#Formats

sheet.save(sheet.open("in.xlsx"), "out.ods");

open and save choose the format by extension, openDetected by the file's bytes. readFormat and writeFormat take a Format, and each codec has its own bytes entry point (readOds, writeXlsx, ...).

sheet.formatOf("sales.xlsx");                  # Format.Xlsx
sheet.formatOfBytes(fs.readBytes("x.bin"));    # by signature
sheet.formatName(sheet.Format.Ods);            # "ods"

Extensions: .ods, .xlsx, .xlsm, .gnumeric, .gnm, .csv, .tsv, .tab. formatOfBytes recognises the two ZIP containers by their members, a gzip stream as Gnumeric, and uncompressed Gnumeric XML by its first 512 bytes. Delimited text has no signature and is not detected. Through readFormat and writeFormat, CSV and TSV are UTF-8, read with typing on, and written from the first sheet.

#What each format keeps

ODS and XLSX keep everything the model holds, in a round trip and in a conversion between them. Gnumeric loses a formula's cached result and turns a date into its serial number. ODS stores a format code as a tree of components, so the code is taken apart on write and rebuilt on read; see Number formats. Tests assert each of these losses.

#OpenDocument (.ods)

A ZIP container of XML. The reader tolerates a missing mimetype member (some writers omit it) and refuses one that declares another document type.

• Run-length encoding. ODS compresses empty space; a row is routinely padded with table:number-columns-repeated="16384". Runs that carry a value are expanded; empty runs only advance the position, so a small file does not become millions of cells. The writer compresses runs again.
• Mixed content. Styled text inside a cell is a <text:span>, a run of spaces is <text:s/>, and tabs and line breaks have their own elements. The reader keeps document order, so a<text:span>B</text:span>b reads as aBb, not abB.
• Number styles are read from both content.xml and styles.xml, and a column's table:default-cell-style-name applies to cells that name no style.

The writer puts mimetype first and uncompressed, as ODF requires; see The ZIP container.

ODF has no error type, so an error is written as the cell's text with no value type, and the reader turns it back into an error. A text cell that looks like an error code (#N/A) therefore reads back as an error.

#Office Open XML (.xlsx)

A ZIP container of XML with [Content_Types].xml, relationship parts, a shared string table and an indexed style sheet.

• Sheets are found through relationships, not by part name. Nothing says sheet1.xml, sheet2.xml are in tab order; the workbook's sheet list is.
• Text is in a shared table, one entry per distinct string. Rich text is split into <r> runs, which the reader joins.
• A date is a number with a date format. The reader resolves c/@s through cellXfs to a numFmtId, treats built-in formats 14-22 and 45-47 as dates, and checks a custom format code for date tokens outside quoted literals. 46297 and 2026-10-02 differ only in that style. The writer adds the two date formats it needs.
• Absent cells are not written. A cell carries its own reference, and the reader trusts r over position.

#Gnumeric

One gzipped XML document; an uncompressed one is read too.

Cells are addressed by zero-based Row and Col attributes. A formula is stored as its expression with a leading =, and Gnumeric recalculates on open, so there is no cached result to store.

Styles are a <gnm:Styles> block of rectangular regions before the cells, each naming a span of rows and columns and the style its cells take. The writer emits regions for styled cells only and merges adjacent cells with the same style into one region, the shape ssconvert writes. The reader never expands a region whose style sets nothing: Gnumeric fills a sheet's unused area with one region over all 16 384 columns and 1 048 576 rows, which would be 17 billion entries.

#Delimited text

See Delimited text.