Surfaced by review during the per-warning-id sweep (PR #497, id [0001]
unused_value).
plan_paragraph_text_edits in docx2html/docx/run_surgery_partial.mbt — the
N0c2 partial token-boundary surgery planner, roughly 900 lines plus its helper
graph — has no production caller. run_surgery_api.mbt documents the addressed
run-replacement boundary as the ONE public seam through which a shipped surface
reaches the surgery, and that seam reaches N0c1. So everything under
plan_paragraph_text_edits is live only under its whitebox and property suites.
The compiler says this as unused_value, and #497 suppresses the diagnostic —
deleting a complete, tested planner would be worse, and moving ~3200 lines of
whitebox tests inline to make it "used" is a test-reorganisation, not a fix. But
the suppression silences a message worth keeping: an implementation with no
seam. A comment at the declaration now records it; this issue tracks the
decision.
Three ways out, in the order they were suggested in review:
- Wire it up — give N0c2 a public seam the way N0c1 has one.
- Isolate it as explicitly experimental/test-only code, so its status is
structural rather than a comment.
- Keep it as it is, with the status conspicuous, until (1) lands.
The same question applies to ParagraphTextEdit (its request type, tracked by
the struct_never_constructed suppression in PR #494) and to
PartialSurgeryLogicalEdit/PartialSurgeryReceipt (whose write-only fields PR
#496 deleted). All four are the same fact seen through different diagnostics.
Surfaced by review during the per-warning-id sweep (PR #497, id [0001]
unused_value).plan_paragraph_text_editsindocx2html/docx/run_surgery_partial.mbt— theN0c2 partial token-boundary surgery planner, roughly 900 lines plus its helper
graph — has no production caller.
run_surgery_api.mbtdocuments the addressedrun-replacement boundary as the ONE public seam through which a shipped surface
reaches the surgery, and that seam reaches N0c1. So everything under
plan_paragraph_text_editsis live only under its whitebox and property suites.The compiler says this as
unused_value, and #497 suppresses the diagnostic —deleting a complete, tested planner would be worse, and moving ~3200 lines of
whitebox tests inline to make it "used" is a test-reorganisation, not a fix. But
the suppression silences a message worth keeping: an implementation with no
seam. A comment at the declaration now records it; this issue tracks the
decision.
Three ways out, in the order they were suggested in review:
structural rather than a comment.
The same question applies to
ParagraphTextEdit(its request type, tracked bythe
struct_never_constructedsuppression in PR #494) and toPartialSurgeryLogicalEdit/PartialSurgeryReceipt(whose write-only fields PR#496 deleted). All four are the same fact seen through different diagnostics.