Win10 compatibility.

[perlindgren]

There seems to be some compatibility issue with Win10 and the generated descriptor(s).

#[gen_hid_descriptor(
    (collection = APPLICATION, usage = 0x01, usage_page = 0xff00) = {
        #[item_settings data] data=input;
    }
)]

#[allow(dead_code)]
struct Data {
    data: u8,
}

The generated descriptor looks like this.

data [
    0x6, 0x0, 0xff,      // USEGE PAGE Vendor
    0x9, 0x1,             // USAGE Vendor
    0xa1, 0x1,           // Collection (Application)
    0x15, 0x0,          // Logical minimum
    0x26, 0xff, 0x0,   // Logical maximum
    0x75, 0x8,          // Report size 8
    0x95, 0x1,          // Report count 1
    0x81, 0x0,          // Input
    0xc0,                 // End application collection
]

This works well under Linux (tested under arch) and OSX. Hower under Win10 the device is not properly detected/configured (USB Device Viewer shows the descriptors correctly but claims the devic to be in low power state.

Adding a usage before Input solves the issue.

data [
    0x6, 0x0, 0xff,      // USEGE PAGE Vendor
    0x9, 0x1,             // USAGE Vendor
    0xa1, 0x1,           // Collection (Application)
    0x15, 0x0,          // Logical minimum
    0x26, 0xff, 0x0,   // Logical maximum
    0x75, 0x8,          // Report size 8
    0x95, 0x1,          // Report count 1
    0x09, 0x01,        // USAGE Vendor <---- manually added here
    0x81, 0x0,          // Input
    0xc0,                 // End application collection
]

Either MS is out of specs (the wrapping usage should be inherited) or Linux/OSX just playing nice guessing, not sure.

Maybe there is already some workaround for this problem that I missed. If not, perhaps a syntax extension to the #[item_settings usage = 0x01, ...] would be possible. Or maybe let the proc macro always emit inherited usage from the wrapping collection (in this case).

Admittedly I'm a USB novice, so not sure what's best here.
/Per

[twitchyliquid64]

I think the wrapping usage is meant to be inherited right?

#8 seems similar.

If we can work out what Windows is doing, I'm happy to implement a workaround like what we did for #[quirks allow_short].

[perlindgren]

My understanding is the wrapping usage should be inherited (and that's the behavior of Linux/OSX). To make the generated descriptor work cross platform I can see two ways. Either add explicit syntax in the macro so that you can annotate the wrapped input/output (and possibly feature as well) with usage info, or alternatively have the macro to emit the inherited usage. Not sure what is the best here.

[twitchyliquid64]

I think a new quirks mode that emits the usage before each input/output element in the report would work?

Not sure how i feel about this as it might make the descriptor huge, but worth a try? wdyt?

[perlindgren]

Hmm, I worked around the problem by manually coding the descriptor, so I don't have a full picture on where its actually required for it to work cross platform. When data is later streamed, does the descriptor size make much difference you think?

[hiszd]

If you make that change to the descriptor and use it in Windows, does it still work in Linux?
Also, specifics on how your workaround was achieved would be great.

[perlindgren]

Hi.
It's been a while since I fiddled with this. As I recall it worked well on both win10, osx, and linux in the end. I don't have the midi stuff up and running currently. We are traveling tomorrow, but will be back later during the week, at that point I can try to check it for you.

Since its been a while, I assume the ecosystem including the usbdevice have progressed a bit, so maybe less of a hack is needed nowadays.

/Per