On Fri, 27 Oct 2023 18:35:33 GMT, Sergey Bylokhov <s...@openjdk.org> wrote:
>> JVM attempts to reuse the buffer for sending MIDI out data when the buffer
>> size is enough. It use `dwBytesRecorded` in `MIDIHDR` structure to indicate
>> the actual size of the data. However, `midiOutLongMsg()` ignores
>> `dwBytesRecorded`, although it did not mentioned in the documentation. I've
>> tested on Windows 7, 10 and 11. All of them have the same behavior.
>>
>> The bug cannot be easily reproduced because some MIDI drivers filter out any
>> malformed MIDI data. The example code below create a special case to make
>> sure all MIDI data are legally when the bug is triggered.
>>
>>
>> import javax.sound.midi.*;
>>
>> public class MidiTest {
>> public static class RawMidiMessage extends MidiMessage {
>> public RawMidiMessage(byte[] data) {
>> super(data);
>> }
>>
>> @Override
>> public Object clone() {
>> return new RawMidiMessage(this.getMessage());
>> }
>> }
>>
>> public static void main(String[] args) {
>> var deviceInfos = MidiSystem.getMidiDeviceInfo();
>> for (var info : deviceInfos) {
>> try (MidiDevice device = MidiSystem.getMidiDevice(info)) {
>> if (device.getMaxReceivers() != 0) {
>> System.out.println("Open MIDI port: " + info.getName());
>> device.open();
>> Receiver receiver = device.getReceiver();
>> // Send two sysex messages at once
>> receiver.send(new RawMidiMessage(new byte[]{
>> (byte) 0xF0, 0x7D, 0x01, (byte) 0xF7,
>> (byte) 0xF0, 0x7D, 0x02, (byte) 0xF7
>> }), -1);
>> // Send another sysex message
>> receiver.send(new RawMidiMessage(new byte[]{(byte) 0xF0,
>> 0x7D, 0x03, (byte) 0xF7}), -1);
>> }
>> } catch (MidiUnavailableException e) {
>> e.printStackTrace();
>> }
>> }
>> }
>> }
>>
>>
>> The expected messages received should be the following three messages
>>
>> F0 7D 01 F7
>> F0 7D 02 F7
>> F0 7D 03 F7
>>
>>
>> But acually four messages was received with the second message repeated
>> twice.
>>
>> F0 7D 01 F7
>> F0 7D 02 F7
>> F0 7D 03 F7
>> F0 7D 02 F7
>>
>>
>> To resolve the issue, I add a new variable to backup the actual buffer size
>> and set `dwBufferLength` of `MIDIHDR` structure to the size of MIDI data.
>> After calling `midiOutLongMsg()`, I restore the original buffer size if the
>> buf...
>
> I think the java code in the description can be added as a testcase, even if
> it works fine before the patch, it is useful that the affected code will be
> triggered.
> As an example you can use the tests in the next folder:
> https://github.com/openjdk/jdk/tree/master/test/jdk/javax/sound
Thanks for the suggestions from Florian Bomers and @mrserb.
I’ve added a new structure to record the actual buffer length. Therefore the
actual buffer length doesn’t need to write back to dwBufferLength.
Additionally, The example code is added as a testcase.
Please let me know if there are any further improvements needed.
-------------
PR Comment: https://git.openjdk.org/jdk/pull/16399#issuecomment-1784126362
