Fluent style processing benchmark

This repo is a small benchmark to compare the status quo of Fluent v9 preprocessing Griffel styles with a proposal to stop shipping pre-processed styles.

Methodology

Tests are run with Tensile, a stress test benchmarking tool.

There are two test categories:

1. "with style processing": the way Fluent is currently shipped with Griffel AOT. This is the baseline case.
2. "no processing": a patched version of Fluent that disables Griffel AOT.
   1. Artifacts are produced by disabling this line and this line.

Components are imported directly from their packages rather than from the Fluent suite package. E.g.:

// import this way
import { Button } from '@fluentui/react-button';
// rather than
import { Button } from '@fluentui/react-components';

Packages are aliased for the "with style processing" case with @fluentui/react-button being aliased to react-button and so forth.

Pacakges are patched for the "no processing" case to eliminate the AOT step in the build.

Setup

1. Clone this repo
2. Install deps: yarn (this will handle aliasing and apply patches for you)

Usage

View the scripts block in package.json for the test cases.

1. Test cases prefixed with bench: are "with style processing"
2. Test cases prefixed with bench-no-process: are "no processing"

Results are written to the .tensile/results folder in the JSON file named for the test. The processedMeasurments node rolls up the various measurements.