Summary
When a link reference definition is followed by a line that opens a title delimiter (", ' or () and the title is never closed before the end of the paragraph, commonmark-java 0.30.0 keeps the partial title on the LinkReferenceDefinition and attaches the source spans of the title lines to the definition. The text of those lines is still parsed as a Paragraph (rendered correctly), but that Paragraph has no source spans.
The spec says a title must be enclosed in matching delimiters (0.31.2, §6.3 "link title") and that "No further character may occur" after a definition (§4.7); when the title does not close, the definition has no title and the following lines are a paragraph (this is what cmark and pulldown-cmark do). Issue #315 (fixed by #318 in 0.23.0, "the title was set to the partially-parsed title and the source spans were wrong") covers the sibling case where the title closes and is followed by garbage on the same line; the end-of-paragraph case is not covered by those tests.
Reproduction (0.30.0, Parser.builder().includeSourceSpans(IncludeSourceSpans.BLOCKS_AND_INLINES).build())
Input 1:
[1]: https://example.org/a
(ver tambem https://example.org/b
[1]
Observed:
LinkReferenceDefinition label="1" dest="https://example.org/a" title="ver tambem https://example.org/b\n" spans=[0,26)[27,60)
Paragraph spans=NONE
Text lit="(ver tambem https://example.org/b" spans=[27,60)
Paragraph spans=[62,65)
Link dest="https://example.org/a" title="ver tambem https://example.org/b\n" spans=[62,65)
Expected: definition with title=null and spans [0,26); the paragraph (ver tambem https://example.org/b with spans [27,60). The HTML output already renders the paragraph; only the title and the spans are wrong.
Input 2 (double-quote variant):
[a]: https://example.org/x
"https://example.org/t
mais
[a]
Observed: title="https://example.org/t\nmais\n", definition spans [0,26)[27,49)[50,54), and the Paragraph has no spans.
Input 3 (the #315 case still has a residual span problem): with "t1\nt2" lixo after the definition, the title is correctly discarded, but the span [27,30) of the first title line stays on the definition while the Paragraph only gets [31,39).
Where
internal/LinkReferenceDefinitionParser.java: title() keeps collecting across lines (:251-256) and finishReference() uses the partial title and all collected spans (:275-291); ParagraphParser.java:58-68 then builds the paragraph from the remaining lines after the definition parser already consumed their spans.
Impact
Tools that map nodes back to the source by SourceSpan (we use it to find which parts of a document are covered by definitions) see the paragraph as part of the definition. We work around it on our side by refusing documents where a renderable node has no source position, until a fix is available.
Reproduced with the commonmark-0.30.0.jar from Maven Central on JDK 17 (0.30.0 is the latest release as of 2026-10-06).
Investigated and written with Claude Code while porting a link checker to Android; the reproduction above was run on our side.
Summary
When a link reference definition is followed by a line that opens a title delimiter (", ' or () and the title is never closed before the end of the paragraph, commonmark-java 0.30.0 keeps the partial title on the LinkReferenceDefinition and attaches the source spans of the title lines to the definition. The text of those lines is still parsed as a Paragraph (rendered correctly), but that Paragraph has no source spans.
The spec says a title must be enclosed in matching delimiters (0.31.2, §6.3 "link title") and that "No further character may occur" after a definition (§4.7); when the title does not close, the definition has no title and the following lines are a paragraph (this is what cmark and pulldown-cmark do). Issue #315 (fixed by #318 in 0.23.0, "the title was set to the partially-parsed title and the source spans were wrong") covers the sibling case where the title closes and is followed by garbage on the same line; the end-of-paragraph case is not covered by those tests.
Reproduction (0.30.0, Parser.builder().includeSourceSpans(IncludeSourceSpans.BLOCKS_AND_INLINES).build())
Input 1:
Observed:
Expected: definition with title=null and spans [0,26); the paragraph (ver tambem https://example.org/b with spans [27,60). The HTML output already renders the paragraph; only the title and the spans are wrong.
Input 2 (double-quote variant):
Observed: title="https://example.org/t\nmais\n", definition spans [0,26)[27,49)[50,54), and the Paragraph has no spans.
Input 3 (the #315 case still has a residual span problem): with "t1\nt2" lixo after the definition, the title is correctly discarded, but the span [27,30) of the first title line stays on the definition while the Paragraph only gets [31,39).
Where
internal/LinkReferenceDefinitionParser.java: title() keeps collecting across lines (:251-256) and finishReference() uses the partial title and all collected spans (:275-291); ParagraphParser.java:58-68 then builds the paragraph from the remaining lines after the definition parser already consumed their spans.
Impact
Tools that map nodes back to the source by SourceSpan (we use it to find which parts of a document are covered by definitions) see the paragraph as part of the definition. We work around it on our side by refusing documents where a renderable node has no source position, until a fix is available.
Reproduced with the commonmark-0.30.0.jar from Maven Central on JDK 17 (0.30.0 is the latest release as of 2026-10-06).
Investigated and written with Claude Code while porting a link checker to Android; the reproduction above was run on our side.