Skip to main content

Include/cpython/floatobject.h

Source:

cpython 3.14 @ ab2d84fe1023/Include/cpython/floatobject.h

Include/cpython/floatobject.h exposes the PyFloatObject struct layout and the endian-safe pack/unpack helpers used by marshal and struct. The public Include/floatobject.h provides the stable ABI.

Map​

LinesSymbolRole
1-20PyFloatObjectStruct layout; ob_fval field
21-45PyFloat_PACK2/4/8, PyFloat_UNPACK2/4/8Endian-aware IEEE 754 pack/unpack
46-60PyFloat_GET_VALUEInline accessor macro

Reading​

PyFloatObject layout​

PyFloatObject has a single meaningful field beyond the object header: ob_fval, a C double. There is no fractional part or sign bit stored separately; the entire IEEE 754 double precision value is stored in ob_fval.

// Include/cpython/floatobject.h:1 PyFloatObject
struct PyFloatObject {
    PyObject_HEAD
    double ob_fval;   /* the actual float value */
};

static inline double
PyFloat_GET_VALUE(PyObject *op) {
    return ((PyFloatObject *)op)->ob_fval;
}

Pack/unpack for serialization​

PyFloat_PACK8(p, x, le) writes the double x to 8 bytes at p in little-endian (le=1) or big-endian (le=0) byte order. These are used by struct.pack format codes d and f, and by marshal to write floats into .pyc files in a platform-independent byte order.

// Include/cpython/floatobject.h:21 PyFloat_PACK8
int PyFloat_PACK8(char *p, double x, int le);
double PyFloat_UNPACK8(const char *p, int le);
int PyFloat_PACK4(char *p, float x, int le);
float PyFloat_UNPACK4(const char *p, int le);

Free list​

Each interpreter maintains a free list of recycled PyFloatObject instances. PyFloat_FromDouble pops from the list before calling PyObject_Malloc; float_dealloc pushes to the free list (up to PyFloat_MAXFREELIST items). The free list is stored in _PyInterpreterState.float_state.

gopy notes​

The gopy float type wraps a Go float64. The pack/unpack helpers are not needed because Go's encoding/binary package handles endian-safe float serialization. The free list is replaced by Go's garbage collector.