angular-cn/modules/benchmarks/src/tree/ng2_static_ftl

Ng2 Static Faster Than BaseLine (FTL) Benchmark

This benchmark was produced in the following way:

1. Run AoT over the root component, one branch component and the leaf component of the ng2_static benchmark
2. Move all 3 component classes to tree.ts
3. Add a depth property to the view factory constructor of the TreeBranchComponent and pass it down to nested view factory calls to be able to use the same view factory for all nesting levels. This is just to make the experiment simpler to not need to change so much code.
4. Optimize the tree_branch.ngfactory.ts and tree_leaf.ngfactory.ts (see below)

The goal of this benchmark is to explore what kind of code the Ng2 compiler should generate in order to get faster. I.e. right now, the Ng2 compiler does not produce this kind of code, but hopefully will in the future.

Optimizations and cleanups

See tree_branch.ngfactory.ts and tree_leaf.ngfactory.ts.

Performance:

• remove view factory functions, call the constructor of the view class directly
• remove createInternal and move the logic into the constructor
• remove AppView parent class
• generate direct calls to detectChangesInternal / destroyInternal of nested components
• remove Renderer and SanitizationService to update the dom directly via document.createElement / element.appendChild / setting Text.nodeValue / Element.style....
• remove AppElement
• create component instance in view constructor

Code size:

• helper methods createElementAndAppend / createTextAndAppend / createAnchorAndAppend
• wrap dirty check for TreeComponent.data into the method updateData in the view class
• remove injectorGetInternal assuming we can detect that nobody is injecting an Injector

Initial results: File size

File size for view class of the component (without imports nor host view factory):

• before: 81 LOC
• after: 53 LOC

Initial results: Deep Tree Benchmark

BENCHMARK deepTree.... Description:

• bundles: false
• depth: 11
• forceGc: false
• regressionSlopeMetric: scriptTime
• runId: f9ae32f0-8580-11e6-914d-9f4f8dbfb5e8
• sampleSize: 20
• userAgent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_11_6) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/53.0.2785.116 Safari/537.36

...createOnly	gcAmount	gcTime	majorGcTime	pureScriptTime	renderTime	scriptTime
ng2.static	10068.17+-33%	9.50+-43%	0.01+-84%	83.01+-18%	57.81+-19%	92.33+-16%
ng2.ftl	493.36+-435%	0.65+-435%	0.00+-435%	28.48+-16%	53.97+-7%	28.48+-16%
baseline	53.81+-435%	0.10+-435%	0.00+-435%	19.79+-20%	52.08+-19%	19.89+-20%
incremental_dom	311.14+-254%	2.43+-248%	0.00+-207%	68.30+-20%	59.31+-19%	70.73+-19%

.....update	gcAmount	gcTime	majorGcTime	pureScriptTime	renderTime	scriptTime
ng2.static	0.00	0.00	0.00	24.65+-12%	31.43+-22%	24.65+-12%
ng2.ftl	0.00	0.00+-435%	0.00+-435%	16.02+-13%	29.15+-9%	16.02+-13%
baseline	509.97+-176%	0.65+-269%	0.40+-434%	28.32+-16%	35.80+-33%	28.32+-16%
incremental_dom	961.48+-246%	0.52+-311%	0.31+-435%	28.94+-19%	36.13+-21%	28.94+-19%

Ratios (create, pureScriptTime)

• 2.9x faster than current implementation
• 2.3x faster than incremental dom
• 1.4x slower than baseline