Files
.CooCoo 4e0c880f58 craft-csv: delimiter detection samples 65,536 characters, not 64 KiB (#2326)
sniff_delimiter read the first 64 KiB of the text. A first field of
three-byte characters longer than about 21,800 characters filled that
sample before its delimiter, so a semicolon or tab file opened as
comma-separated. The sample is now the first 65,536 characters, the
amount detection read before it moved into craft-csv.

The new test fails without the change: a first field of 65,535
three-byte characters followed by a semicolon is detected as comma.
2026-11-18 23:20:42 -05:00
..

craft-csv

Delimited text (CSV, TSV and their relatives) and SYLK (.slk). No unsafe; the only dependency is thiserror.

Fields are text. Reading a field as a number or a date, deciding how a value is written, and mapping a SYLK sheet onto a document are the caller's work.

Delimited text

Function What it does
decode_text(bytes) UTF-8 and UTF-16 byte order marks; without one, valid UTF-8 is kept; other bytes are Windows-1256 when they read as Arabic, else Windows-1252
sniff_delimiter(text, quote) comma, semicolon or tab: the one with the most consistent field count over the first 20 records with other text in a sample of the first 65,536 characters; when none separates those records, the candidates whose records hold only their delimiters and whitespace (rows of empty cells) are compared the same way, leaving out any candidate that found a record with other text; when no candidate finds coherent complete records with a delimiter in each, a candidate that finds no complete record with other text counts the delimiters of the first such record up to the end of the sample, or up to a quoted field still open there (after its rows of empty cells, a record without its delimiter is left out), and a quote inside a field that does not start with one makes that record incoherent; quote boundaries must be coherent, and ties prefer tab, comma, then semicolon
parse_records(text, delim, quote, max_fields, on_record) records of fields, one callback per record (return false to stop); CR, LF and CRLF line ends; quoted fields with doubled quotes and line breaks; a quote in the middle of a field is text
records(text, delim, quote, max_records, max_fields) the same, collected
push_field(out, field, delim), field_len(field, delim) one field as written, quoted only when it holds the delimiter, a quote or a line break; its exact length, to enforce a size cap before writing

max_fields bounds what one record keeps, so a row of endless delimiters cannot allocate a string per field. The delimiter between fields and the line end are the writer's own.

SYLK

sylk::read(bytes) gives a sylk::Sheet: cells by (row, column) with a value, the formula in A1 notation (the file's R1C1 text with its references rewritten, not checked for syntax) and the number format code; column widths in characters; and the number of cell records skipped because their position is outside the grid. sylk::r1c1_formula_to_a1 is the rewrite on its own.

Files over 512 MB are refused, positions outside 1,048,576 rows by 16,384 columns are skipped, and at most 1,024 column-width records are kept.

Origin

Moved from GridCraft: crates/xlsx/src/csv.rs (decoding, sniffing, the record parser and the quoting of its writer) and crates/engine/src/io/sylk.rs, at 5cf52c5f. GridCraft keeps the parts that depend on its model: typed values and dates from fields, and the sheet it builds.