Currently we convert objects to strings using `'' + value` which is quickest,
but it stringifies the value using its `valueOf`, rather than `toString`. These
changes switch to using `String(value)` which has identical performance
and calls the `toString` method as expected. Note that another option
was calling `toString` directly, but benchmarking showed it to be slower.
I've included the benchmark I used to verify the performance so we have it
for future reference and we can reuse it when making changes to `renderStringify`
in the future.
Also for reference, here are the results of the benchmark:
```
Benchmark: renderStringify
concat: 2.006 ns(0%)
concat with toString: 2.201 ns(-10%)
toString: 237.494 ns(-11741%)
toString with toString: 121.072 ns(-5937%)
constructor: 2.201 ns(-10%)
constructor with toString: 2.201 ns(-10%)
toString mono: 14.536 ns(-625%)
toString with toString mono: 9.757 ns(-386%)
```
Fixes#38839.
PR Close#39843
`ViewContainerRef` is declared in ViewEngine but it sub-classed in Ivy. This creates a circular
dependency between ViewEngine `ViewContainerRef` which needs to declare `__NG_ELEMENT_ID__` and
ivy factory which needs to create it. The workaround used to be to pass the `ViewContainerRef`
through stack but that created a very convoluted code. This refactoring simply bundles the
two files together and removes the stack workaround making the code simpler to follow.
PR Close#39621
`TemplateRef` is declared in ViewEngine but it sub-classed in Ivy. This creates a circular
dependency between ViewEngine `TemplateRef` which needs to declare `__NG_ELEMENT_ID__` and
ivy factory which needs to create it. The workaround used to be to pass the `TemplateRef`
through stack but that created a very convoluted code. This refactoring simply bundles the
two files together and removes the stack workaround making the code simpler to follow.
PR Close#39621
`ElementRef` is declared in ViewEngine but it sub-classed in Ivy. This creates a circular
dependency between ViewEngine `ElementRef` which needs to declare `__NG_ELEMENT_ID__` and
ivy factory which needs to create it. The workaround used to be to pass the `ElementRef`
through stack but that created a very convoluted code. This refactoring simply bundles the
two files together and removes the stack workaround making the code simpler to follow.
PR Close#39621
This commit updates micro benchmarks to use relative path to Ivy runtime code. Keeping absolute
locations caused issues with build optimizer that retained certain symbols and they appeared in the
output twice.
PR Close#39142
This commit adds micro benchmarks to run micro benchmarks for i18n-related logic in the
following scenarios:
- i18n static attributes
- i18n attributes with interpolations
- i18n blocks of static text
- i18n blocks of text + interpolations
- simple ICUs
- nested ICUs
First 4 scenarios also have baseline scenarios (non-i18n) so that we can compare i18n perf with
non-i18n logic.
PR Close#39142
`TNodeType.View` was created to support inline views. That feature did
not materialize and we have since removed the instructions for it, leave
an unneeded `TNodeType.View` which was still used in a very
inconsistent way. This change no longer created `TNodeType.View` (and
there will be a follow up chang to completely remove it.)
Also simplified the mental model so that `LView[HOST]`/`LView[T_HOST]`
always point to the insertion location of the `LView`.
PR Close#38707
Host `TNode` was passed into `getOrCreateTNode` just so that we can
compute weather or not we are a root node. This was needed because
`previousOrParentTNode` could have `TNode` from `TView` other then
current `TView`. This is confusing mental model. Previous change
ensured that `previousOrParentTNode` must always be part of `TView`,
which enabled this change to remove the unneeded argument.
PR Close#38707
Refactors the `ng_rollup_bundle` rule to a macro that relies on
the `@bazel/rollup` package. This means that the rule no longer
deals with custom ESM5 flavour output, but rather only builds
prodmode ES2015 output. This matches the common build output
in Angular projects, and optimizations done in CLI where
ES2015 is the default optimization input.
The motiviation for this change is:
* Not duplicating rollup Bazel rules. Instead leveraging the official
rollup rule.
* Not dealing with a third TS output flavor in Bazel.The ESM5 flavour has the
potential of slowing down local development (as it requires compilation replaying)
* Updating the rule to be aligned with current CLI optimizations.
This also _fixes_ a bug that surfaced in the old rollup bundle rule.
Code that is unused, is not removed properly. The new rule fixes this by
setting the `toplevel` flag. This instructs terser to remove unused
definitions at top-level. This matches the optimization applied in CLI
projects. Notably the CLI doesn't need this flag, as code is always
wrapped by Webpack. Hence, the unused code eliding runs by default.
PR Close#37623
Update docs in the micro benchmarks to include:
* How to run with no turbo inlining
* Where to find the profiles in the DevTools
* Best way to debug benchmarks (using the profile_in_browser rather than --inspect-brk)
PR Close#37140
Switches our tslint setup to the standard `tslint.json` linter excludes.
The set of files that need to be linted is specified through a Yarn script.
For IDEs, open files are linted with the closest tslint configuration, if the
tslint IDE extension is set up, and the source file is not excluded.
We cannot use the language service plugin for tslint as we have multiple nested
tsconfig files, and we don't want to add the plugin to each tsconfig. We
could reduce that bloat by just extending from a top-level tsconfig that
defines the language service plugin, but unfortunately the tslint plugin does
not allow the use of tslint configs which are not part of the tsconfig project.
This is problematic since the tslint configuration is at the project root, and we
don't want to copy tslint configurations next to each tsconfig file.
Additionally, linting of `d.ts` files has been re-enabled. This has been
disabled in the past and a TODO has been left. This commit fixes the
lint issues and re-enables linting.
PR Close#35800
This commit updates the host bindings micro benchmark to run tests with mutliple directives (where each directive contains host bindings). The number of directives is configurable as a constant in the micro benchmark file. This change is needed to have an ability to measure/compare perf in different scenarios.
PR Close#35736
This commit adds micro benchmark for host bindings, so that we can assess the impact of changes related to host bindings (for example PR #35568).
PR Close#35705
- Adds `TView` into `LFrame`, read the `TView` from `LView` on `enterView`.
- Before this change the `TView` was ofter looked up from `LView` as `lView[TVIEW]`. This is suboptimal since reading from an Array, requires that the read checks array size before the read. This means that such a read has a much higher cost than reading from the property directly. By passing in the `TView` explicitly it makes the code more explicit and faster.
- Some rearrangements of arguments so that `TView` would come before `LView` for consistency.
PR Close#35069
These tests are used for perf testing and don't run as part of CI, as a result they bit-rotted. This fixes that. Long term these tests should be run as part of CI.
PR Close#35071
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
This change introduces class/style reconciliation algorithm for DOM elements.
NOTE: The code is not yet hooked up, it will be used by future style algorithm.
Background:
Styling algorithm currently has [two paths](https://hackmd.io/@5zDGNGArSxiHhgvxRGrg-g/rycZk3N5S)
when computing how the style should be rendered.
1. A direct path which concatenates styling and uses `elemnent.className`/`element.style.cssText` and
2. A merge path which uses internal data structures and uses `element.classList.add/remove`/`element.style[property]`.
The situation is confusing and hard to follow/maintain. So a future PR will remove the merge-path and do everything with
direct-path. This however breaks when some other code adds class or style to the element without Angular's knowledge.
If this happens instead of switching from direct-path to merge-path algorithm, this change provides a different mental model
whereby we always do `direct-path` but the code which writes to the DOM detects the situation and reconciles the out of bound write.
The reconciliation process is as follows:
1. Detect that no one has modified `className`/`cssText` and if so just write directly (fast path).
2. If out of bounds write did occur, switch from writing using `className`/`cssText` to `element.classList.add/remove`/`element.style[property]`.
This does require that the write function computes the difference between the previous Angular expected state and current Angular state.
(This requires a parser. The advantage of having a parser is that we can support `style="width: {{exp}}px" kind of bindings.`)
Compute the diff and apply it in non destructive way using `element.classList.add/remove`/`element.style[property]`
Properties of approach:
- If no out of bounds style modification:
- Very fast code path: Just concatenate string in right order and write them to DOM.
- Class list order is preserved
- If out of bounds style modification detected:
- Penalty for parsing
- Switch to non destructive modification: `element.classList.add/remove`/`element.style[property]`
- Switch to alphabetical way of setting classes.
PR Close#34004
This commit fixes a couple issues that prevent `class_binding` benchmark from running: moving constants requires by the `benchmark` function before function declaration and referencing correct consts in template instructions.
PR Close#34242
Since config=ivy now sets the define=compile flag and the define=angular_ivy_enabled
flag to cause usage of Ivy, we can update all of the documentation and scripts that
reference compile=aot to use config=ivy.
PR Close#33983
Micro-benchmarks were broken after we've introduced concept of
DECLARATION_COMPONENT_VIEW on LView (after this change embedded
views must have a pointer to a parent LView).
PR Close#34031
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
Prior to this patch all the styling benchmarks only tested for
template map-based style/class bindings. Because of each of the bindings
being only present in the template, there was no possibility of
there being any duplicate map-based styling bindings.
This benchmark introduces benchmarking for map-based style/class bindings
that are evaluated from both template bindings as well as directives.
This benchmark can be executed by calling:
```
bazel build //packages/core/test/render3/perf:duplicate_map_based_style_and_class_bindings_lib.min_debug.es2015.js
node dist/bin/packages/core/test/render3/perf/duplicate_map_based_style_and_class_bindings_lib.min_debug.es2015.js
```
The benchmark is also run via the `profile_all.js` script (found in
`packages/core/test/render3/perf/`)
PR Close#33608
Prior to this patch all the styling benchmarks only tested for
template-based style/class bindings. Because of each of the bindings
being only present in the template, there was no possibility of
there being any duplicate bindings. This benchmark introduces
style/class bindings being evaluated from both a template and from
various directives.
This benchmark can be executed by calling:
```
bazel build //packages/core/test/render3/perf:duplicate_style_and_class_bindings_lib.min_debug.es2015.js
node dist/bin/packages/core/test/render3/perf/duplicate_style_and_class_bindings_lib.min_debug.es2015.js
```
The benchmark is also run via the `profile_all.js` script (found in
`packages/core/test/render3/perf/`)
PR Close#33600
When debugging `LView`s it is easy to get lost since all of them have
the same name. This change does three things:
1. It makes `TView` have an explicit type:
- `Host`: for the top level `TView` for bootstrap
- `Component`: for the `TView` which represents components template
- `Embedded`: for the `TView` which represents an embedded template
2. It changes the name of `LView` to `LHostView`, `LComponentView`, and
`LEmbeddedView` depending on the `TView` type.
3. For `LComponentView` and `LEmbeddedView` we also append the name of
of the `context` constructor. The result is that we have `LView`s which
are name as: `LComponentView_MyComponent` and `LEmbeddedView_NgIfContext`.
The above changes will make it easier to understand the structure of the
application when debugging.
NOTE: All of these are behind `ngDevMode` and will get removed in
production application.
PR Close#33449
The 4b81bb5c97 patch seemingly broke the
`profile_all.js` file due to the file renaming. This patch restores the
functionality of said script.
PR Close#33494
this makes running and profiling tests much easier. Example usage:
```
yarn bazel run --define=compile=aot //packages/core/test/render3/perf:noop_change_detection
```
See README.md update for more info.
PS: I considered moving the ng_rollup bundle into the macro but I didn't want to make
too many changes in this PR. If we find running benchmarks in this way useful, we
should refactor the build file more, and move the ng_rollup_bundle targets into the
macro.
PR Close#33389
The styling algorithm requires that the `RNode` has a `className`
property in order to execute the fast-path. This changes adds the
emulation of this property.
PR Close#33392
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
`LFrame` stores information specifice to the current `LView` As the code
enters and leaves `LView`s we use `enterView()` and `leaveView()`
respectively to build a a stack of `LFrame`s. This allows us to easily
restore the previous `LView` instruction state.
PR Close#33178
Decrease `MIN_SAMPLE_DURATION` to make it more likely that we cane fit into single time slice.
Increase `MIN_SAMPLE_COUNT_NO_IMPROVEMENT` to make it more likely to find the best
PR Close#33341
A PR that updates one of the benchmarks and another one that changes the signature for `elementStart` got in around the same time which is causing a compilation error. These changes fix the error.
PR Close#33067
Kristiyan Kostadinov
Misko Hevery
Andrew Kushnir
Paul Gschwendtner
Joey Perrott
Andrew Scott
Joey Perrott
Pawel Kozlowski
Matias Niemelä
Igor Minar