Excel: a percent column rejects European decimals that a number column accepts

[dvejsada]

Found while working on #126.

What happens. _apply_column_type() in xlsx_tools/helpers.py normalises thousands and decimal separators for the number and currency branches, but not for percent. The percent branch calls float() on the raw text, so a European decimal fails to parse and the cell silently stays text — no number, no percent format, no warning.

Reproduced against b36ebfb:

types:number       cell '4,3'          -> NUMBER 4.3
types:percent      cell '4,3'          -> text   '4,3'
types:number       cell '1.234,56'     -> NUMBER 1234.56
types:percent      cell '1.234,56'     -> text   '1.234,56'
types:currency:€   cell '1.234,56 €'   -> NUMBER 1234.56
types:percent      cell '4.3'          -> NUMBER 0.043

The same value, in the same spreadsheet, is a number in one column and a string in the next — decided only by which type the column declares. A 4,3% written by a European author lands as text, so it does not sum, chart or sort.

Why. helpers.py:963-966:

if type_lower.startswith('percent'):
    explicit = type_spec.split(':', 1)[1].strip() if ':' in type_spec else ''
    numeric_str = clean.rstrip('%').strip()
    try:
        cell.value = float(numeric_str) / 100

number (:930) and currency (:916) both pass their text through _strip_thousands_separators() first. That helper already handles English 1,234.56, European 1.234,56 and bare thousands 1,234, and its docstring records the 1,5 → 15 silent 10x error that motivated it. The percent branch never got the same treatment.

Fix. Route the percent branch through _strip_thousands_separators() like the other two:

numeric_str = _strip_thousands_separators(clean.rstrip('%').strip())

Two things to check while doing it, since they are the same omission one layer over:

• resolve_cell() (:403-410) detects a trailing % outside the types: directive entirely and calls float(percent_body) directly — same gap, different entry point. An untyped 4,3% in a plain table has the same problem.
• _percent_decimals() (:162-165) already does .replace(',', '.') before counting fraction digits, so it is more lenient than the coercion that feeds it. Once the coercion accepts 4,3%, confirm the two agree on the resulting format — a mismatch between a learner and its coercion is exactly what xlsx: a formula in a percent column gets a flat 0% format #126 fixed for a different pair.

Warning channel. Whatever the fix, a percent cell that cannot be parsed should say so rather than quietly becoming text — xlsx_tools/warnings.py exists for this since #114, and "the caller asked for a percent and got a string" is a warning-severity substitution by the rubric in AGENTS.md.

Tests. tests/test_xlsx_percent_columns.py covers the types: percent path; a European-decimal case belongs beside the existing ones, and should be confirmed to fail on the commit before the fix.