linexcel 1.4.0

[1.4.0] — 2026-08-25

Fixed

• A comma where the engine writes a dot is no longer a divergence. A
  workbook saved by a French Excel stores 6,7 €, linexcel recalculates
  6.7 €, and the report used to show a red disagreement on a cell that is
  right — the separator comes from the regional settings of the machine that
  saved the file, and is written nowhere in it. Underneath were two
  contradictory comparisons: the Python one returned "no difference" for
  every pair of strings, so a recalculated oui over a stored non raised
  no warning at all, while the viewer compared rendered text and called any
  difference a disagreement. There is now one rule, with three answers —
  same, format (the same value written with another locale's separators)
  and differ — decided once and carried on the node as cachedAgreement.
  German 1.234,5 and Swiss 1'234.5 grouping are read too; grouping is held
  to groups of three digits, so Salle 1 2 against Salle 12 stays a
  disagreement. Both readings are still shown side by side; only the verdict
  changes, and the card names the case.
• A text result that genuinely differs is now reported. It never was: the
  old comparison ignored every pair of strings. Expect warnings on workbooks
  that were silent before — they were silent wrongly.

Added

• A run that will be long says so before it starts. Past a few seconds,
  the estimate is printed and the two ways out are named:

  16 MB of formulas: this should take about 10 seconds. --dry-run says
  what is in the file without analysing it; -v shows progress.
  

  It comes from the uncompressed weight of the sheet parts, which the zip
  index already carries, so asking costs about a twentieth of a millisecond
  and every run can afford to. An order of magnitude, not a stopwatch reading:
  it knows nothing about the machine it runs on, and a workbook of mostly
  values finishes faster than it says. Smaller files say nothing. --dry-run
  states the estimate too, and inspect_workbook() returns it as
  estimatedSeconds alongside the new sheetBytes.

• linexcel.analyzer.sheet_bytes(): the uncompressed size of a workbook's
  sheet parts, read from the zip index without unpacking one.