Conformance

CSV looks simple, but it is full of edge cases. csv-pipe handles them, and three layers of tests run on every commit.

A curated corpus

Over 70 cases, each paired with its expected output. They include what trips parsers up: quoted commas and line breaks inside a field, doubled quotes, byte-order marks, the various line endings, ragged rows, and Unicode.

A small sample, shown with header: false so each row is a plain array:

Input	Parsed
a,"b,c",d	["a", "b,c", "d"]
"a""b"	['a"b']
"line 1\nline 2"	["line 1\nline 2"]
a,b\r\nc,d (CRLF)	["a","b"], ["c","d"]
a\rb (lone CR)	["a"], ["b"]
a,,c	["a", "", "c"]
"a"junk,b	["a", "b"]
"ab (unterminated)	["ab"]
a,b (leading BOM)	["a", "b"]
café,日本,😀	["café", "日本", "😀"]

The full set lives in conformance.cases.ts.

The csv-spectrum suite

csv-pipe also passes csv-spectrum, a widely used third-party CSV test suite.

Property-based testing

Examples only reach the cases we anticipated. For the rest, fast-check runs a few properties over thousands of random inputs:

• It never throws. Any string produces a well-formed result.
• It round-trips. Encoding data and reading it back returns the original, even with commas, quotes, and newlines in the fields.
• It is unaffected by chunking. The same input split into varying chunk sizes parses identically to the whole.

Ambiguous inputs

A few inputs have no single right answer, so here is what csv-pipe does:

• It is forgiving by default: a short row is padded, a long row loses the extras, anything after a closing quote is ignored, and an unterminated quote runs to the end. Switch on strict to reject malformed rows.
• trim strips whitespace from every field, quoted ones included.
• A line that is only an empty quoted field ("") becomes one empty field. The default skipEmptyLines drops it with the blanks; skipEmptyLines: false keeps it.

See Options for each setting in full.

Run it yourself

git clone https://github.com/martsinlabs/csv-pipe
cd csv-pipe
npm install
npm test