Re: [Libusb-win32-devel] usb_interrupt_read timeouts

[Dan E. <dan...@ne...>]

Dave Higton wrote:
>
> In the Linux version of libusb, the usb_interrupt_read () function can
> accept a value of 0 for the timeout, which is interpreted as "infinite".
> In the win32 version, this is not accepted.  OK, I can put any finite
> timeout that I choose, and loop round infinitely on a separate thread,
> putting any resulting input into a FIFO.
>
> But what happens if a USB message comes in immediately after the
> timeout, but before the new call?  Is any such message lost?
>
After the timeout, the host will no longer be sending IN requests, so 
the device will not be able to return any data.

There's possibly a tiny hole - after the timeout has elapsed, libusb 
then cancels the IO. The IO could have validly completed between these 2 
events, but the data would be lost.

  if(WaitForSingleObject(c->ol.hEvent, timeout) == WAIT_TIMEOUT)
    {
       /* request timed out */
      if(cancel)
        {
          _usb_cancel_io(c);

       /* wait until the request is cancelled */
       WaitForSingleObject(c->ol.hEvent, 0);
        }

      usb_error("usb_reap: timeout error");
      return -ETIMEDOUT;
    }


Dan.