winerror_to_errno implementation #81886
Comments
|
OSError and _Py_fstat_noraise rely on winerror_to_errno() in PC/errmap.h in order to translate a Windows system error code into a POSIX errno value. PC/errmap.h is supposed to be generated by PC/generrmap.c, which is based on the old CRT's _dosmapperr() function. However, this function isn't implemented in the Universal CRT, and ucrt's corresponding function, __acrt_errno_from_os_error, isn't exported for public use. If generrmap.c is effectively dead, then it's no longer possible to add custom mappings there, as was done for bpo-12802 (ERROR_DIRECTORY -> ENOTDIR) and bpo-13063 (ERROR_NO_DATA -> EPIPE). Also, errmap.h hasn't been regenerated in 8 years, and since then the CRT added a new mapped value: ERROR_NO_UNICODE_TRANSLATION (1113) -> EILSEQ. Unless someone can suggest a way to continue automatically generating errmap.h via generrmap.c, then I think winerror_to_errno should be manually implemented in a more readable and maintainable way, and updated from the ucrt source (ucrt\misc\errno.cpp) with each major Python release. The implementation could use a switch statement like it currently does, but with named error codes, grouped by result. For example: int
winerror_to_errno(int winerror)
{
switch(winerror) {
case ERROR_FILE_NOT_FOUND: // 2
case ERROR_PATH_NOT_FOUND: // 3
case ERROR_INVALID_DRIVE: // 15
case ERROR_NO_MORE_FILES: // 18
case ERROR_BAD_NETPATH: // 53
case ERROR_BAD_NET_NAME: // 67
case ERROR_BAD_PATHNAME: // 161
case ERROR_FILENAME_EXCED_RANGE: // 206
return ENOENT; case ERROR_INVALID_FUNCTION: // 1
case ERROR_INVALID_ACCESS: // 12
case ERROR_INVALID_DATA: // 13
case ERROR_INVALID_PARAMETER: // 87
case ERROR_NEGATIVE_SEEK: // 131
default:
return EINVAL;
}
} |
eryksun
added
3.8 (EOL)
|
Special casing of Winsock error codes should be handled in winerror_to_errno rather than in oserror_parse_args, so it's consistently applied everywhere. And it should use the proper range for Winsock, 10000-11999 -- not 10000-2147483647. Codes above the Winsock range should map to EINVAL. It's not a non-issue since this range has the following blocks allocated: Another thing winerror_to_errno should do is unwrap HRESULT codes for Win32 errors (0x80070000-0x8007FFFF). In other words, 0x80070005 is effectively the same as ERROR_ACCESS_DENIED (5). It should be mapped to EACCES and raise PermissionError. For example: int
winerror_to_errno(int winerror)
{
/* Unwrap FACILITY_WIN32 HRESULT errors. */
if ((winerror & 0xFFFF0000) == 0x80070000) {
winerror &= 0x0000FFFF;
}
/* Winsock error codes (10000-11999) are errno values. */
if (winerror >= 10000 && winerror < 12000) {
return winerror;
}
switch(winerror) {
/* ... */
}
} |
|
While I'm inclined to think it's okay to find the CRT sources (e.g. "C:\Program Files (x86)\Windows Kits\10\Source\10.0.18362.0\ucrt\misc\errno.cpp") and extract the table from there as part of build, the problem I have is that it is woefully incomplete. I'd much rather have a manually managed table. It'd also be nice to expose it through the errno module somehow, probably just as a dict. It also seems like at least some of the WSA* constants (e.g. WSAEBADF==10009) are the equivalent errno plus 10000 (e.g.EBADF==9). Should we be mapping those back to the errno value? |
Mapping WSAEINTR (10004) -> EINTR (4) and WSAEACCES (10013) -> EACCES (13) would allow special casing them as InterruptedError and PermissionError. This is different from the aliases such as EWOULDBLOCK = WSAEWOULDBLOCK, but it should be fine since the six affected values are singled out in WinSock2.h as C errno values. So the Winsock section becomes a switch: /* Winsock error codes (10000-11999) are errno values. */
if (winerror >= 10000 && winerror < 12000) {
switch(winerror) {
case WSAEINTR:
case WSAEBADF:
case WSAEACCES:
case WSAEFAULT:
case WSAEINVAL:
case WSAEMFILE:
/* Winsock definitions of errno values. See WinSock2.h */
return winerror - 10000;
default:
return winerror;
}
} |
|
I have created a pull request for this issue. Please take a look. |
|
I'm going to merge and backport this all the way to 3.7. Adding a publicly accessible dict for use in Python code should be 3.9 only. |
|
New changeset 38e0b9e by Oleg Iarygin in branch 'main': |
Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.
Show more details
GitHub fields:
bugs.python.org fields:
The text was updated successfully, but these errors were encountered: