Refactor to LET — PR 4: promotable literals + drop-input UX polish (spec 0008)

[jimmytacks]

Summary

Final slice of the Refactor to LET feature (spec 0008). Adds literal promotion to the promotable section and polishes the drop-input UX + dialog status bar.

Closes #205 · Part of #201

What changed

Literal tokenizer (RefactorEngine.WalkLiterals)

Walks the formula text in source order, masking cell-ref token spans (new CellRefExtractor.MatchSpans) so the row digits of A1 aren't mistaken for a literal, and skipping string contents / sheet qualifiers. Yields:

• Numeric literals (100, 0.20, 1.5e-10) — dedupe by parsed value, so 0.20 and 0.2 collapse to one row (first spelling preserved). No leading sign is captured, so A1*-5 correctly promotes 5 and keeps - inline.
• String literals — dedupe by un-escaped value (embedded "" handled); token keeps the original escaped spelling.
• Boolean literals — TRUE/FALSE case-insensitive, excluding function calls (TRUE() and sheet qualifiers.

Each unique value becomes a default-off RefactorPromotableRow(kind: Literal) with an occurrence count.

Promotion + rewrite

Promoting a literal materialises an input binding whose RHS is the original token text (preserves 0.20 vs 0.2). A new RewriteLiterals pass — run after the cell-ref and identifier passes so a promoted-string binding name is never re-scanned as an identifier — swaps every occurrence for the binding name.

Drop-input UX

Unticking a promoted literal or named range drops it from the rewrite map, restoring the original token inline everywhere (no warning even if it reintroduces duplication). Covered by new tests.

Status bar

Now shows a summary count — 3 inputs, 2 promotable, 1 calculation binding — plus a · N merged note, with validation errors taking priority.

Note on ordering

Literals are surfaced in source order, matching the first-occurrence convention used everywhere else in the engine. For the manual-test formula this means 50 appears between "high" and "medium" rather than after "low" as the issue's example listing sketched — the source-order placement is the principled one.

Tests

12 new RefactorEngineTests cover: numeric/string(with "")/boolean promotion, two distinct numerics staying distinct, value-dedupe (0.20/0.2), digits-inside-identifier safety, the issue's manual-example token set, drop-input restoring literals + named ranges, existing-LET literal scoping (value-binding RHS excluded), and round-trip safety via LetParser + LetToLambdaBuilder. Two pre-existing promotable tests were narrowed to their kind (their formulas now also surface a literal). Full suite: 980 passing.

Manual Excel test (for Tim)

Type =IF(A1 > 100, "high", IF(A1 > 50, "medium", "low")) + IF(B1 > 100, 1, 0), run /Refactor. The promotable section should list 100 (2 uses), "high", 50, "medium", "low", 1, 0. Promote 100 only, Save, and confirm both 100s were replaced and the formula evaluates identically.

🤖 Generated with Claude Code