NOTE: This is very much a WORK IN PROGRESS.
This is a JS-focused micro-benchmark of the React framework using the matrix-react-sdk library to provide a more "real-world" example while still being focused and repeatable(-ish).
The benchmark mocks out some chat events and then switches rooms back and force rapidly. The matrix-react-sdk uses a different key per room which triggers a nearly complete update in React.
An additional goal is to be able to run this directly in the SpiderMonkey js-shell using a minimal polyfill/mock of the DOM APIs. This is easier to do detailed performance testing with and gives additional data points on how much the DOM API interaction plays a role in performance of this.
- Setup NodeJS/NPM tools. If building Firefox, you can also use the version
from
mach boostrap
which can be found in$HOME/.mozbuild/node/bin
. - Run
npm install
to fetch dependencies using thepackage.json
file. - Run
npm run build
to generate the WebPack/Babel output files. - Optionally host the HTML and dist/main.js file using a simple server (such as
python3 -m http.server
) 5.a) Load the matrix_demo.html page and check console after done. 5.b) Run js-shell on dist/main.js and look at output.
matrix_demo.html - This loads the main JS file and has dummy CSS.
babel.config.json - Controls the level of transpiling. Modelled after the practical settings used by the Element Web (Matrix) client.
webpack.config.js - WebPack configuration to enable Babel. It also uses the null-loader
to disable a eagerly loaded modules of the matrix-element-sdk to avoid features we are not focused on such as WASM compile time. This can be used to switch between production and development configurations.
package.json - Dependency files for build and execution. Mostly just Babel and React packages.
dist/main.js - The main (generated) JS file that is loaded in browser or run directly in a JS shell.
src/shell-polyfill-hack.js - Mocked functions for SpiderMonkey shell with just enough detail to run the application. This is ignored in browser environments which use the real DOM APIs.
The dist/main.js
file can be run directly in the SpiderMonkey shell. It will
automatically mock some web-platform APIs to allow things to run (mostly)
correctly. This polyfill will use plain-objects as DOM nodes and maintains the
parent/child relationships as well as small amounts of additional state. This
will have different performance characteristics than a real DOM
implementation.
If the matrix code encounters errors, it may render an error page instead of
throwing an exception to the top level. When making changes, it is worth dumping
the mock DOM tree and seeing if it is reasonable. The src/DumpDOMTree.js
code
can help out here. Note that Promise microtasks need to be flushed for an issued
render to complete.
It's possible to run with the webpack mode set to 'development' however that
causes react to do a bunch of additional checks which will change the
performance characteristics. Unfortuantely, the react npm package seems to
contain a preminified version that's used for production. To work around this
it's necessary to rebuild the 'react' package from source. This can be done by
cloning https://github.com/facebook/react (checking out
46b68eaf626f924093952125bd75ba16df0fe204 because the v17.0.2 seems wrong
(facebook/react#24079)) and running yarn build --pretty
to avoid the minification process. Then run npm link
in the
resulting build/node_modules/react
and build/node_modules/react-dom
directories. This will install the unminfied versions globally. This version
can then be used by matrix-react-bench
by running npm link react react-dom
in the matrix-react-bench
directory. Finally, add optimization: { minimize: false }
to webpack.config.js