Typing and validation

Parsed fields are strings by default, and parse<T> is a type assertion, like JSON.parse(...) as T: it shapes the result type but does not change values at runtime. Two options bridge that gap.

Dynamic typing

Off by default. When on, a value is coerced to a number or boolean only when it round-trips exactly, so no data is silently lost. Only finite numbers are coerced.

parse('id,active\n42,true', { dynamicTyping: true });
// [{ id: 42, active: true }]

Field	With dynamicTyping	Why
42	42 (number)	round-trips exactly
true	true (boolean)	recognized literal
007	"007" (string)	would lose the leading zero
1.50	"1.50" (string)	would lose the trailing zero
1e3	"1e3" (string)	would not round-trip to 1e3
Infinity	"Infinity" (string)	not a finite number
TRUE	"TRUE" (string)	only lowercase true/false

Because typing is per field, it is best for known-numeric data. For full control, use the row hook below.

Validation with the row hook

For real runtime safety, pass a row hook. It receives each raw record and returns the final element, so a schema library or a hand-written guard enforces the type. This is the parsing analogue of the format hook on the encoder.

import { parse } from 'csv-pipe';

const users = parse('id\n1\n2', {
  row: (record) => ({ id: Number(record.id) })
});
// [{ id: 1 }, { id: 2 }]

The hook also receives a context with the zero-based rowIndex of the record, useful for error messages:

parse('email\na@b.com', {
  row: (record, { rowIndex }) => {
    if (!record.email.includes('@')) {
      throw new Error(`Invalid email on row ${rowIndex}`);
    }
    return record;
  }
});

A schema library fits the same slot. For example, with a validator that exposes a parse method, return schema.parse(record) from the hook to get validated, fully typed records.