_PyTime_FromDouble() fails to detect an integer overflow when converting a C double to a C int64_t

Python 3.7.5 (default, Nov 20 2019, 09:21:52) 
[GCC 9.2.1 20191008] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import time
>>> time.sleep(9223372036.854777)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: sleep length must be non-negative

It's a bug in _PyTime_FromDouble() which fails to detect the integer overflow when casting a C double to a C _PyTime_T (int64_t): bug in _Py_InIntegralTypeRange(_PyTime_t, d) where d is a C double.

On my Fedora 31, double is a 64-bit IEEE 754 float, _PyTime_t is int64_t (64-bit signed integer).

_PyTime_t maximum is 9223372036854775807. But casted as C double, it becomes 9223372036854775808:

>>> int(float(9223372036854775807))
9223372036854775808

_PyTime_FromDouble() checks if!(_Py_DoubleInIntegralTypeRange(_PyTime_t, d)) with the macro:

#define _Py_InIntegralTypeRange(type, v) (_Py_IntegralTypeMin(type) <= v && v <= _Py_IntegralTypeMax(type))

and _Py_IntegralTypeMax(type)=2**63-1.

"v <= _Py_IntegralTypeMax(type)" compares a C double to a C int64_t: the compiler casts the C int64_t to a C double.

The problem is that 2**63-1 casted to a C double rounds using ROUND_HALF_EVEN rounding mode which gives a number *greater* than 2**63-1: we get 2**63.

To implement "v <= max", we have to round max towards zero (ROUND_DOWN), not round it using ROUND_HALF_EVEN.

I didn't find a way to control the rounding mode of casting C int64_t to C double, but we can round it *afterwards* using nextafter(max, 0.0) (ROUND_DOWN).

Similar issue in HHVM:

* https://github.com/facebook/hhvm/issues/5932
* https://github.com/PPC64/hhvm/commit/7cdb76b4f495aa7aa40a696379862916c27f5828

Oh, clang on FreeBSD spotted a similar bug in float.__trunc__()!

static PyObject *
float___trunc___impl(PyObject *self)
/*[clinic end generated code: output=dd3e289dd4c6b538 input=591b9ba0d650fdff]*/
{
    double x = PyFloat_AsDouble(self);
    double wholepart;           /* integral portion of x, rounded toward 0 */

    (void)modf(x, &wholepart);
    /* Try to get out cheap if this fits in a Python int.  The attempt
     * to cast to long must be protected, as C doesn't define what
     * happens if the double is too big to fit in a long.  Some rare
     * systems raise an exception then (RISCOS was mentioned as one,
     * and someone using a non-default option on Sun also bumped into
     * that).  Note that checking for >= and <= LONG_{MIN,MAX} would
     * still be vulnerable:  if a long has more bits of precision than
     * a double, casting MIN/MAX to double may yield an approximation,
     * and if that's rounded up, then, e.g., wholepart=LONG_MAX+1 would
     * yield true from the C expression wholepart<=LONG_MAX, despite
     * that wholepart is actually greater than LONG_MAX.
     */
    if (LONG_MIN < wholepart && wholepart < LONG_MAX) {  /* <=== HERE */
        const long aslong = (long)wholepart;
        return PyLong_FromLong(aslong);
    }
    return PyLong_FromDouble(wholepart);
}


Objects/floatobject.c:881:45: warning: implicit conversion from 'long' to 'double' changes value from 9223372036854775807 to 9223372036854775808
      [-Wimplicit-int-float-conversion]
    if (LONG_MIN < wholepart && wholepart < LONG_MAX) {
                                          ~ ^~~~~~~~
/usr/include/sys/limits.h:64:18: note: expanded from macro 'LONG_MAX'
#define LONG_MAX        __LONG_MAX      /* max for a long */
                        ^~~~~~~~~~
/usr/include/x86/_limits.h:64:20: note: expanded from macro '__LONG_MAX'
#define __LONG_MAX      0x7fffffffffffffff      /* max for a long */
                        ^~~~~~~~~~~~~~~~~~
1 warning generated.

Is it a bug in float.__trunc__()? It seems to me that the detailed comment accounts for case of rounding up.

That's why < is used instead of <=.

C99, 6.3.1.4 Real floating and integer:

"""
When a value of integer type is converted to a real floating type, if the value being
converted can be represented exactly in the new type, it is unchanged. If the value being
converted is in the range of values that can be represented but cannot be represented

exactly, the result is either the nearest higher or nearest lower representable value, chosen
in an implementation-defined manner. If the value being converted is outside the range of
values that can be represented, the behavior is undefined.
"""

So int64_t => double is implementation defined, but not undefined.

float___trunc___impl() looks correct to me (of course we can squash
that warning).

New changeset d3e53bc5321a9f08c7ed5b9383eefb2e03dfa6e2 by Victor Stinner in branch 'main':
bpo-39277: Fix PY_TIMEOUT_MAX cast in _threadmodule.c (GH-31195)
https://github.com/python/cpython/commit/d3e53bc5321a9f08c7ed5b9383eefb2e03dfa6e2