Re: Compile errors in libgcj while building cross-compiler for MinGW

[Ranjit Mathew <rmathew at hotmail dot com>]

Ranjit Mathew wrote:
> and w32api-2.0. I'm posting this to find out how Adam and the
> other MinGW hackers, if any, have got around these issues and are
> able to build a cross-compiler.
Am I the only one left here who plays with the mingw32 target?
Adam, where art thou?

:-(


> Second, java/net/natPlainDatagramSockImpl.cc uses the ECONNREFUSED
> errno code to check for refused connection. However, on Windows, not 
> only is
> this not defined, according to WinSock2 APIs the last error code can be
> retrieved using WSAGetLastError( ) and not checking for errno - then again
> it is WSAECONNREFUSED and not ECONNREFUSED.
More details on what I point out above can be found here:

http://msdn.microsoft.com/library/en-us/winsock/winsock/error_codes_errno_h_errno_and_wsagetlasterror_2.asp
http://msdn.microsoft.com/library/en-us/winsock/winsock/windows_sockets_error_codes_2.asp

The Windows Sockets functions sendto(), recv(), etc. return
SOCKET_ERROR if they fail, which (fortunately) is defined as "-1",
the same as that for BSD Sockets.

However, to check for the actual error code, you're supposed to
invoke WSAGetLastError( ) and not errno, and then again the
error constants are WSAECONNREFUSED, WSAENOTSOCK, etc. and
not ECONNREFUSED, ENOTSOCK, etc.

What I propose is the following:

1. We define an errno alias for networking code, say "sockerrno"
   and use it consistently in networking code that uses the Sockets
   API.

2. "posix.h" merely says "#define sockerrno errno"

3. "win32.h" says "#define sockerrno WSAGetLastError( )"

4. "win32.h" additionally says "#define ECONNREFUSED WSAECONNREFUSED",
   etc.

NOTE: This makes "sockerrno" an rvalue (a non-lvalue), because
to set it, on Windows you must call WSASetLastError( ).

How does this look? The existing code would be impacted minimally,
I feel, by this change.

If it looks all right, I can post a patch for the same.

Lastly, I guess this should have gone to the Classpath list as
well, especially Michael Koch, but I didn't copy it to
that list as I think there's a lot of overlapping subscriptions
between the lists. If you think I'm wrong, I'll also forward
it to at least Michael.

Sincerely Yours,
Ranjit.

--
Ranjit Mathew          Email: rmathew AT hotmail DOT com

Bangalore, INDIA.      Web: http://ranjitmathew.tripod.com/