ADR 0009: openpyxl reads spreadsheets, through its own API¶
.xlsx and .xlsm are read with openpyxl in read_only mode, iterating rows
directly. pandas.read_excel is not used: it has no chunksize, so the whole
sheet materializes and the bounded-memory invariant is lost.
python-calamine was measured and rejected. It collapses formula, blank, and
empty-string cells to '', and types the same integer as int from a .xls
and as float from an .xlsx; that information is gone before this package
sees it, so no rule here can recover it. It is also a compiled extension, which
ties availability to wheels published for each interpreter in the supported
range.
openpyxl is pure Python, streams rows, and keeps the distinctions that matter: an integer stays an integer, formula text survives, and a blank cell is distinguishable from an empty string. Its one measured defect is dimension inflation – a single formatting-only cell at H40 reported a 2x3 sheet as 40x8 – which ADR 0010 repairs by trimming to the used range.
Legacy .xls is out of scope and raises a typed error naming the conversion.
Supporting it would add a second engine, a second coercion rule, and a second
fixture generator for a format last current in 2003. If it returns, xlrd
is the engine to use: it keeps the pure-Python property and leaves this path
untouched, and it needs its own boolean rule because it reports True as 1.