Modules/_csv.c (part 3)

Source:

cpython 3.14 @ ab2d84fe1023/Modules/_csv.c ↗

This annotation covers the writer side. See modules_csv2_detail for csv.reader, csv.DictReader, and the dialect object.

Map

Lines	Symbol	Role
1-80	csv.writer	Create a writer that formats rows
81-180	Writer.writerow	Format and write one row
181-260	Writer.writerows	Write multiple rows
261-360	Quoting modes	QUOTE_ALL, QUOTE_MINIMAL, QUOTE_NONNUMERIC, QUOTE_NONE
361-500	Dialect registry	csv.register_dialect, csv.get_dialect

Reading

csv.writer

// CPython: Modules/_csv.c:980 csv_writer
static PyObject *
csv_writer(PyObject *module, PyObject *args, PyObject *keyword_args)
{
    WriterObj *self = PyObject_GC_New(WriterObj, &Writer_Type);
    self->writeline = NULL;
    PyObject *output_file;
    PyArg_UnpackTuple(args, "writer", 1, 1, &output_file);
    self->writeline = PyObject_GetAttrString(output_file, "write");
    self->dialect = (DialectObj *)_call_dialect(dialect_inst, keyword_args);
    return (PyObject *)self;
}

csv.writer(f) stores f.write for later calls. The output file only needs a write(str) method; it need not be an actual file. StringIO works: csv.writer(io.StringIO()) captures CSV output into a string.

Writer.writerow

// CPython: Modules/_csv.c:1040 Writer_writerow
static PyObject *
Writer_writerow(WriterObj *self, PyObject *seq)
{
    DialectObj *dialect = self->dialect;
    Py_ssize_t len = PySequence_Length(seq);
    for (Py_ssize_t i = 0; i < len; i++) {
        PyObject *field = PySequence_GetItem(seq, i);
        /* Convert to string */
        PyObject *str_field = PyObject_Str(field);
        /* Apply quoting based on dialect.quoting */
        join_append(self, str_field, i == len - 1);
        Py_DECREF(str_field);
        Py_DECREF(field);
    }
    PyObject *line = join_finalize(self);
    /* Append line terminator and write */
    PyObject *written = PyObject_CallOneArg(self->writeline, line);
    Py_DECREF(line);
    return written;
}

Each field is converted to a string via str(). Numeric fields with QUOTE_NONNUMERIC are quoted. The dialect's delimiter, quotechar, lineterminator govern formatting. join_append handles escaping of embedded quotes and delimiters.

Quoting modes

// CPython: Modules/_csv.c:680 quoting modes
/* QUOTE_MINIMAL (0): quote only if field contains delimiter, quotechar, or lineterminator */
/* QUOTE_ALL (1): always quote every field */
/* QUOTE_NONNUMERIC (2): quote non-numeric fields; reader converts unquoted to float */
/* QUOTE_NONE (3): never quote; use escapechar for special chars */
/* QUOTE_NOTNULL (4): like QUOTE_MINIMAL but skip None/empty */
/* QUOTE_STRINGS (5): quote all string (non-numeric) fields */

QUOTE_NONNUMERIC is useful for CSV interchange with tools that auto-detect types: unquoted fields will be parsed as floats by csv.reader. QUOTE_NONE requires an escapechar to be set, otherwise special characters cause Error.

Dialect registry

// CPython: Modules/_csv.c:1220 csv_register_dialect
static PyObject *
csv_register_dialect(PyObject *module, PyObject *args, PyObject *kwargs)
{
    PyObject *name_obj;
    PyArg_UnpackTuple(args, "register_dialect", 1, 2, &name_obj, ...);
    if (!PyUnicode_Check(name_obj)) {
        PyErr_SetString(PyExc_TypeError, "dialect name must be a string");
        return NULL;
    }
    DialectObj *dialect = (DialectObj *)_call_dialect(dialect_inst, kwargs);
    PyDict_SetItem(module_state->dialects, name_obj, (PyObject *)dialect);
    Py_RETURN_NONE;
}

csv.register_dialect('pipes', delimiter='|') stores a dialect by name. The built-in names are 'excel' (default), 'excel-tab', and 'unix'. csv.get_dialect(name) retrieves a registered dialect; csv.list_dialects() returns all registered names.

gopy notes

csv.writer is module/csv.Writer in module/csv/module.go (not yet ported; csv module uses the pure-Python fallback). The dialect registry is a Go map[string]*Dialect. Writer.writerow builds the output string using strings.Builder and calls the file's write method via objects.CallMethod.