, then we need
to recover the analysis from the previous compilation run. To
support this, the `IncrementalState` class has been updated to
expose the `MetadataReader` and `MetadataRegistry` interfaces.
This is included in the `metaRegistry` object to capture these analyses,
and also in the `localMetaReader` as a fallback to use if the
current compilation analysis was skipped.
PR Close#30238
As part of incremental compilation performance improvements, we need
to track the dependencies of files due to expressions being evaluated by
the `PartialEvaluator`.
The `PartialEvaluator` now accepts a `DependencyTracker` object, which is
used to track which files are visited when evaluating an expression.
The interpreter computes this `originatingFile` and stores it in the evaluation
`Context` so it can pass this to the `DependencyTracker.
The `IncrementalState` object implements this interface, which allows it to be
passed to the `PartialEvaluator` and so capture the file dependencies.
PR Close#30238
Moves most of the tests in `output_spec` into `acceptance`. Note that one test is left in `render3`, because it's testing mixing in custom logic with instructions which we can't replicate using `TestBed`.
PR Close#30372
Moves all manual `render3` content projection tests that use
the `ɵɵelementProperty` to acceptance tests. Additionally a
few other content projection tests were moved over to
acceptance. Eventually we'll be able to move all remaining
content projection tests to acceptance.
PR Close#30357
This commit moves the delegated constructor detection to a helper
function and also adds more test coverage.
The original code for this came from https://github.com/angular/angular/pull/24156
thanks to @ts2do.
Closes#24156Closes#27267
// FW-1310
PR Close#30368
After this PR is merged, maintainers no longer need to update .bazelrc
file, toolchain and platform related flags for RBE builds and tests
(unless there is a breaking change in Bazel related to those flags).
Maintainers just need to update the pin of @bazel-toolchains repo
regularly in the packages/bazel/package.bzl file according to
https://releases.bazel.build/bazel-toolchains.html to include the
latest checked-in toolchain configs. If rbe_autoconfig() cannot find
appropriate toolchain configs for the version of Bazel in the version of
@bazel_toolchains repo that is currently used by this project, it will pull
down the container and generate the configs on the fly as the beginning
of the build/test.
PR Close#29336
Currently we always just set the timing to `false` if we aren't
able to analyze a given call expression or new expression. e.g.
```ts
ngOnInit() {
thirdPartyCallSync(() => this.query.doSomething())
}
```
In that case the `thirdPartyCallSync` function comes from the `node_modules`
and is only defined through types while there is no code for the
actual function logic that can be analyzed. This makes it impossible
to tell whether the given call expression actually causes the specified
arrow function to be executed synchronously or not. In order to be able
to make this better, we now peek into the passed arrow function and
check for a synchronous query usage. If so, we set the query timing to
static and mark it as ambiguous. This ensures that the usage strategy is
less "magical" and more correct with third-party code.
Additionally since functions like `setTimeout` are not analyzable but known
to be asynchronous, there is a hard-coded list of known functions which
shouldn't be marked as ambiguous.
Resolves FW-1214. As planned within https://hackmd.io/hPiLWpPlQ4uynC1luIBdfQ
PR Close#30215
Even though we don't run the size-tracking test on CI
right now, we update the golden size map as part of
the size-tracking tool update. The size-map difference
should generally be kept up-to-date to be able to
determine which PRs contribute size to a given file.
PR Close#30257
Based on discussion that happened on the PR that introduced
the size-tracking tool, we want to have another threshold for
the raw byte difference. This allows us to better control for
which changes the size-tracking tool should report a difference.
See: https://github.com/angular/angular/pull/30070#discussion_r278332315
PR Close#30257
- splits existing property acceptance tests into property_binding and property_interpolation
- ports tests from render3 instructions tests to acceptance tests
- removes redundant or unnecessary tests that are covered by existing acceptance tests
:)
PR Close#30321
Currently the static-query migration ignores queries declared on getters
or setters as these are not part of a `PropertyDeclaration`. We need to
handle these queries in order to cover all queries within a given project.
The usage strategy is not able to detect timing for queries on accessors,
so we add a TODO and print a message. The template strategy is able
to detect the proper timing for such queries because it's not dependent
on detecting the usage of the query.
Resolves FW-1215
PR Close#30327
We have an issue where we would like to be able to test perf counter metrics in acceptance tests, but we are unable to do so, because it will break when those same tests are run with ViewEngine. This PR adds a testing utility to `onlyInIvy` that allows for testing of performance counters, and even gives readable errors for what value on `ngDevMode` is incorrect. Has typings for decent autocompletion as well.
PR Close#30339
This is the final patch to migrate the Angular styling code to have a
smaller instruction set in preparation for the runtime refactor. All
styling-related instructions now work both in template and hostBindings
functions and do not use `element` as a prefix for their names:
BEFORE:
elementStyling()
elementStyleProp()
elementClassProp()
elementStyleMap()
elementClassMap()
elementStylingApply()
AFTER:
styling()
styleProp()
classProp()
styleMap()
classMap()
stylingApply()
PR Close#30318
This patch removes all host-specific styling instructions in favor of
using element-level instructions instead. Because of the previous
patches that made sure `select(n)` worked between styling calls, all
host level instructions are not needed anymore. This patch changes each
of those instruction calls to use any of the `elementStyling*`,
`elementStyle*` and `elementClass*` styling instructions instead.
PR Close#30336
The static-query template strategy leverages the AOT compiler
in order to determine the query timing. Unfortunately the AOT
compiler has open bugs that can cause unexpected failures which
make the template strategy unusable in rare cases. These rare
exceptions need to be handled gracefully in order to avoid confusion
and to provide a more smooth migration.
Additionally migration strategy setup failures are now reported with
stack traces as the `ng update` command does not print stack traces.
This makes it easier to reproduce and report migration issues.
PR Close#30269
In an Angular CLI project scenario where projects only reference
top-level source-files through the `tsconfig` `files` option, we currently
do not migrate referenced source-files. This can be fixed checking all
referenced source-files which aren't coming from an external library.
This is similar to how `tslint` determines project source-files.
PR Close#30269
Currently when someone has a call expression within the `ngOnInit` call
and we try to peek into that function with respect to the current function
context, the schematic errors because a call expression argument is
undefined. This is valid because the target function declaration defines
that parameter with a default value. In order to fix this, we need to
respect parameter default values.
PR Close#30269
Fixes not being able to bind a `SafeStyle` as a camel cased style property (e.g. `[style.backgroundImage]="someSafeStyle"`). The issue was due to the fact that we only check the dash case property names to determine whether to sanitize a value.
This PR resolves FW-1279.
PR Close#30328
572b54967c changed how the schematic
tests are executed. Tests no longer use the schematic collection
that is also used by the CLI `ng update` command and therefore
the migration collection could technically be invalid.
In order to ensure that the public migration collection is guaranteed
to work and to avoid duplication within two schematic collections, the
changes are partially reverted and only the disabled `injectable-pipe`
schematic has its own collection.
PR Close#30198
Currently we always prompt when the static-query migration runs. This is not
always needed because some applications do not even use `ViewChild` or
`ContentChild` queries and it just causes confusion if developers need to
decide on a migration strategy while there is nothing to migrate.
In order to avoid this confusion, we no longer prompt for a strategy
if there are no queries declared within the project.
PR Close#30254
This patch is one commit of many patches that will unify all styling instructions
across both template-level bindings and host-level bindings. This patch in particular
removes the `elementIndex` param because it is already set prior to each styling
instruction via the `select(n)` instruction.
PR Close#30313
Instead of launching a Node.js process that in turn spawns Bazel binary,
the Builder could now directly spawn the native binary. This makes the
bootup process slightly more efficient, and allows the Builder to
control spawn options. This works with both Bazel and iBazel.
PR Close#30306
Currently, in jit mode, `ngInjectableDef`, `ngDirectiveDef`, `ngPipeDef` and `ngModuleDef` use `ng://`,
which display them in the top domain in Chrome Dev Tools, whereas `ngComponentDef` uses `ng:///` which display components in a separate domain.
You can currently see:
```
AppModule
UserService
ng://
|_ AppComponent
|_ template.html
|_ AppComponent.js
...
```
This commits replaces all `ng://` with `ng:///` to display every Angular entity in the `ng://` domain.
```
ng://
|_ AppModule
|_ UserService
|_ AppComponent
...
```
PR Close#29826
Prior to this patch, the `select(n)` instruction would only be generated
when property bindings are encountered which meant that styling-related
bindings were skipped. This patch ensures that all styling-related bindings
(i.e. class and style bindings) are always prepended with a `select()`
instruction prior to being generated in AOT.
PR Close#30311
This patch breaks up the existing `elementStylingMap` into
`elementClassMap` and `elementStyleMap` instructions. It also breaks
apart `hostStlyingMap` into `hostClassMap` and `hostStyleMap`
instructions. This change allows for better tree-shaking and reduces
the complexity of the styling algorithm code for `[style]` and `[class]`
bindings.
PR Close#30293
If an entry-point has a missing dependency then all the entry-points
that would have pointed to that dependency are also removed from
the dependency graph.
Previously we were still processing the dependencies of an entry-point
even if it had already been removed from the graph because it depended
upon a missing dependency that had previously been removed due to another
entry-point depending upon it.
This caused the dependency processing to crash rather than gracefully
logging and handling the missing invalid entry-point.
Fixes#29624
PR Close#30270
There was a problem with a combination of the `eager` URL update, browser `back` button, and hybrid applications. Details provided in internal ticket http://b/123667227.
This fix handles the problem by setting `router.browserUrlTree` when all conditions have failed, meaning the browser doesn't do anything with the navigation other than update internal data structures. Without this change, the problem was an old value was stored in `router.broserUrlTree` causing some new navigations to be compared to an old value and breaking future navigations.
PR Close#30160
In the existing implementation the `elementPropertyInternal` function (meant to
set element properties) was executed even if a bound value didn't change. The
`elementPropertyInternal` was inspecting the incoming value and after comparing it
to `NO_CHANGE` - exiting early. All in all it meant that we were unnecessarily
invoking the `elementPropertyInternal` function for cases where bound value didn't
change.
Based on my bencharks (running change detection without any model update in a tight
loop) this unnecessary function call was causing ~5% slowdown in the change detection
process.
PR Close#30255
This commit fixes a regression introduced in PR 29692 where
the interpolate symbol in View Engine was improperly prefixed
with the ɵɵ that signifies private instructions for Ivy. It
resulted in interpolations of 10+ values not working correctly
in AOT mode. This commit removes the prefix.
PR Close#30243
Previously, `R3TestBedCompiler` was dynamically defining an
`@NgModule`-decorated `CompilerModule` class inside a method call.
Since ngcc only processes top-level classes, this class was not
transformed causing failures in unit tests (see #30121 for details).
This commit fixes it by using `compileNgModuleDefs()` directly (similar
to the fix in #30037).
Fixes#30121
PR Close#28530
Sometimes we need to override module resolution behaviour.
We do this by implementing the optional method `resolveModuleNames()`
on `CompilerHost`.
This commit ensures that we always try this method first before falling
back to the standard `ts.resolveModuleName`
PR Close#30017
Packages that do not follow APF may have the declaration files in the same
directory as one source format, typically ES5. This is problematic for ngcc,
as it needs to create a TypeScript program with all JavaScript sources of
an entry-point, whereas TypeScript's module resolution mechanism would have
resolved an internal module import to the external facing .d.ts declaration
file, instead of the JavaScript source file. This behavior results in the
program to be analysed being incomplete.
This commit introduces a custom compiler host that recognizes the above
scenario and rewires the resolution of a .d.ts declaration file to its
JavaScript counterpart, if applicable.
Fixes#29939
PR Close#30017
Ivy uses R3Injector, but we are currently pulling in both the StaticInjector
(View Engine injector) and the R3Injector when running with Ivy. This commit
adds an ivy switch so calling Injector.create() pulls in the correct
implementation of the injector depending on whether you are using VE or Ivy.
This saves us about 3KB in the bundle.
PR Close#30219
Previously, we were supporting injection flags for provider deps, but only
if they fit the format `new Optional()`. This commit fixes resolution of
provider deps to also support `Optional` (without the new). This keeps us
backwards compatible with what View Engine supported.
PR Close#30216
Stores the views that are part of a container directly on the `LContainer`, rather than maintaining a dedicated sub-array.
This PR resolves FW-1288.
PR Close#30179
Currently, we are not properly resolving forward refs when they appear
in deps for providers created with the useFactory strategy. This commit
wraps provider deps in the resolveForwardRef call so the tokens are
passed into the inject function as expected.
PR Close#30201
Fixes `HostBinding` and `HostListener` declarations not being inherited from base classes that don't have an Angular decorator.
This PR resolves FW-1275.
PR Close#30158
Currently the `static-query` migrations fails at the final step of
updating a query when the query already specifies options which
cannot be transformed easily. e.g. the options are computed through
a function call: `@ViewChild(..., getQueryOpts());` or `@ViewChild(..., myOptionsVar)`.
In these cases we technically could add additionally logic to update
the query options, but given that this is an edge-case and it's
potentially over-engineering the migration schematic, we just
always add a TODO for the timing and print out the determined
query timing in the console. The developer in that case just needs
to manually update the logic for the query options to contain the
printed query timing.
Potentially related to: https://github.com/angular/angular-cli/issues/14298
PR Close#30178
This PR adds the implementation for `definitionAndBoundSpan` because
it's now preferred over `definition`. vscode would send this new request
for `Go to definition`. As part of this PR the implementation for
`definition` is refactored and simplified. Goldens for both methods are
checked in.
PR Close#30125
This commit introduces a new interface, which abstracts access
to the underlying `FileSystem`. There is initially one concrete
implementation, `NodeJsFileSystem`, which is simply wrapping the
`fs` library of NodeJs.
Going forward, we can provide a `MockFileSystem` for test, which
should allow us to stop using `mock-fs` for most of the unit tests.
We could also implement a `CachedFileSystem` that may improve the
performance of ngcc.
PR Close#29643
Pete Bacon Darwin
Ben Lesh
Jason Aden
Alex Rickabaugh
Paul Gschwendtner
cexbrayat
Keen Yee Liau
Alan
Alex Eagle
Kristiyan Kostadinov
Olivier Combe
JiaLiPassion
Xin Gao
Piotr Tomiak
Ly-lns
Matias Niemelä
Frederik Prijck
Pawel Kozlowski
likui
benbot1
Vinit Tomar
Filipe Silva
Kara Erickson
George Kalpakas
JoostK