fix(core): tracked span replace inside another author's insertion

[kklimuk]

🤖 docx-cli triage

What was wrong

replace "disputed amount" "disputed sum" --track --author Editor on text inside Reviewer A's pending <w:ins> left "disputed sum" bare inside A's insertion, so it was credited to A. When A's insertion also held Reviewer B's nested <w:del>, offsets slid 24 characters: "The Client may " was deleted instead of the match. Both cases exited 0 with 1 occurrence replaced. edit --at pN:S-E --track shares the same code path (replaceSpanInParagraph) and had the same bug.

What changed

• Under tracking, a match that sits entirely inside a tracked-change wrapper now splits that wrapper where it is. Editor's <w:del> stays nested in A's insertion, "disputed sum" goes in Editor's own <w:ins> between the halves, and the trailing half keeps A's author/date with a new unique w:id, which is how Word records it. Offsets count from the wrapper's start, and B's deletion is moved whole, untouched.
• Code review on the rewrite also fixed these, each with tests:
  • Tracked replacements inside a hyperlink, smartTag or field used to land as bare text. They now get their own <w:ins>, and cut link text becomes a tracked <w:del>.
  • Insertion-point spans (pN:S-S) used to be appended untracked at the end of the paragraph. They now land at the point, tracked.
• Files: mostly src/core/find/replace-span.tsx (a ~595-line rewrite), a new isTrackedChangeWrapper helper in src/core/parser/run-ops.ts, a filter option on wrapContiguousTrackable, and a notes section in src/core/track-changes/CLAUDE.md.
• Fixture: multi-tracked.docx gets a cross-author p3 (Editor edits inside Reviewer A's still-pending insertion). Two tracked-change count assertions in track-changes.test.ts went up to match; tc0–tc5 are unchanged.

How it works

flowchart TD
  A[match span in paragraph] --> B{tracked and span inside one ins/del wrapper?}
  B -- no --> C[paragraph-level del + ins, as before]
  B -- yes --> D[split wrapper at span end; new w:id on tail half]
  D --> E[wrap matched plain runs in Editor w:del, still inside head half]
  E --> F[insert Editor w:ins between halves]

Loading

Follow-up commit (maintainer decisions)

• Same author merges: when the editor is the author of the enclosing <w:ins>/<w:moveTo>, a span wholly inside it is edited in place: the cut text is removed, and the new text joins that insertion (same w:id). This matches Word editing your own pending insertion. Different authors keep the split shape.
• --baseline = all changes rejected: runs now record their enclosing revision (within in the AST), and a run is hidden in a view if any revision around it is. Read, wc, find, the markdown renderer and iterateBlocks({view}) all share isRevisionVisible.
• Code review in that pass also fixed:
  • Nested wrappers: a link, smart tag or insertion nested inside a split wrapper is now split recursively, so its matched text is cut.
  • revN pairing: a del and an ins now pair only when both are by the same author.
  • Untracked edge insertions no longer join the neighbouring author's insertion.
  • Fenced code blocks now respect --accepted/--baseline.

Verification

• Same author: <w:ins 1 A>may withhold a disputed sum.</w:ins>, with no del and one revision.
• Different author (case 1): the XML is as shown above. read --baseline → "The Client ", identical to track-changes reject --all; read --accepted → "The Client may withhold a disputed sum."
• Nested-deletion doc (case 2): --baseline → "Fees are due. ", identical to reject --all; --accepted → "Fees are due. The Client may withhold disputed sum."
• A new test asserts read --baseline ≡ reject --all and read --accepted ≡ accept --all for del-in-ins and ins-in-del docs, plus wc/find --baseline.
• bun run check ✅ bun run test:unit ✅ (1733 tests).

Decisions (settled)

• Same-author edits merge into the existing insertion.
• --baseline shows the document with every change rejected.
• Split out as follow-ups: Untracked replace across a hyperlink nested in another author's insertion corrupts text and duplicates w:id #16 (an untracked replace across a hyperlink inside another author's insertion duplicates w:id and corrupts text) and Whole-paragraph edit --text --track silently resolves other authors' pending changes #17 (whole-paragraph edit --text --track resolves other authors' changes).

Not verified

• The same-author merge on a <w:moveTo> (a tracked move) hasn't been checked against Word.

Fixes #13

🤖 Generated with Claude Code

https://claude.ai/code/session_012GoxD36rnNfg6uArkVxtJL

---

Generated by Claude Code

[kklimuk]

🤖 docx-cli triage

Opened this for #13. A tracked replace inside another author's pending insertion now splits that insertion and puts the new text in the editor's own <w:ins>, the way Word does. The nested-deletion offset shift is fixed, and both repro cases match the issue's expected output. Unit tests (1715) and bun run check pass. This is a sizable rewrite of replace-span.tsx, and it has one behavior change for same-author edits plus three known gaps, all listed in the description.

Action needed: approve/merge — decide the same-author split question and whether the three gaps become follow-up issues, then merge #15.

---

Generated by Claude Code

[kklimuk]

🤖 docx-cli triage

Closing at the maintainer's request. Three review rounds on this ~1,300-line rewrite each surfaced ~15 further issues, several of them regressions against main, so it is being replaced by a minimal, narrowly gated fix for #13 built from scratch on current main. The branch stays as a reference.

---

Generated by Claude Code