Victor Stinner pointed out that on x86 Gentoo Installed with X 3.x buildbot, there is a compiler warning:

Python/pystate.c:1483:18: warning: cast to pointer from integer of
different size [-Wint-to-pointer-cast]

(https://buildbot.python.org/all/#/builders/103/builds/2067)

This warning reveals a bug:

static int
_long_shared(PyObject *obj, _PyCrossInterpreterData *data)
{
    int64_t value = PyLong_AsLongLong(obj);
    if (value == -1 && PyErr_Occurred()) {
        if (PyErr_ExceptionMatches(PyExc_OverflowError)) {
            PyErr_SetString(PyExc_OverflowError, "try sending as bytes");
        }
        return -1;
    }
    data->data = (void *)value;

A 64-bit value is converted to void *, which is 32-bit on 32-bit platforms.

I've implemented a PR that uses Py_ssize_t instead to match the pointer size and to preserve the ability to work with negative numbers.

Is "long long" (AKA int64_t) 32 bits or 64 bits on a 32-bit platform?  It it's always 32 bits then there is no problem here.  Otherwise I agree we have a problem to fix.  My understanding is that it is the former (always 32 bits).

"long long" is mandated to be at least 64-bit by C99 (5.2.4.2.1 Sizes of integer types). If it were 32-bit, no warnings would have been issued.

What matters here is that _PyCrossInterpreterData.data type is void*. So the largest integer that you can store is intptr_t (signed) or uintptr_t (unsigned). In practice, sizeof(void*) == sizeof(intptr_t) == sizeof(uintptr_t) == sizeof(size_t) == sizeof(ssize_t). IMHO using size_t or ssize_t is a good solution, since it's well supported in Python.

Having slept on it and given the extra info you've provided I'm more comfortable with the proposed solution now. :)

So the result will be that on 32-bit architectures only 32-bit (signed) ints will be allowed through channels.  On 64-bit platforms (and higher) it will still be 64-bit ints.  That's a livable outcome, particularly since 32-bit platforms already have to deal with such variation already. :)  Plus, the consistency with sys.maxsize makes sense.

The alternative (on 32-bit arch) would be to send through a struct containing the long long, similar to what we do for str and bytes.  I don't think that's worth it.  If it matters we can do something like that later (since it's an implementation detail).

Consequently, I've approved the PR.  I'll merge it in a little while.

Thanks for working on this.

FTR, issue #34569 covered similar ground.

New changeset 16f842da3c862d76a1177bd8ef9579703c24fa5a by Eric Snow (Alexey Izbyshev) in branch 'master':
bpo-35972: _xxsubinterpreters: Fix potential integer truncation on 32-bit in channel_send() (gh-11822)
https://github.com/python/cpython/commit/16f842da3c862d76a1177bd8ef9579703c24fa5a

Thanks, Alexey!

> New changeset 16f842da3c862d76a1177bd8ef9579703c24fa5a by Eric Snow (Alexey Izbyshev) in branch 'master':
> bpo-35972: _xxsubinterpreters: Fix potential integer truncation on 32-bit in channel_send() (gh-11822)

It seems like this change introduced a reference leak: bpo-35984.

ack