Modules/_csv.c

Modules/_csv.c is the complete C implementation of the csv module. There is no pure-Python fallback for the core parsing and formatting logic. The file owns three public types: Dialect, Reader, and Writer.

Map

Lines	Symbol	Role
1–120	DialectObj, dialect_check_*	Dialect type definition and validation
121–400	parse_process_char, parse_save_field	reader state machine
401–600	Reader_iternext	reader iteration driver
601–850	join_append, join_append_data	writer field quoting
851–1100	csv_writerow, csv_writerows	writer public methods
1101–1800	csv_reader, csv_writer, module init	constructors and module def

Reading

Dialect: settings object

DialectObj stores all formatting parameters. Each field is validated at construction time by a dedicated dialect_check_* helper so that the hot parse loop can read them without extra checks.

// CPython: Modules/_csv.c:83 DialectObj
typedef struct {
    PyObject_HEAD
    char delimiter;
    char quotechar;
    char escapechar;
    int doublequote;
    int skipinitialspace;
    int strict;
    int quoting;
    PyObject *lineterminator;
} DialectObj;

parse_process_char: state machine core

The reader advances one character at a time through a switch on the current parser state. The states are START_RECORD, START_FIELD, IN_FIELD, IN_QUOTED_FIELD, QUOTE_IN_FIELD, ESCAPED_CHAR, and EAT_CRNL. This is the only place where the dialect settings are consulted during parsing.

// CPython: Modules/_csv.c:187 parse_process_char
static int
parse_process_char(ReaderObj *self, module_state *module_state, Py_UCS4 c)
{
    switch (self->state) {
    case IN_FIELD:
        if (c == self->dialect->delimiter) {
            if (parse_save_field(self) < 0) return -1;
            self->state = START_FIELD;
        }
        /* ... other transitions ... */
        break;
    case IN_QUOTED_FIELD:
        if (c == self->dialect->quotechar)
            self->state = QUOTE_IN_FIELD;
        break;
    }
    return 0;
}

Reader_iternext: pulling one row

Reader_iternext fetches the next line from the underlying iterator, feeds each character to parse_process_char, then collects the accumulated fields into a Python list. End-of-file is signaled by StopIteration.

// CPython: Modules/_csv.c:455 Reader_iternext
static PyObject *
Reader_iternext(ReaderObj *self)
{
    PyObject *lineobj = PyIter_Next(self->input_iter);
    if (lineobj == NULL) return NULL;
    /* feed each UCS4 code point through parse_process_char */
    fields = self->fields;
    self->fields = NULL;
    return fields;
}

join_append: quoting decision for the writer

join_append decides whether a field needs quoting based on the dialect's quoting setting and the presence of special characters, then delegates to join_append_data for the actual buffer construction.

// CPython: Modules/_csv.c:668 join_append
static int
join_append(WriterObj *self, PyObject *field, int quoted)
{
    Py_ssize_t rec_len;
    if (quoted)
        rec_len = join_append_data(self, kind, data, field_len, 1);
    else
        rec_len = join_append_data(self, kind, data, field_len, 0);
    return rec_len < 0 ? -1 : 0;
}

gopy notes

• The Go port lives in module/csv/. The state machine constants (START_FIELD, IN_QUOTED_FIELD, etc.) are reproduced as typed int constants in reader.go.
• DialectObj maps to a Dialect struct in Go. Validation helpers are ported one-for-one rather than replaced with a single generic validator.
• join_append_data performs in-place reallocation of a C buffer. The Go side uses strings.Builder and avoids manual sizing.

CPython 3.14 changes

• Dialect.lineterminator now raises TypeError (instead of silently accepting a non-string) when set to a non-str value, closing a long-standing inconsistency with the documented type.
• parse_process_char gained a QUOTE_NONE_ESCAPE branch to handle quoting=QUOTE_NONE together with an escapechar more precisely, fixing a regression introduced in 3.12.
• The module now exposes csv.QUOTE_STRINGS and csv.QUOTE_NOTNULL constants (added in 3.12 for the writer side) in the C layer rather than patching them in from Python.