Skip to content

README.md

Latest commit

 

History

History
72 lines (53 loc) · 2.78 KB

README.md

File metadata and controls

72 lines (53 loc) · 2.78 KB
Raw

Real-world transpilation benchmark

This benchmark runs Babel standalone inside MiniRacer and transpiles a pinned real-world modern JavaScript module toward an ES2015-ish target.

No large benchmark inputs are committed. resources.json stores only resource locations, byte sizes, and SHA-256 checksums. The benchmark downloads resources into tmp/benchmark/transpile on first run and verifies each checksum before use.

Default input:

  • pdfjs_worker_4_10_38
  • URL: https://unpkg.com/pdfjs-dist@4.10.38/build/pdf.worker.mjs
  • SHA-256: 7c237f83fa56bce645d8af51d183c9c56ba7b2d2928ff42754dc7020bea36323
  • Size: 2,209,730 bytes (~2.1 MiB)

Compiler:

  • @babel/standalone@7.26.10
  • URL: https://unpkg.com/@babel/standalone@7.26.10/babel.min.js
  • SHA-256: 7acf405141ca9e611b6d40aeef52071432d5f387cf906fc3556fb9f0b1a28601

Run from the repository root:

bundle exec ruby benchmark/transpile/bench.rb

Useful modes:

# Show pinned downloadable resources
bundle exec ruby benchmark/transpile/bench.rb --list-resources

# Download and checksum resources without running the benchmark
bundle exec ruby benchmark/transpile/bench.rb --fetch-only

# Quick one-iteration smoke run; use rounds=1 to opt out of the stable-sample
# floor used for multi-round transpile runs.
BENCH_ITERATIONS=1 BENCH_ROUNDS=1 BENCH_WARMUP=1 bundle exec ruby benchmark/transpile/bench.rb

# Run on MiniRacer's single-threaded platform
BENCH_SINGLE_THREADED=1 bundle exec ruby benchmark/transpile/bench.rb
# or:
bundle exec ruby benchmark/transpile/bench.rb --single-threaded

# Use the smaller pdf.js API module instead of the worker
bundle exec ruby benchmark/transpile/bench.rb --source pdfjs_api_4_10_38

# Run only the case that returns the generated code to Ruby
bundle exec ruby benchmark/transpile/bench.rb --only 'transpile_call_source_to_es6_code'

# Machine-readable output
bundle exec ruby benchmark/transpile/bench.rb --json --output /tmp/mini_racer-transpile.json

Before timing each benchmark case, the suite performs untimed warmup iterations. Multi-round runs collect at least seven timed samples because Babel/pdf.js transpilation has regular V8 GC outliers; with only two or three samples, the median can report an arbitrary GC phase instead of a stable center. Use --rounds 1 only for smoke runs.

Benchmark cases:

  • babel/load_standalone: creates a context and loads Babel standalone.
  • babel/transpile_cached_source_to_es6_length: source is already stored in JS; returns only generated code length.
  • babel/transpile_call_source_to_es6_length: passes the large source string from Ruby each iteration; returns only generated code length.
  • babel/transpile_call_source_to_es6_code: passes source from Ruby and returns the generated code string to Ruby, exercising both large input and output boundary costs.