I've just seen something that I've never seen before:
1. reboot Windows (with keyboard switched off).
2. start Chrome.
(see screenshot)
a) The TS22 sound card appeared as an available, selected input device (!) (The keyboard is plugged in to the TS22's MIDI IN, but is switched off.)
b) There are multiple copies of the VirtualMIDISynth in the list of output devices. 

Tried again: Rebooted Windows, and restarted Chrome. The behaviour was as it should be with the input device selector set to "choose a MIDI input device", the output device selector set to "choose a MIDI output device" and the available output devices correct (as in the "Initial state" in the original report (above).

I've also noticed that repeatedly turning the keyboard on and off increases the number of entries for the VirtualMIDISynth in the list of available output devices.

Deleted: 1. ChromeInitialDevices.png 4.9 KB

	1. ChromeInitialDevices.png 4.9 KB View Download

I get occasional problems with the Windows either not recognising the keyboard at all or not finding it again after being in sleep mode. So, suspecting that the problem may lie in a flaky keyboard driver, I tried updating it using the Windows device manager. (screenshot 1)
After searching the web, I was told that the driver is already up to date (screenshot 2).

It seems a bit odd that Windows classes "Audio, Video and Games controllers" as loudspeakers (output devices) rather than as input devices (see screenshot 1 and screenshot 2). Maybe that's not important, but I just thought I'd mention it.

Translation of screenshot 2: "USB-Audiogerät" means "USB Audio device". (My keyboard is NOT an audio device.)

Deleted: DeviceManager.png 18.0 KB

	DeviceManager.png 18.0 KB View Download

Deleted: KeyboardDriverUpdateDialog.png 9.5 KB

	KeyboardDriverUpdateDialog.png 9.5 KB View Download

+toyoshim

toyoshim@, is this still an unconfirmed issue? Could you please help triage or route this appropriately?

I confirmed that this issue happens on Windows 10 (probably with a specific Windows update) when the Windows MIDI system goes in a wrong state. The system actually returns multiple device entries and system reboot could fix this problem in my case.

Also VirtualMIDISynth is a third party driver that behaves differently from what our implementation assumes. I.e., the port can be opened even after the port is exclusively opened. That could be a reason of this problem.

I'm currently rewriting the backend to fix several problems (issue 672793). I believe it will fix this problem.

j.ingram:
Can you try Canary channel with following flag explicitly being flipped like below?
Implementation wasn't finished yet, but you could confirm if device enumeration works correctly. (sending and receiving functionalities may be not in the Canary yet)

chrome://flags/#enable-midi-manager-dynamic-instantiation => Enabled
chrome://flags/#use-winrt-midi-api => Disabled

Yes, I can confirm that device enumeration is now working correctly in Canary with the flags set as above.
More precisely: Live switching my keyboard works correctly in both input and output menus.
The VirtualMIDISynth is also fine.

Thanks!

Today I submitted all planned changes for m58.
So, probably I could enable this new backend on the m58.
(Now, it is enabled on Canary and Dev for 50% users)

I implemented new Web MIDI backend for Windows from scratch, and I believe this new backend can fix all long-living problems.

I want to figure out and fix all critical bugs while m58 is in Canary, Dev, and Beta. So, I'm very happy if you continue to play with this new mode.

@toyoshim: Please keep me posted about changes to the Canary implementation. I normally develop using Chrome, but will be happy to try things out as they appear. As you say, sending and receiving functionalities are not there yet...

You can track this design change progress by monitoring the issue 672793 (please check the star icon for mail notification).

All chromium side changes including new mode support for other platforms will be posted there, and I will post comments when field trial configuration is changed.

Thanks!

Understood. :-)