astral-html v0.0.4

Released on 2026-10-20.

Bug fixes

• Select text modes according to SVG and HTML integration context so well-formed
  SVG no longer causes Reader or Document to consume following HTML as text
  (astral-sh/astral-html#85)
• Honor self-closing SVG elements so they no longer remain open and incorrectly
  exhaust Document's depth limit
  (astral-sh/astral-html#85)
• Preserve CDATA sections as text while parsing SVG content
  (astral-sh/astral-html#85)
• Close an immediately open HTML a or option when another start tag has the
  same name, preventing sibling runs from incorrectly hitting Document's depth
  limit
  (astral-sh/astral-html#86)

Performance

• Borrow unchanged comment text during tokenizer recovery, avoiding allocations
  unless newline or NUL normalization is required
  (astral-sh/astral-html#75)

Other changes

• Document where Reader and Document differ from browser parsing and warn
  against using their output to approve HTML for rendering
  (astral-sh/astral-html#78)
• Clarify that Doctype::force_quirks == false does not by itself indicate
  standards mode
  (astral-sh/astral-html#78)
• Add README highlights for the crate's runtime dependency, unsafe-code policy,
  and html5lib tokenizer test coverage
  (astral-sh/astral-html#77)
• Round README benchmark results to match their measurement precision
  (astral-sh/astral-html#83)

astral-html v0.0.3

Released on 2026-10-05.

Performance

• Speed parsing after tags with many attributes by discarding oversized
  deduplication tables instead of repeatedly clearing them
  (astral-sh/astral-html#72)

Other changes

• Make parser performance easier to compare with benchmark tables covering link
  extraction and document construction across five workloads
  (astral-sh/astral-html#71)

astral-html v0.0.2

Released on 2026-10-05.

Other changes

• Show the full repository README, including examples and benchmarks, on
  crates.io
  (astral-sh/astral-html#69)

astral-html v0.0.1

Released on 2026-10-05.

This is the first release of astral-html, a high-performance HTML parser
designed for document traversal.

Enhancements

• Add read-only document and element views with attribute lookup, descendant
  iteration, text extraction, and configurable resource limits
  (astral-sh/astral-html#8)
• Add a streaming event reader that automatically selects the correct text mode
  for HTML elements
  (astral-sh/astral-html#7)
• Add a UTF-8 tokenizer that borrows unchanged input and supports data, RCDATA,
  raw-text, script, plaintext, and CDATA states
  (astral-sh/astral-html#5)
• Decode named and numeric HTML character references according to their text or
  attribute context
  (astral-sh/astral-html#3)
• Allow documents with any number of attributes while retaining input, node, and
  nesting limits
  (astral-sh/astral-html#34)
• Add the initial astral-html crate without unsafe code or a mandated
  allocator
  (astral-sh/astral-html#2)

Bug fixes

• Treat text-mode closing tags with non-letter names as text, matching HTML
  tokenizer rules
  (astral-sh/astral-html#10)

Performance

• Reduce routine parsing allocations by reusing attribute storage and borrowing
  unchanged comments, decoded strings, and common normalized names
  (astral-sh/astral-html#16)
  (astral-sh/astral-html#21)
• Make unknown and unterminated named references substantially faster with early
  attribute-context rejection and a compact trie
  (astral-sh/astral-html#45)
  (astral-sh/astral-html#52)
• Borrow unchanged punctuation-heavy comments instead of allocating strings for
  them
  (astral-sh/astral-html#55)
• Normalize newlines and NUL characters in script, raw-text, and plaintext
  content in a single pass
  (astral-sh/astral-html#54)
• Coalesce runs of literal < characters, sharply reducing memory use for
  adversarial text
  (astral-sh/astral-html#24)
• Reuse owned buffers when replacing NUL characters, reducing peak memory use
  for large tokens
  (astral-sh/astral-html#30)
• Normalize comment newlines while scanning to avoid a second full-buffer
  allocation
  (astral-sh/astral-html#28)
• Release excessive capacity after character-reference decoding so small results
  do not retain input-sized allocations
  (astral-sh/astral-html#27)
• Avoid indexing balanced open elements until malformed nesting requires
  recovery
  (astral-sh/astral-html#53)
• Scan comment and script content in bulk
  (astral-sh/astral-html#36)
• Delay attribute-name indexing until it is beneficial and reuse its allocation
  between tags
  (astral-sh/astral-html#38)
  (astral-sh/astral-html#46)
• Speed named-reference decoding with direct handling for common names and
  narrower searches for uncommon names
  (astral-sh/astral-html#39)
  (astral-sh/astral-html#44)
• Reuse the tokenizer’s first character-reference offset instead of rescanning
  unchanged text prefixes
  (astral-sh/astral-html#40)
• Avoid rescanning tag, attribute, and doctype names already marked for
  normalization
  (astral-sh/astral-html#56)
• Speed element and attribute queries for short normalized names
  (astral-sh/astral-html#58)
• Reduce buffered token-dispatch overhead in Reader and Tokenizer
  (astral-sh/astral-html#57)
• Use compact parent indices to reduce each document node by eight bytes on
  64-bit platforms
  (astral-sh/astral-html#42)
• Avoid allocating names when entering built-in script, raw-text, and RCDATA
  modes
  (astral-sh/astral-html#43)