Wrong offset in MIDIPacketList struct

[davidgiga1993]

Issue details

The struct definition for MIDIPacketList is wrong and only seems to work by accident on 32bit platforms.
When calling getPacket() on a MIDIPacketList struct the returned packet is off by 4 byte on a 64bit platform.

Example:
On 32bit

• Pointer of list: 6106767000
• Pointer of getPacket().getHandle(): 6106767004

On 64bit

• Pointer of list: 6106767000
• Pointer of getPacket().getHandle(): 6106767008
  Therefore all the data of the MIDIPacket is invalid on 64bit platforms.

The issue is that the packet is not referenced by ref in the original header - therefore the offset is always 4 byte (the numPackets field)

Reproduction steps/code

• Receive midi data using CoreMidi
• The length of any midi packet is way too long on 64bit platforms which shows the off by 4.

Configuration

Build Tools:

• Gradle plugin

Versions:

• Robovm: 2.3.7-SNAPSHOT
• XCode: 10.2.1
• JDK: 1.8.0_181

Build Targets:
iPhone 8 (64 bit)

Workaround

As of now I'm reading the packet with an hardcoded 4 byte offset but that's pretty ugly..

long listPointer = pktlist.getHandle();
MIDIPacket packet = MIDIPacket.toStruct(MIDIPacket.class, listPointer + 4);

[dkimitsa]

MIDIPacket uses non native struct member packing (#pragma pack(push, 4)), this is not supported by robovm itself. so all data gots natural alignment, which is 8 byte in case of LP64, will check if it hard to tweak struct member compiler a little to support this.

[davidgiga1993]

I see, thanks for the feedback! As a short term solution: Do you know how I could workaround this issue when creating a MIDIPacketList struct? The workaround I mentioned only applies when reading but not when creating this type.

[dkimitsa]

you can try manually build a structure in allocated memory (with VM.allocateMemory), and then creating java object for needed memory using MIDIPacket.toStruct(

[florianf]

@dkimitsa is this also fixed via #375?

[dkimitsa]

hi @florianf, no it requires packed struct support in compiler. I will deliver fix as another PR

[dkimitsa]

hi @davidgiga1993 , sadly it took more than month since reported. but there is a support for packed structs in #378 . Could you please test it and report any issues. Thanks

[dkimitsa]

@florianf @Tom-Ski it was already merged to master, can be closed