`Include/cpython/floatobject.h`

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

Map

Symbol	Kind	Purpose
`PyFloatObject`	struct	Internal float object: single `fval` C `double` field
`PyFloat_AS_DOUBLE`	macro	Zero-check-free `fval` read; callers must guarantee type
`_PyFloat_IsInfinity`	function	True when the value is positive or negative infinity
`_PyFloat_IsNan`	function	True when the value is a NaN
`_PyFloat_FormatAdvancedWriter`	function	Format a float into a `_PyUnicodeWriter` with format spec
`_PyFloat_Pack2`	function	Encode a double as IEEE 754 binary16 into a byte buffer
`_PyFloat_Pack4`	function	Encode a double as IEEE 754 binary32 into a byte buffer
`_PyFloat_Pack8`	function	Encode a double as IEEE 754 binary64 into a byte buffer
`_PyFloat_Unpack2`	function	Decode IEEE 754 binary16 bytes to a C double
`_PyFloat_Unpack4`	function	Decode IEEE 754 binary32 bytes to a C double
`_PyFloat_Unpack8`	function	Decode IEEE 754 binary64 bytes to a C double

Reading

`PyFloatObject` layout and `PyFloat_AS_DOUBLE`

The struct is deliberately minimal: one PyObject_HEAD and one double. The macro PyFloat_AS_DOUBLE exposes this field without a type check, matching the convention used by PyTuple_GET_ITEM and friends.

/* Include/cpython/floatobject.h */
typedef struct {
    PyObject_HEAD
    double fval;
} PyFloatObject;

#define PyFloat_AS_DOUBLE(op) (((PyFloatObject *)(op))->fval)

In gopy the equivalent is the F field on objects.Float. The public AsDouble() accessor corresponds to the safe PyFloat_AsDouble() in the stable API; the field read directly matches PyFloat_AS_DOUBLE.

IEEE 754 pack and unpack helpers

The six _PyFloat_Pack* / _PyFloat_Unpack* functions implement Python's struct module binary encoding. Each function accepts a little-endian flag so the same code handles both byte orders.

int _PyFloat_Pack2(double x, char *p, int le);
int _PyFloat_Pack4(double x, char *p, int le);
int _PyFloat_Pack8(double x, char *p, int le);

double _PyFloat_Unpack2(const char *p, int le);
double _PyFloat_Unpack4(const char *p, int le);
double _PyFloat_Unpack8(const char *p, int le);

The Pack functions return 0 on success and -1 (with an exception set) on overflow. The Unpack functions return the decoded value or -1.0 with an exception on malformed input. In gopy these map to functions in the struct module port, not directly to objects/float.go, because they operate on raw byte slices rather than on Float objects.

Special-value predicates

static inline int
_PyFloat_IsInfinity(double x) {
    return isinf(x);
}

static inline int
_PyFloat_IsNan(double x) {
    return isnan(x);
}

Go's math.IsInf and math.IsNaN provide direct equivalents. objects/float.go delegates to those rather than reimplementing the bit patterns.

gopy mirror

objects/float.go defines type Float struct with a single F float64 field, mirroring PyFloatObject.fval. The PyFloat_AS_DOUBLE macro has no Go equivalent because field access is already safe and inlineable. The _PyFloat_FormatAdvancedWriter logic is handled by the string-formatting path in the compiler and is not part of the core Float object.

CPython 3.14 changes

3.14 does not change the PyFloatObject struct or the PyFloat_AS_DOUBLE macro. The only visible update in this header is that _PyFloat_FormatAdvancedWriter is annotated with _Py_NO_SANITIZE_MEMORY to silence memory-sanitizer false positives from the double-to-bytes reinterpretation in the pack functions.

Map​

Reading​

PyFloatObject layout and PyFloat_AS_DOUBLE​

IEEE 754 pack and unpack helpers​

Special-value predicates​

gopy mirror​

CPython 3.14 changes​

Map