The cached file-system was implemented to speed up ngcc
processing, but in reality most files are not accessed many times
and there is no noticeable degradation in speed by removing it.
Benchmarking `ngcc -l debug` for AIO on a local machine
gave a range of 196-236 seconds with the cache and 197-224
seconds without the cache.
Moreover, when running in parallel mode, ngcc has a separate
file cache for each process. This results in excess memory usage.
Notably the master process, which only does analysis of entry-points
holds on to up to 500Mb for AIO when using the cache compared to
only around 30Mb when not using the cache.
Finally, the file-system cache being incorrectly primed with file
contents before being processed has been the cause of a number
of bugs. For example https://github.com/angular/angular-cli/issues/16860#issuecomment-614694269.
PR Close#36687
Currently destroy hooks are stored in memory as `[1, hook, 5, hook]` where
the numbers represent the index at which to find the context and `hook` is
the function to be invoked. This breaks down for `multi` providers,
because the value at the index will be an array of providers, resulting in
the hook being invoked with an array of all the multi provider values,
rather than the provider that was destroyed. In ViewEngine `ngOnDestroy`
wasn't being called for `multi` providers at all.
These changes fix the issue by changing the structure of the destroy hooks to `[1, hook, 5, [0, hook, 3, hook]]` where the indexes inside the inner array point to the provider inside of the multi provider array. Note that this is slightly different from the original design which called for the structure to be `[1, hook, 5, [hook, hook]`, because in the process of implementing it, I realized that we wouldn't get passing the correct context if only some of the `multi` providers have `ngOnDestroy` and others don't.
I've run the newly-added `view_destroy_hooks` benchmark against these changes and compared it to master. The difference seems to be insignificant (between 1% and 2% slower).
Fixes#35231.
PR Close#35840
Rebuild the yarn lock file from scratch to collapse instances where
one package is able to satisfy multiple dependencies. Currently we
have some situations where we have multiple versions when one would
work.
Example:
```
"@babel/code-frame@^7.0.0":
version "7.0.0"
resolved "https://registry.yarnpkg.com/@babel/cod
integrity sha512-OfC2uemaknXr87bdLUkWog7nYuliM9Ij
dependencies:
"@babel/highlight" "^7.0.0"
"@babel/code-frame@^7.5.5":
version "7.5.5"
resolved "https://registry.yarnpkg.com/@babel/cod
integrity sha512-27d4lZoomVyo51VegxI20xZPuSHusqbQ
dependencies:
"@babel/highlight" "^7.0.0"
"@babel/code-frame@^7.8.3":
version "7.8.3"
resolved "https://registry.yarnpkg.com/@babel/cod
integrity sha512-a9gxpmdXtZEInkCSHUJDLHZVBgb1QS0j
dependencies:
"@babel/highlight" "^7.8.3"
```
becomes
```
"@babel/code-frame@^7.0.0", "@babel/code-frame@^7.5.5", "@babel/code-frame@^7.8.3":
version "7.8.3"
resolved "https://registry.yarnpkg.com/@babel/cod
integrity sha512-a9gxpmdXtZEInkCSHUJDLHZVBgb1QS0j
dependencies:
"@babel/highlight" "^7.8.3"
```
PR Close#36377
In version 10, undecorated base classes that use Angular features need
to be decorated explicitly with `@Directive()`. Additionally, derived
classes of abstract directives need to be decorated.
The migration already handles this for undecorated classes that are
not explicitly decorated, but since in V9, abstract directives can be
used, we also need to handle this for explicitly decorated abstract
directives. e.g.
```
@Directive()
export class Base {...}
// needs to be decorated by migration when updating from v9 to v10
export class Wrapped extends Base {}
@Component(...)
export class Cmp extends Wrapped {}
```
PR Close#35339
We don't have an integration test for the `undecorated-classes-with-decorated-fields
migration. For consistency and to cover for the latest changes, we add
it to the `ng update` integration test.
PR Close#35339
This has a couple benefits:
- we now use a .bazelversion file rather than package.json to pin the version of bazel we want. This means even if you install bazel on your computer rather than via yarn, you'll still get a warning if your bazel version is wrong.
- you no longer end up downloading three copies of bazel due to bugs in both npm and yarn where they download all tarballs before checking the metadata to see which are usable on the local platform.
- bazelisk correctly handles the tools/bazel trick for wrapping functionality, which we want to use to instrument developer build latencies
PR Close#36078
Prior to this commit, i18n runtime logic relied on the assumption that provided translation is syntactically correct, specifically around ICU syntax. However provided translations might contain some errors that lead to parsing failure. Specifically when translation contains curly braces, runtime i18n logic tries to parse them as an ICU expression and fails. This commit validates ICU parsing result (making sure it was parsed correctly) and throws an error if parsing error happens. The error that is thrown also contains translated message text for easier debugging.
Note: the check and the error message introduced in this PR is a safeguard against the problem that led to unhandled i18n runtime logic crash. So the framework behavior remains the same, we just improve the error message and it should be safe to merge to the patch branch.
Resolves#35689.
PR Close#35923
* integration tests target definitions in integration/BUILD.bazel updated to use a single dict
* payload tracking for integration tests updated to work under Bazel
* legacy integration_test CI job removed
* integration/run_tests.sh script no longer used in CI so it has been updated for running integration tests locally in the legacy way
PR Close#35985
Currently, when running the ngcc binary directly and provide an invalid option ngcc will not error out and the user might have a hard time telling why ngcc is behaving not as expected.
With this change we now output an actionable error:
```
yarn ngcc --unknown-option
Options:
--version Show version number [boolean]
-s, --source A path (relative to the working directory)
of the `node_modules` folder to process.
[default: "./node_modules"]
-p, --properties An array of names of properties in
package.json to compile (e.g. `module` or
`es2015`)
Each of these properties should hold the
path to a bundle-format.
If provided, only the specified properties
are considered for processing.
If not provided, all the supported format
properties (e.g. fesm2015, fesm5, es2015,
esm2015, esm5, main, module) in the
package.json are considered. [array]
-t, --target A relative path (from the `source` path) to
a single entry-point to process (plus its
dependencies).
--first-only If specified then only the first matching
package.json property will be compiled.
[boolean]
--create-ivy-entry-points If specified then new `*_ivy_ngcc`
entry-points will be added to package.json
rather than modifying the ones in-place.
For this to work you need to have custom
resolution set up (e.g. in webpack) to look
for these new entry-points.
The Angular CLI does this already, so it is
safe to use this option if the project is
being built via the CLI. [boolean]
--legacy-message-ids Render `$localize` messages with legacy
format ids.
The default value is `true`. Only set this
to `false` if you do not want legacy
message ids to
be rendered. For example, if you are not
using legacy message ids in your
translation files
AND are not doing compile-time inlining of
translations, in which case the extra
message ids
would add unwanted size to the final source
bundle.
It is safe to leave this set to true if you
are doing compile-time inlining because the
extra
legacy message ids will all be stripped
during translation.
[boolean] [default: true]
--async Whether to compile asynchronously. This is
enabled by default as it allows
compilations to be parallelized.
Disabling asynchronous compilation may be
useful for debugging.
[boolean] [default: true]
-l, --loglevel The lowest severity logging message that
should be output.
[choices: "debug", "info", "warn", "error"]
--invalidate-entry-point-manifest If this is set then ngcc will not read an
entry-point manifest file from disk.
Instead it will walking the directory tree
as normal looking for entry-points, and
then write a new manifest file.
[boolean] [default: false]
--help Show help [boolean]
Unknown arguments: unknown-option, unknownOption
```
PR Close#36010
This commit adds support in the Angular monorepo and in the Angular
compiler(s) for TypeScript 3.8. All packages can now compile with
TS 3.8.
For most of the repo, only a handful few typings adjustments were needed:
* TS 3.8 has a new `CustomElementConstructor` DOM type, which enforces a
zero-argument constructor. The `NgElementConstructor` type previously
declared a required `injector` argument despite the fact that its
implementation allowed `injector` to be optional. The interface type was
updated to reflect the optionality of the argument.
* Certain error messages were changed, and expectations in tests were
updated as a result.
* tsserver (part of language server) now returns performance information in
responses, so test expectations were changed to only assert on the actual
body content of responses.
For compiler-cli and schematics (which use the TypeScript AST) a major
breaking change was the introduction of the export form:
```typescript
export * as foo from 'bar';
```
This is a `ts.NamespaceExport`, and the `exportClause` of a
`ts.ExportDeclaration` can now take this type as well as `ts.NamedExports`.
This broke a lot of places where `exportClause` was assumed to be
`ts.NamedExports`.
For the most part these breakages were in cases where it is not necessary
to handle the new `ts.NamedExports` anyway. ngtsc's design uses the
`ts.TypeChecker` APIs to understand syntax and so automatically supports the
new form of exports.
The View Engine compiler on the other hand extracts TS structures into
metadata.json files, and that format was not designed for namespaced
exports. As a result it will take a nontrivial amount of work if we want to
support such exports in View Engine. For now, these new exports are not
accounted for in metadata.json, and so using them in "folded" Angular
expressions will result in errors (probably claiming that the referenced
exported namespace doesn't exist).
Care was taken to only use TS APIs which are present in 3.7/3.6, as Angular
needs to remain compatible with these for the time being.
This commit does not update angular.io.
PR Close#35864
ng_update_migrations will still access the global yarn cache on its `ng update` call and there is no way to avoid this that I can see but if no other integration tests access the global yarn cache then that one test can have free reign over it.
PR Close#35877
This commit moves the build-related scripts
(`build-ivy-npm-packages.js`, `build-packages-dist.js` and
`package-builder.js`) to a dedicated directory to keep the `scripts/`
directory cleaner.
It also moves the logic for building the `zone.js` package to a separate
script, `zone-js-builder.js`, to make it re-usable. A subsequent commit
will use it to build the `zone.js` package when building the Ivy Angular
packages as well.
PR Close#35780
Because the WORKSPACE file is generated JIT by schematics in this integration test, we need to patch the schematics to add the work-around.
PR Close#35808
`ɵɵgetInheritedFactory()` is called from generated code for a component which extends another class. This function is detected by Closure to have a side effect and is not able to tree shake the component as a result. Marking it with `noSideEffects()` tells Closure it can remove this function under the relevant tree shaking conditions.
PR Close#35769
The flakiness in integration/bazel-schematics is going to be a bit tricker as the WORKSPACE file is JIT generated by the architect build layer
PR Close#35804
Move bazel-in-bazel them to test job & increase it is 2xlarge+. test_integration_bazel is removed. Overall CI credit usage is reduced.
test: include ng_elements_schematics in legacy integration tests temporarily
This test was recently added and use a new pattern that doesn't work with npm_integration_test out of the box. It needs some refactoring to work. Left a TODO for this
PR Close#33927
* it's tricky to get out of the runfiles tree with `bazel test` as `BUILD_WORKSPACE_DIRECTORY` is not set but I employed a trick to read the `DO_NOT_BUILD_HERE` file that is one level up from `execroot` and that contains the workspace directory. This is experimental and if `bazel test //:test.debug` fails than `bazel run` is still guaranteed to work as `BUILD_WORKSPACE_DIRECTORY` will be set in that context
* test //integration:bazel_test and //integration:bazel-schematics_test exclusively
* run "exclusive" and "manual" bazel-in-bazel integration tests in their own CI job as they take 8m+ to execute
```
//integration:bazel-schematics_test PASSED in 317.2s
//integration:bazel_test PASSED in 167.8s
```
* Skip all integration tests that are now handled by angular_integration_test except the tests that are tracked for payload size; these are:
- cli-hello-world*
- hello_world__closure
* add & pin @babel deps as newer versions of babel break //packages/localize/src/tools/test:test
@babel/core dep had to be pinned to 7.6.4 or else //packages/localize/src/tools/test:test failed. Also //packages/localize uses @babel/generator, @babel/template, @babel/traverse & @babel/types so these deps were added to package.json as they were not being hoisted anymore from @babel/core transitive.
NB: integration/hello_world__systemjs_umd test must run with systemjs 0.20.0
NB: systemjs must be at 0.18.10 for legacy saucelabs job to pass
NB: With Bazel 2.0, the glob for the files to test `"integration/bazel/**"` is empty if integation/bazel is in .bazelignore. This glob worked under these conditions with 1.1.0. I did not bother testing with 1.2.x as not having integration/bazel in .bazelignore is correct.
PR Close#33927
Before this change content queries with the `descendants: false` option, as implemented in ivy,
would not descendinto `<ng-container>` elements. This behaviour was different from the way the
View Engine worked. This change alligns ngIvy and VE behaviours when it comes to queries and the
`<ng-container>` elements and fixes a common bugs where a query target was placed inside the
`<ng-container>` element with a * directive on it.
Before:
```html
<needs-target>
<ng-container *ngIf="condition">
<div #target>...</div> <!-- this node would NOT match -->
</ng-container>
</needs-target>
```
After:
```html
<needs-target>
<ng-container *ngIf="condition">
<div #target>...</div> <!-- this node WILL match -->
</ng-container>
</needs-target>
```
Fixes#34768
PR Close#35384
Previously, we needed to manually specify a ChromeDriver version to
download on CI that would be compatible with the browser version
provided by the docker image used to run the tests. This was kept in the
`CI_CHROMEDRIVER_VERSION_ARG` environment variable.
With recent commits, we use the browser provided by `puppeteer` and can
determine the correct ChromeDriver version programmatically. Therefore,
we no longer need the `CI_CHROMEDRIVER_VERSION_ARG` environment
variable.
NOTE:
There is still one place (the `bazel-schematics` integration project)
where a hard-coded ChromeDriver version is necessary. Since I am not
sure what is the best way to refactor the tests to not rely on a
hard-coded version, I left it as a TODO for a follow-up PR.
PR Close#35381
Root cause is that for perf reasons we cache `LFrame` so that we don't have to allocate it all the time. To be extra fast we clear the `LFrame` on `enterView()` rather that on `leaveView()`. The implication of this strategy is that the deepest `LFrame` will retain objects until the `LFrame` allocation depth matches the deepest object.
The fix is to simply clear the `LFrame` on `leaveView()` rather then on `enterView()`
Fix#35148
PR Close#35156
Brings in feat: builtin: expose @npm//foo__all_files filegroup that includes all files in the npm package (https://github.com/bazelbuild/rules_nodejs/commit/8d77827) that is needed for npm_integration_test @npm//puppeteer pkg_tar on OSX (as the OSX Chrrome libs are extracted to paths that contain spaces)
PR Close#35430
This means integration tests no longer need to depend on a $CI_CHROMEDRIVER_VERSION_ARG environment variable to specify which chromedriver version to download to match the locally installed chrome. This was bad DX and not having it specified was not reliable as webdriver-manager would not always download the chromedriver version to work with the locally installed chrome.
webdriver-manager update --gecko=false --standalone=false $CI_CHROMEDRIVER_VERSION_ARG is now replaced with node webdriver-manager-update.js in the root package.json, which checks which version of chrome puppeteer has come bundled with & downloads informs webdriver-manager to download the corresponding chrome driver version.
Integration tests now use "webdriver-manager": "file:../../node_modules/webdriver-manager" so they don't have to waste time calling webdriver-manager update in postinstall
"// resolutions": "Ensure a single version of webdriver-manager which comes from root node_modules that has already run webdriver-manager update",
"resolutions": {
"**/webdriver-manager": "file:../../node_modules/webdriver-manager"
}
This should speed up each integration postinstall by a few seconds.
Further, integration test package.json files link puppeteer via file:../../node_modules/puppeteer which is the ideal situation as the puppeteer post-install won't download chrome if it is already downloaded. In CI, since node_modules is cached it should not need to download Chrome either unless the node_modules cache is busted.
NB: each version of puppeteer comes bundles with a specific version of chrome. Root package.json & yarn.lock currently pull down puppeteer 2.1.0 which comes with chrome 80. See https://github.com/puppeteer/puppeteer#q-which-chromium-version-does-puppeteer-use for more info.
Only two references to CI_CHROMEDRIVER_VERSION_ARG left in integration tests at integration/bazel-schematics/test.sh which I'm not entirely sure how to get rid of it
Use a lightweight puppeteer=>chrome version mapping instead of launching chrome and calling browser.version()
Launching puppeteer headless chrome and calling browser.version() was a heavy-handed approach to determine the Chrome version. A small and easy to update mappings file is a better solution and it means that the `yarn install` step does not require chrome shared libs available on the system for its postinstall step
PR Close#35049
Fixes issue with yarn_install not following yarn-path in .yarnrc when bazel run from yarn with `yarn bazel ...` (rules_nodejs: fix: unset YARN_IGNORE_PATH in yarn_install before calling yarn #1588)
PR Close#34961
There are different `DebugNode`/`DebugElement` implementations (and
associated helper functions) for ViewEngine and Ivy. Additionally, these
classes/functions, which are defined inside the `core` package, are
imported by the `platform-browser` package.
Previously, this code was not tree-shaken as expected in Ivy. #30130
partially addressed the issue, but only for the case where `core` and
`platform-browser` end up in the same closure after webpack's scope
hoisting. In cases where this is not the case, our webpack/terser based
tooling is not capable of tree-shaking it.
This commit fixes the problem, by ensuring that the code retained in Ivy
mode (due to the cross-package import) does not unnecessarily reference
`DebugNode`/`DebugElement`, allowing the code to be tree-shaken away.
This results in a 7.6KB reduction in the size of the main angular.io
bundle.
Jira issue: [FW-1802](https://angular-team.atlassian.net/browse/FW-1802)
PR Close#35003
Previously we would write to class/style as strings `element.className` and `element.style.cssText`. Turns out that approach is good for initial render but not good for updates. Updates using this approach are problematic because we have to check to see if there was an out of bound write to style and than perform reconciliation. This also requires the browser to bring up CSS parser which is expensive.
Another problem with old approach is that we had to queue the DOM writes and flush them twice. Once on element advance instruction and once in `hostBindings`. The double flushing is expensive but it also means that a directive can observe that styles are not yet written (they are written after directive executes.)
The new approach uses `element.classList.add/remove` and `element.style.setProperty/removeProperty` API for updates only (it continues to use `element.className` and `element.style.cssText` for initial render as it is cheaper.) The other change is that the styling changes are applied immediately (no queueing). This means that it is the instruction which computes priority. In some circumstances it may result in intermediate writes which are than overwritten with new value. (This should be rare)
Overall this change deletes most of the previous code and replaces it with new simplified implement. The simplification results in code savings.
PR Close#34804
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
Pete Bacon Darwin
Greg Magolan
crisbeto
Joey Perrott
Paul Gschwendtner
JiaLiPassion
George Kalpakas
Alex Eagle
Andrew Kushnir
Alan Agius
Alex Rickabaugh
Doug Parker
Miško Hevery
Pawel Kozlowski
Igor Minar
Joey Perrott