[herniqeu]

Root cause

magic/compat.py builds a ctypes.c_int (4 bytes) buffer for magic_setparam/magic_getparam, but libmagic dereferences that pointer as a size_t (ms->bytes_max = *(const size_t *)value in file's func.c):

    def getparam(self, param):
        v = c_int()
        i = _getparam(self._magic_t, param, byref(v))
...
    def setparam(self, param, value):
        v = c_int(value)
        return _setparam(self._magic_t, param, byref(v))

On 64-bit this has three real effects: setparam silently truncates values above 32 bits and still reports success, getparam lets libmagic write 8 bytes into a 4-byte buffer, and the write path hands libmagic 4 uninitialized adjacent bytes. magic/__init__.py already uses c_size_t (it was fixed there before); only the compat layer, added later, was left with c_int.

Reproduction (Python 3.11, libmagic 5.44, Linux x64)

import magic.compat as c
m = c.open(c.MAGIC_NONE)
m.setparam(c.PARAM_BYTES_MAX, 2**33)   # returns 0 ("success")
m.getparam(c.PARAM_BYTES_MAX)          # master: 0  (bytes_max became 0, which
                                       #         zeroes libmagic's read limit)
                                       # expected: 8589934592

Setting bytes_max to 0 breaks identification for every later from_* call on that handle, so the truncation is not cosmetic.

Fix

Use c_size_t in the compat getparam/setparam, as magic/__init__.py already does.

Coverage

New test_compat_params_use_size_t in test/python_magic_test.py: sets PARAM_BYTES_MAX to 2**33 and reads it back. It fails on master (AssertionError: 8589934592 != 0) and passes with the patch. Full suites: python_magic_test 14 tests OK (2 skipped), libmagic_test 3 tests OK.

(The test file requires export LC_ALL=en_US.UTF-8, as the suite's own guard asks.)