llm_tag_parser 0.2.1

A robust, reactive streaming tag parser for LLM streams that isolates thought, code, or custom blocks chunk-by-chunk in real-time.

0.2.1

• Feature: Added XmlTagUtilities class and TagNode.tagName getter to cleanly parse tag names (e.g. Material.Card or ui:button) from XML-like tag definitions.

• Robustness: Added 20+ comprehensive stress tests validating escape characters, backtrack boundary splits, and dot-notation namespace formatting.

• Docs: Updated README to document the 0.2.0 API surface.

  • Added Parallel Tag Instances section covering .instances and TagNode routing.
  • Added Chronological Node Stream section covering the low-level nodes stream and TextNode/TagNode types.
  • Expanded Attribute Extraction section with .getAttributeStream() and .getAttributeFuture() convenience helpers.
  • Expanded API Reference with full LlmTagParser.nodes, LlmTagContent, and LlmNode type tables.
  • Added Instance Isolation row to the Robustness table.
  • Updated Quick Start version pin to ^0.2.0.

0.2.0

• Feature: Unified Chronological Node Event Stream (nodes).
  • Re-architected parser around a chronological stream of LlmNodes (TextNode and TagNode), preserving the exact timeline order of the response.
• Feature: True Tag Instance Isolation.
  • Every matched tag occurrence (e.g. parallel <interface> blocks) is represented as a distinct TagNode instance with its own completely isolated stream and future.
  • Zero content bleeding, leakage, or multiplexing issues between identical sibling tag occurrences.
• Feature: Dynamic Tag Instance Routing (instances).
  • Exposed a .instances reactive stream on LlmTagContent to easily listen to and route specific tag occurrences as they open.
• Feature: Stack-Based Nested Tag Routing.
  • Uses an internal active instance stack to correctly route nested tags (like <tool_use> inside <thinking>), ensuring parent instances correctly contain their children's content.
• Feature: Standard Replay Buffering.
  • Re-implemented subscriber buffering utilizing Dart's standard Stream.multi, providing robust, native support for late and multiple subscribers.
• API Extension: Added .getAttributeStream(name) and .getAttributeFuture(name) convenience methods.

0.1.2

• Feature: Replaced regex attribute parser with an extremely robust state-machine scanner.
  • Full support for hyphens, namespaces, periods, and numbers in attribute key names (e.g. data-id, xml:lang, ns:a.b-c_d).
  • Supports unquoted attribute values (e.g. <interface id=main>).
  • Correctly parses escaped quotes (e.g. \", \') inside attribute values without truncating them.
  • Automatically identifies boolean / key-only attributes (e.g. disabled or checked) and maps them to standard flags.
  • Dynamic quote-aware tag header boundaries: handles mathematical operators (like age > 21), nested brackets, and closing tag sequences occurring inside attribute values seamlessly.

0.1.1

• Feature: Implemented prefix-matching dynamic lookahead buffer.
  • Replaced the static sliding-window lookahead with an adaptive suffix-prefix scanner.
  • Reverts lookahead latency to 0 characters for standard conversational text.
  • Automatically buffers only when the stream mimics a potential tag boundary, and instantly flushes buffered text if a mismatch occurs (0-latency backtracking).

0.1.0

• Initial version.