Skip to content

v0.10.10

@zizixmas zizixmas tagged this 07 Jun 05:09
LinkReferenceDefinitionGroup blocks fell through MarkdownDocumentReader.ExtractBlock's
default arm into NormalizeRenderer, leaking the raw definition plus a trailing []:
into chunk content (Filer saw trailing []: after tables/lists in osi-model.md /
tcp-vs-udp.md). Reference definitions are markdown metadata, not body text.

Skip LinkReferenceDefinitionGroup / LinkReferenceDefinition in ExtractBlock so they
never reach chunk content. Inline links keep their display text and resolved target,
so no content is lost.

Regression tests (TDD, each verified RED with the fix stashed):
- MarkdownDocumentReaderTests: labeled-def strip + inline-link preservation,
  table-then-def trailing []: guard, multiple-defs strip-all.
- NativeProcessAsyncMarkerStripTests: the exact Filer path
  (IDocumentProcessorFactory.Create -> ProcessAsync -> Result.Chunks[].Content)
  asserts zero []: / [label]: url and no content loss.

README "Clean chunk content" section documents the reference-definition exclusion.
Assets 2
Loading