midiOutUnprepareHeader intermittently hangs on Windows during repeated large sysex sends

[insolace]

I tried submitting a pull request but did not have permissions.

Starting at line 3184 of RtMidi.cpp:

// Unprepare the buffer and MIDIHDR.
while ( MIDIERR_STILLPLAYING == midiOutUnprepareHeader( data->outHandle, &sysex, sizeof ( MIDIHDR ) ) ) Sleep( 1 );

This line of code is problematic and can cause our application to crash during our firmware update process, when we are sending multiple reboot commands in between large sysex payloads. Our current workaround is to delete the RtMidi in/out objects and then recreate them, otherwise depending on the state of the update process things can hang.

We'd prefer if there was better error handling and/or a timeout inside of this loop. Here is a simple timeout that would be better if there was a method to change the duration:

int timeOut = 0;
// Unprepare the buffer and MIDIHDR.
while ( MIDIERR_STILLPLAYING == midiOutUnprepareHeader( data->outHandle, &sysex, sizeof ( MIDIHDR ) ) )
{
    if (timeOut++ > 25000)
        break;
    Sleep( 1 );
}

[insolace]

It looks like another issue references this loop but did not encounter the same bug the we did:

#250

Debugging this further, my suggested timeout code does not actually prevent the application from hanging/crashing. It would seem that having a device reboot after receiving a large sysex payload can create situations where calling midiOutUnprepareHeader hangs, and there's no method in the API to check the validity of the handle or the buffer before calling it.

Our current workaround is to send the firmware, wait for the midi ports to change to confirm that the device has rebooted, and then stop the application with a user dialog/ok that then kills and relaunches the application. Otherwise if we try to send sysex without the kill/relaunch, there's a 50/50 chance that calling midiOutUnprepareHeader will crash the application, even if we delete / recreate the rtmidi object.