After further investigation, both the original code and Nigel's patch
seem to work on my machine (WinXP, MAudio 1x1 USB MIDI), although I
don't have a USB analyzer to see if garbage is being spewed at the USB
device. The proposed code is essentially:
m->hdr->dwBytesRecorded = 0;
m->hdr->dwBufferLength = m->sysex_byte_count;
m->error = midiOutLongMsg(m->handle.out, m->hdr,
sizeof(MIDIHDR));
A relatively minor problem with this is that if you set dwBufferLength,
you effectively reduce the available buffer space, which might force
needless allocations of larger buffers. An improvement is to restore
dwBufferLength after sending the message. Also, it makes me nervous to
set dwBytesRecorded to zero when Microsoft says to set it to the buffer
length. The code works either way in my tests, but it seems more
conservative to set BOTH dwBufferLength and dwBytesRecorded to the
desired count. I'm changing the code to something like this:
long real_length = m->hdr->dwBufferLength;
m->hdr->dwBytesRecorded = m->sysex_byte_count;
m->hdr->dwBufferLength = m->sysex_byte_count;
m->error = midiOutLongMsg(m->handle.out, m->hdr,
sizeof(MIDIHDR));
m->hdr->dwBufferLength = real_length; /* restore length */
Even stranger is that for the case of sysex messages via the
midiStreamOut() function, m->hdr->dwBytesRecorded is definitely
important (zero does not work), and the line of code that rounds
dwBytesRecorded up to a multiple of 4 is also necessary. I'm not
changing that.
Stay tuned for updates to PortMidi. If you want the latest code with
these sysex changes immediately, let me know.
Thanks again to Nigel Brown for pointing out this subtle problem.
-Roger
_______________________________________________
media_api mailing list
[email protected]
http://www.create.ucsb.edu/mailman/listinfo/media_api
