#Reading large sheets

open builds the whole workbook in memory. For a sheet of hundreds of thousands of cells, walk it a row at a time instead:

def r as sheet.RowReader init sheet.openRows("huge.xlsx", "Sheet1");
while (sheet.hasRow($r)) {
    $r = sheet.nextRow($r);
    process(sheet.rowStrings($r));
}

rowValues($r) gives the current row as Values and rowStrings($r) as text. rowNumber($r) is one-based and 0 before the first row. For XLSX it is the row's own r number, so skipped empty rows show as gaps. For ODS it counts <table:table-row> elements, so a run written as one element with table:number-rows-repeated counts once.

"" as the sheet name takes the first sheet. openRowsFromBytes(raw, kind, name) does the same over bytes already in memory. The reader is a value over those bytes, not a file handle: advance it with $r = nextRow($r), and there is nothing to close.

You see each row once, in order. For random access, read the workbook.

#Memory

A 25 000-row, 20-column XLSX (500 000 cells, 15 MB of sheet XML), on one host:

Reading the whole workbook at this size is slower still, since it also builds a cell map entry for every cell. See Memory.

#Formats

ODS and XLSX stream. Gnumeric and delimited text are read whole and handed out row by row, so the same loop works for every format.