Word and LibreOffice comparison handoffs

Pinned source: TortoiseGit 7338078f8ddd924b8cddee35f512f2286072136d, contrib/diff-scripts/diff-doc.js and merge-doc.js.

Connected source behavior

Source behavior	Mac adaptation
`diff-doc.js` falls back to OpenOffice when creating Word fails	An unavailable Word installation selects the existing LibreOffice UNO adapter using fresh editable copies of the same captured bytes. Permission denial, compile failure and uncertain automation do not start another application.
Modern Word branch swaps inputs before its comparison	Captured base is opened and compared against captured new, preserving the existing modern direction, target-new, author and formatting options.
Outline/master/reading view normalization	Normalize outline unconditionally; normalize master/reading only when no subdocuments exist. The condition preserves source operator precedence.
Mark comparison result saved, then close source	Discover a unique new window and reject preexisting document names before marking its document saved; close only the captured private input. Ambiguous results remain available for review.
Suppress automatic macros	Disable and read back automation security before opening inputs, then restore and read back the previous setting on success and failure. Failure to confirm restoration fails the handoff.
Modern three-way Word merge	Preserve base/theirs and base/mine tracked comparisons, then merge theirs into private mine. Close captured base/theirs on failure; preserve a possible edited mine once revision editing starts.
Merge uses Word only	No automatic LibreOffice merge fallback. Its explicit adapter retains the source mine/theirs UNO workflow and existing Save As/Mark Resolved instructions.

Native ownership

Word and LibreOffice comparison requests keep captured access leases and Quit accounting until their workers finish. Owner closure, task cancellation and app cleanup cancel the captured token. Word receives a private marker and checks it between commands; cancellation cannot undo a command already in flight. The serialized handler still attempts input cleanup and security restoration before returning a typed receipt. Macro-security failure, malformed completion or an uncertain result retains private files. Proven unopened/retired requests remove only their own preview directory.

Input closure uses the exact private full path. If a partial open loses its returned reference, cleanup may reacquire only a unique exact private-path match that was absent before dispatch. Unknown or ambiguous ownership retains files. No cancellation path quits Word or closes its active document by selection.

The installed Word dictionary has no result value for compare. New-window and preexisting-document exclusion is a conservative adaptation, not an atomic result provenance guarantee during concurrent user-created/closed windows. That physical acceptance case remains open; the port does not claim exact Word ownership solely from a successful injected receipt.

Validation boundaries

The grouped runner is scripts/test-word-office-family.py. It uses private preferences, temporary repositories/files and hidden offscreen windows. The existing dispatcher fixtures cover captured bytes, fallback selection, denied and uncertain handoffs, Save As/explicit resolution, stale responses and worker/Quit lifetime. The additional automation fixture runs the actual compile/event/typed receipt bridge with injected replies, including malformed/security/cancellation receipts and merge error mapping. Script-contract assertions inspect the fixed ownership and view/security statements. They do not execute those statements in Word. An optional dictionary check compiles both handlers without executing an Office command.

Live Word output, macro behavior/restoration, concurrent window activity, Word format converters, LibreOffice runtime compatibility, real TCC and file grants, physical interaction, minimum OS and signed App Store acceptance remain open. PowerPoint shape-level revision decisions and editable merged decks remain a separate implementation gap. See the user guides for current workflows: Word comparison, Word merge, LibreOffice.

Batch evidence records the passing Debug build and eight native case/engine combinations, compiled Word terminology, owned receiver cleanup and remaining physical/signed limits. Core was unchanged.