Summary

Fixes #36. When a table directly follows paragraph text without a blank line, Marko's GFM parser doesn't recognize it as a Table element—the table rows are parsed as inline content within a Paragraph. The line wrapper then breaks wide rows at the wrap width (88 chars), producing invalid markdown like Geo:\nIntl % |. This PR makes the line wrapper always preserve table rows on single lines.

Additionally fixes table separator rendering to preserve original dash widths instead of normalizing everything to | --- |.

Changes

• src/flowmark/linewrapping/tag_handling.py: Modified add_tag_newline_handling() to:
  • Always treat table rows (|-prefixed lines) as segment boundaries, regardless of whether template tags are present (previously only activated when tags were detected)
  • Pass table-row segments through verbatim instead of sending them to the base wrapper, preventing any width-based wrapping
  • Preserve existing behavior for list items (only segment boundaries when tags are present)
• src/flowmark/formats/flowmark_markdown.py: Modified render_table() to preserve original table separator dash widths (ensuring a minimum of 3 dashes) instead of normalizing all separators to exactly ---
• tests/test_wrapping.py: Added 4 new unit tests + updated 1 existing test:
  • test_table_rows_preserved_without_tags — table rows in paragraphs stay on own lines
  • test_wide_table_rows_not_wrapped — exact reproduction of issue Wide table header rows broken mid-cell by line wrapping #36
  • test_table_rows_with_semantic_wrapping — semantic mode also preserves rows
  • test_table_rows_only_no_surrounding_text — standalone table rows preserved
  • test_block_heuristics_only_with_tags_for_lists — updated regression guard (renamed from test_block_heuristics_only_with_tags)
• tests/test_filling.py: Added 2 integration tests via fill_markdown():
  • test_wide_table_adjacent_to_paragraph — end-to-end reproduction of Wide table header rows broken mid-cell by line wrapping #36
  • test_standalone_wide_table — regression guard for properly-parsed tables
• tests/testdocs/: Added "Wide Table Adjacent to Paragraph" section to testdoc.orig.md and updated all 4 expected output files (plain, semantic, cleaned, auto) — now with preserved separator dash widths

Test Plan

• All 287 tests pass (full suite: python -m pytest tests/ -v)
• All 4 inline doctests pass (python -m pytest src/flowmark/linewrapping/block_heuristics.py)
• All 4 golden reference doc tests pass (plain, semantic, cleaned, auto modes)
• Table separator dashes are preserved (not normalized to minimum 3)
• Manual testing: run flowmark --auto on a file containing a wide table directly after paragraph text (no blank line) — verify table rows are preserved on single lines
• Manual testing: run flowmark --auto on a file with a standalone wide table (blank line before it) — verify separator dashes are preserved
• Manual testing: verify normal paragraph text still gets reflowed at wrap width
• Edge case: table rows inside template tags ({% field %}) — verify existing behavior preserved
• Edge case: table rows inside blockquotes — verify no interference
• Edge case: table with alignment markers (:---:, :---, ---:) — verify alignment preserved
• Edge case: list items without tags — verify they are NOT treated as segment boundaries (regression guard)

Related Beads

• fm-1422 (epic): Fix table rows broken by line wrapping when parsed as paragraph content (issue Wide table header rows broken mid-cell by line wrapping #36)
• fm-625c: Core fix — remove has_tags guard for table row detection
• fm-y6np: Unit tests in test_wrapping.py
• fm-gt51: Integration tests in test_filling.py
• fm-vzb9: Reference doc test updates
• fm-ngl0: Regression guard verification
• fm-3ahm: Full test suite run

https://claude.ai/code/session_01JESDwDj2UYgf9v5yttYyxr