Fix silent drop of named LAMBDA / non-range defined names on round-trip

[senoff]

Original Problem

Modern Excel workbooks can define named LAMBDA functions and other formula expressions at the workbook level (definedName elements whose value is a formula like LAMBDA(x,x*2) rather than a range address). ExcelJS silently dropped these on load because DefinedNamesXform.extractRanges() returned an empty array for non-range values, and parseClose() discarded values with no ranges. On round-trip, all named LAMBDAs were gone.

Cause

extractRanges() was designed for range lists and had no concept of formula expressions. Values like LAMBDA(x,x*2) were split by comma, each fragment tested as a range (isValidRange), and — since none passed — the result was an empty array. The caller then stored {name: 'MyDouble', ranges: []} and threw the body away. Additionally, DefinedNamesXform.render() always called model.ranges.join(','), which produced an empty string for non-range names.

Fix

Three surgical changes:

lib/xlsx/xform/book/defined-name-xform.js

1. extractRanges(): added a quote-aware formula heuristic. A ( that appears outside single-quoted sheet names (i.e. not in 'Data (2026)'!$A$1) signals a formula expression. Return [] immediately to prevent incorrectly splitting the formula body.

   Heuristic: count single-quote characters before the first (. If the count is even (zero included), the ( is not inside a quoted sheet name — treat as formula.

2. parseClose(): when ranges.length === 0 and the text is non-empty, store model.formula = text verbatim.

3. render(): write model.formula verbatim instead of model.ranges.join(',') when formula is set.

lib/doc/workbook.js

4. Added this._formulaDefinedNames = [] field to store formula-typed defined names separately (the DefinedNames/CellMatrix layer does not handle non-range entries).

5. get model(): concatenates _formulaDefinedNames into the definedNames array.

6. set model(value): splits incoming definedNames into range-based (→ _definedNames) and formula-based (→ _formulaDefinedNames).

Files changed

• lib/xlsx/xform/book/defined-name-xform.js — render, parseClose, extractRanges
• lib/doc/workbook.js — constructor, get model, set model
• spec/unit/xlsx/xform/book/defined-name-xform.spec.js — 3 new test cases (LAMBDA, LAMBDA multi-param, sheet-with-parens regression)
• spec/integration/pr/lambda-defined-name.spec.js — new integration test (3 round-trip cases)

Test Run

Unit: 16 passing (10ms)
Integration: 3 passing (32ms)

Regression test for sheet names containing parentheses ('Data (2026)'!$A$1) confirms the formula heuristic does not misclassify valid range references.

Cross-PR check

lib/doc/workbook.js is also touched by PR #67 (getWorksheet case-insensitive). PR #67 touches only getWorksheet() body (lines ~112–116); this fix touches the constructor (line 31), get model (line 163), and set model (line 223–225). No overlap. lib/xlsx/xform/book/defined-name-xform.js is not touched by any other open senoff PR.

Note: .prettierrc edit removed from this PR. The canonical .prettierrc / .eslintrc fix is in PR #69.

Excel/soffice verification

This fix touches the serialization of workbook-level defined names (xl/workbook.xml). soffice is not available in this worker environment. The round-trip tests confirm LAMBDA definitions survive the write/load cycle via JSZip XML inspection. Recommend maintainer open a workbook with named LAMBDAs in Excel after merge to verify the functions are usable.

grace-review summary

Run 1 — MEDIUM (gpt-5.5): extractRanges ( heuristic misclassified 'Data (2026)'!$A$1 as formula. Fixed: heuristic is now quote-aware (counts single quotes before first (). Added regression test.

Run 2 — openai:gpt-5.5: No defects found. gemini: LOW (whitespace-only defined name body not preserved) — deferred, whitespace-only defined names don't appear in real Excel files, current behavior (drop) is safe.

Note: committed with --no-verify per AGENTS.md Rule 1.