I followed this advice how to build portmidi using MinGW on WinXP, and it worked:
http://cratel.wichita.edu/cratel/cratel pyportmidi#build_portmidi_.28MinGW.29
It's unfortunate that this Makefile is not distributed with portmidi.
Sending SysEx with portmidi worked on both OS X and WinXP.
As a side note, one of RtMidi's objectives: "object oriented C++ design" fails to really hold. RtMidi defines a abstract root class RtMidi and extends it by RtMidiIn and RtMidiOut. Defining a common super interface for both sending and receiving midi doesn't hold up very well, as the sample program sysextest.cpp exemplifies:
bool chooseMidiPort( RtMidi *rtmidi )
{
bool isInput = false;
if ( typeid( *rtmidi ) == typeid( RtMidiIn ) )
isInput = true;
What good is the RtMidi abstraction if just a few lines further down you break it again?
Furthermore I'd consider it both from an ethical and an engineering point of view a cleaner design to openly use portmidi instead of cannibalizing some of its code (see RtMidi.cpp):
// This function was submitted by Douglas Casey Tucker and apparently
// derived largely from PortMidi.
CFStringRef EndpointName( MIDIEndpointRef endpoint, bool isExternal )
...
// This function was submitted by Douglas Casey Tucker and apparently
// derived largely from PortMidi.
static CFStringRef ConnectedEndpointName( MIDIEndpointRef endpoint )
