In Windows, os.device_encoding() is hard coded to map file descriptor 0 and descriptors 1 and 2 respectively to the console's input and output code page if isatty(fd) is true. But isatty() is true for any character device, such as "NUL". Also any fd might be a console, by way of dup(), dup2() -- or open() with the device names "CON", "CONIN$", and "CONOUT$".
The correct device encoding of a console file is needed for use with os.read() and os.write(). It's also necessary for io.TextIOWrapper() if PYTHONLEGACYWINDOWSSTDIO is set.
_Py_device_encoding() in Python/fileutils.c should use _get_osfhandle() to get the OS handle, and, if it's a character-device file, determine the code page to return, if any, depending on whether it's an input or output console file. For example:
PyObject *
_Py_device_encoding(int fd)
{
#if defined(MS_WINDOWS)
HANDLE handle;
DWORD temp;
UINT cp = 0;
_Py_BEGIN_SUPPRESS_IPH
handle = (HANDLE)_get_osfhandle(fd);
_Py_END_SUPPRESS_IPH
if (handle == INVALID_HANDLE_VALUE ||
GetFileType(handle) != FILE_TYPE_CHAR)
Py_RETURN_NONE;
Py_BEGIN_ALLOW_THREADS
/* GetConsoleMode requires a console handle. */
if (!GetConsoleMode(handle, &temp)) {
/* Assume access denied implies output. */
if (GetLastError() == ERROR_ACCESS_DENIED)
cp = GetConsoleOutputCP();
} else {
if (GetNumberOfConsoleInputEvents(handle, &temp)) {
cp = GetConsoleCP();
} else {
cp = GetConsoleOutputCP();
}
}
Py_END_ALLOW_THREADS
if (cp == CP_UTF8) {
return PyUnicode_FromString("UTF-8");
} else if (cp != 0) {
return PyUnicode_FromFormat("cp%u", (unsigned int)cp);
} else {
Py_RETURN_NONE;
}
#else
if (isatty(fd)) {
return _Py_GetLocaleEncodingObject();
} else {
Py_RETURN_NONE;
}
#endif /* defined(MS_WINDOWS) */
}
|