evanw / tla-fuzzer

A fuzzer for various top-level await bundling strategies

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Top-level await correctness fuzzer

This is a fuzzer to test the correctness of various top-level await JavaScript bundling strategies. Fuzzing is done by randomly generating module graphs and comparing the evaluation order of the bundled code with V8's native module evaluation order.

How to run

  1. Install dependencies with npm ci
  2. Run the fuzzer with node ./fuzzer.js

Current results

"Same" here means that the bundled code behaves exactly the same as the unbundled code. "Different" here means that the bundled code behaves differently (i.e. is evaluated in a different order) than unbundled code. The same percentage means how many runs were same out of 300 total runs.

Note: Both the specification and V8/node currently have subtle bugs that cause undesirable behavior. So it's not really the case that matching V8/node 100% exactly is desirable. But it is desirable to match V8/node at least almost exactly (~99%) as the bugs are very subtle and only affect a few edge cases. Hopefully the various implementations of top-level await will converge on the same behavior in the future.

Variant: simple

  • Custom module registry algorithm: βœ… Same (100% same)
  • Webpack 5.75.0: βœ… Same (100% same)
  • Import becomes inline await: 🚫 Different (80% same)
  • Rollup 3.7.4: 🚫 Different (80% same)
  • SystemJS 6.13.0: 🚫 Different (67% same)

Variant: trailing promise

  • Webpack 5.75.0: βœ… Same (100% same)
  • Custom module registry algorithm: 🚫 Different (100% same)
  • SystemJS 6.13.0: 🚫 Different (40% same)
  • Rollup 3.7.4: 🚫 Different (14% same)
  • Import becomes inline await: 🚫 Different (11% same)

Variant: cyclic

  • Custom module registry algorithm: 🚫 Different (99% same)
  • Webpack 5.75.0: 🚫 Different (99% same)
  • SystemJS 6.13.0: 🚫 Different (85% same)
  • Import becomes inline await: 🚫 Different (67% same)
  • Rollup 3.7.4: 🚫 Different (67% same)

Variant: cyclic, trailing promise

  • Custom module registry algorithm: 🚫 Different (99% same)
  • Webpack 5.75.0: 🚫 Different (99% same)
  • SystemJS 6.13.0: 🚫 Different (46% same)
  • Rollup 3.7.4: 🚫 Different (22% same)
  • Import becomes inline await: 🚫 Different (20% same)

About

A fuzzer for various top-level await bundling strategies


Languages

Language:JavaScript 100.0%