Follow-on from the DocxDiff N-way consolidate work (PR #228), surfaced during the pre-D4 hardening (native move + per-cell table composition).
Current behavior
When a reviewer moves a paragraph to a position that crosses a table boundary (e.g. relocating a trailing paragraph to between a lead paragraph and a table), the block aligner can read the table as the displaced block. The consolidate then collapses that table to a whole-table block-level conflict (under BaseWins the base table is kept), so a second reviewer's disjoint table-cell edit in that table is surfaced as part of the block conflict rather than composed per-cell.
This is not a correctness failure: there is no silent loss, reject ≡ base holds, and the contested table edit is recorded in the conflict (not dropped). It is a precision/quality limitation in how a paragraph move adjacent to a table interacts with per-cell table composition.
Repro (sketch)
- Base:
paragraph A + 2×2 table + paragraph B.
- Reviewer Alice: move
paragraph B to just before the table (crossing the table boundary).
- Reviewer Bob: edit one table cell.
- Observed: whole-table block conflict; Bob's cell edit not composed. (If Alice instead moves
B to the very top so the lead+table stay contiguous, both compose cleanly — confirming the boundary-crossing is the trigger.)
Desired
A paragraph move adjacent to / crossing a table boundary should not contest the table block; the move and the disjoint table-cell edit should each apply independently (native move + per-cell compose), conflicts == 0, reject ≡ base.
Pointers
Docxodus/Ir/Diff/IrCompositeMerger.cs (move planning PlanMoves, table-compose branch / ComposeTableDiffs)Docxodus/Ir/Diff/IrBlockAligner.cs (which block is chosen as the relocated/displaced one)- Design context:
docs/architecture/ir_diff_engine.md § N-way composite / Consolidate.
Acceptance criteria
- A reviewer move that crosses a table boundary + another reviewer's disjoint table-cell edit both apply (native move + composed cell),
conflicts == 0, reject ≡ base, no content loss.
- Regression: base 2-way parity stays 179/179; full suite green.
Follow-on from the DocxDiff N-way consolidate work (PR #228), surfaced during the pre-D4 hardening (native move + per-cell table composition).
Current behavior
When a reviewer moves a paragraph to a position that crosses a table boundary (e.g. relocating a trailing paragraph to between a lead paragraph and a table), the block aligner can read the table as the displaced block. The consolidate then collapses that table to a whole-table block-level conflict (under
BaseWinsthe base table is kept), so a second reviewer's disjoint table-cell edit in that table is surfaced as part of the block conflict rather than composed per-cell.This is not a correctness failure: there is no silent loss,
reject ≡ baseholds, and the contested table edit is recorded in the conflict (not dropped). It is a precision/quality limitation in how a paragraph move adjacent to a table interacts with per-cell table composition.Repro (sketch)
paragraph A+2×2 table+paragraph B.paragraph Bto just before the table (crossing the table boundary).Bto the very top so the lead+table stay contiguous, both compose cleanly — confirming the boundary-crossing is the trigger.)Desired
A paragraph move adjacent to / crossing a table boundary should not contest the table block; the move and the disjoint table-cell edit should each apply independently (native move + per-cell compose),
conflicts == 0,reject ≡ base.Pointers
Docxodus/Ir/Diff/IrCompositeMerger.cs(move planningPlanMoves, table-compose branch /ComposeTableDiffs)Docxodus/Ir/Diff/IrBlockAligner.cs(which block is chosen as the relocated/displaced one)docs/architecture/ir_diff_engine.md§ N-way composite / Consolidate.Acceptance criteria
conflicts == 0,reject ≡ base, no content loss.