[rmorgans]

RevisionProcessor.RejectRevisions keeps the inserted text and drops the deleted text when the revisions sit inside a w:hyperlink, the opposite of Reject All. Accept is correct.

Reproduction

<w:p><w:hyperlink w:anchor="x">
  <w:r><w:t xml:space="preserve">Alpha </w:t></w:r>
  <w:del ...><w:r><w:delText>beta</w:delText></w:r></w:del>
  <w:ins ...><w:r><w:t>delta</w:t></w:r></w:ins>
  <w:r><w:t xml:space="preserve"> gamma</w:t></w:r>
</w:hyperlink></w:p>

Reject gives Alpha delta gamma; Word gives Alpha beta gamma.

Cause. Reject inverts w:del/w:ins and then accepts. ReverseRevisionsTransform only inverts revisions whose nearest ancestor (skipping sdt, sdtContent, smartTag) is w:p, so inside a hyperlink nothing is inverted.

Fix. Skip w:hyperlink as well, like smartTag.

Test. RP053_RevisionsInsideHyperlink (accept and reject).

---

Found while using RevisionProcessor to check that Reject All of a tracked-change redline gives back the original document. Expected results are Microsoft Word for Mac 16.113's own Accept All / Reject All on the same markup. The regression test fails on current master and passes with the fix; ./build.sh passes.

This is one of six small, independent RevisionProcessor fixes (#548, #549, #550, #551, #552, #553). Each adds its test after RP001 in RevisionProcessorTests.cs, so whichever merges second will need a trivial rebase there; I'll rebase the rest as they land.

Prepared with an AI coding agent (Claude); I reviewed the change and ran the tests.

[rmorgans]

Folded into #554. All three fixes added a name to the same ancestor predicate in ReverseRevisionsTransform, so they could only merge one at a time with a rebase after each. #554 now covers w:hyperlink, w:fldSimple, w:dir and w:bdo in one change, with one parameterized test. Closing this one to save you the extra rebase cycle.