v0.1.13113 Aug 2026·Rendering, Web APIs & DOM, Media

Brotli, from scratch: the engine now speaks the web's best compression

Until now the engine asked servers for uncompressed pages: it could decode gzip when a CDN forced it, but it never advertised compression, so almost every page travelled at full size. This release adds a complete Brotli decoder (RFC 7932) written from scratch, like the rest of the stack - the format every major server and browser prefers for text, typically 15 to 25 percent smaller than gzip. Brotli is a serious format: canonical prefix codes with run-length-coded code lengths, three interleaved symbol streams with block-switching, context modeling that picks a literal code from the previous two bytes, a sliding window, and a built-in 122,784-byte dictionary whose words pass through 121 textual transforms. The static data is machine-extracted from the RFC itself and verified against the checksums the RFC publishes. Correctness is proven the only honest way: a 28-fixture corpus compressed by an independent implementation decodes byte identically, and a live check fetched a real CDN page served as Brotli and got the exact HTML. The engine now sends Accept-Encoding gzip, deflate, br on every request, so pages simply arrive smaller and faster. Networking moves 54 to 55; overall holds at 52.

Parity with the reference browser

74overall, as of v0.1.343

Web APIs / DOM surface42

Security architecture66

GPU / graphics acceleration68

DevTools / extensions72

Networking85

Media85

Accessibility85

Real-world site compatibility99

HTML parsing100

CSS parsing + selectors100

Layout: block / flex / grid / tables / writing-modes100

Paint + visual100

JavaScript language100

JavaScript performance (JIT)100

HTML parsing100

What changed 3 changes

Media	A from-scratch RFC 7932 Brotli decoder (core-media/brotli.rs): bit reader, canonical prefix codes (simple and complex forms with accumulating run-length code lengths), meta-block headers, block-type switching across three symbol streams, literal context modeling (LSB6/MSB6/UTF8/Signed), distance ring buffer with the spec's exact update rules, the static dictionary with all 121 word transforms, uncompressed and metadata meta-blocks, and a decompression-bomb output cap.
Rendering	Static data extracted from the RFC itself: scripts/gen_brotli_data.py parses the dictionary, transform list and context tables out of RFC 7932's own text and verifies each against the CRC-32 check values the RFC publishes, then emits the Rust tables - no hand transcription anywhere.
Web APIs & DOM	Proven and wired: 28 reference-compressed fixtures (text, HTML, UTF-8, incompressible noise, 1.2MB repetitive streams at two window sizes, empty, one byte) decode byte-identically; truncated and garbage inputs error rather than panic; trailing garbage is rejected. The engine now advertises Accept-Encoding: gzip, deflate, br and decodes at the single seam all transports share - verified live against a real CDN serving Brotli.