Also have a new node type for queries.
This leads to less memory usage and better performance.
Deep Tree Benchmark results (depth 11):
- createAndDestroy (view engine vs current codegen):
* pureScriptTime: 78.80+-4% vs 72.34+-4%
* scriptTime: 78.80+-4% vs 90.71+-9%
* gc: 5371.66+-108% vs 9717.53+-174%
* i.e. faster when gc is also considered and about 2x less memory usage!
- update unchanged
Part of #14013
PR Close#14120
The new view engine allows our codegen to produce less code,
as it can interpret view definitions during runtime.
The view engine is not feature complete yet, but already
allows to implement a tree benchmark based on it.
Part of #14013
Detailed changes:
- remove `UNINITIALIZED`, initialize change detection fields with `undefined`.
* we use `view.numberOfChecks === 0` now everywhere
as indicator whether we are in the first change detection cycle
(previously we used this only in a couple of places).
* we keep the initialization itself as change detection get slower without it.
- remove passing around `throwOnChange` in various generated calls,
and store it on the view as property instead.
- change generated code for bindings to DOM elements as follows:
Before:
```
var currVal_10 = self.context.bgColor;
if (jit_checkBinding15(self.throwOnChange,self._expr_10,currVal_10)) {
self.renderer.setElementStyle(self._el_0,'backgroundColor',((self.viewUtils.sanitizer.sanitize(jit_21,currVal_10) == null)? null: self.viewUtils.sanitizer.sanitize(jit_21,currVal_10).toString()));
self._expr_10 = currVal_10;
}
var currVal_11 = jit_inlineInterpolate16(1,' ',self.context.data.value,' ');
if (jit_checkBinding15(self.throwOnChange,self._expr_11,currVal_11)) {
self.renderer.setText(self._text_1,currVal_11);
self._expr_11 = currVal_11;
}
```,
After:
```
var currVal_10 = self.context.bgColor;
jit_checkRenderStyle14(self,self._el_0,'backgroundColor',null,self._expr_10,self._expr_10=currVal_10,false,jit_21);
var currVal_11 = jit_inlineInterpolate15(1,' ',self.context.data.value,' ');
jit_checkRenderText16(self,self._text_1,self._expr_11,self._expr_11=currVal_11,false);
```
Performance impact:
- None seen (checked against internal latency lab)
Part of #13651
- add ng2_switch benchmark to track `ngFor` over `ngSwitch`
- measure create only, createDestroy and update
- simplify the created dom
- always add a style binding
Tobias Bosch
Victor Berchet
Matias Niemelä
Bowen Ni
Bowen Ni
vsavkin
Joao Dias
Pawel Kozlowski
Alex Eagle