[BRLTTY] [issue] HID braille device very slow under certain conditions

[Yannick PLASSIARD]

Dear all,

For a few weeks now, I have to use a Mac to do my job. I have two braille devices:
- a Focus 5 USB device
- A VarioUltra 40 USB-HID device

Before that, I used these two devices on Windows and Linux without any particular issues, either on a real Linux system with Brltty 6.6, or in a Linux VM hosted by Windows with the same Brltty 6.6.

On MacOS however I noticed something strange with my VarioUltra device: the transfer speed of the device (sending braille input / receiving braille dots to print) is very very slow. For example, if I type characters quickly, the device will freeze for several seconds and then the characters will be processed by BRLTTY, making it very inconfortable to use.

About my current setup:
- Running latest macOS Sonoma on an ARM64 (m2Pro) chip
- Running latest VMWare Fusion software version to use a Linux ARM64 Debian virtual machine
- Connect my braille device(s) to the VM and not to the Mac, so that BRLTTY can see and interact with them.


My observations:
- Using the Focus 40 5th generation connected to USB, everything works very well, nothing to complain about. I’m able to read and more importantly write very quickly  (typing at around 4 to 6 chars per second), and everything is responsive.
- Using the VarioULTRA 40 connected to USB-HID to the same VM, things are very sluggish: typing at the same speed will result in the situation described above, i.e. slowness / freeze of the device until either a key is pressed, or a certain time.


What I tried:
1) With the Focus 40:
- Works well everywhere: in a VM hosted by Windows/Mac, Linux host, and MacOS Host using the screen driver.
2) With the VarioULTRA device: Works well on a Linux VM hosted by Windows and linux Host, become sluggish on a Linux VM hosted by MacOS.
3) I tried to change hypervisors, and every software behaves the same, either with UTM or VMWare Fusion (UTM can use either QEMU backend for virtualisation, or Apple’s builtin virtualisation support).
4) I A few years ago, I noticed the very same behaviour with another USB-HID device (Brilliant BI 40X) connected to a real Linux host, but unfortunately was unable to investigate furthermore as I had this device for a couple of hours.

What I understood from the sources:
1) seems there is no Darwin HID support for now, so unable to test it with a compiled version for this OS with a screen driver, like I did with the Focus 40 5th Generation.
2) It seems that the async_io stack which is responsible for scheduling input/ouput events (and other things as well) does some polling so that it does not actually « sleeps » for a given time before reading data from the device. I tried to understand how timeouts are computed, to see if the issue could come from this computation, but I’m still in the early stage of understanding this stack and I currently did not find anything regarding this.

Therefore, here are my questions:
1) Does anyone using an USB-HID device already experienced such behaviours in the past, and if so, did you manage to solve it?
2) Which logs would be necessary to investigate this furthermore (log categories / level)?
3) If you’re thinking about something I did not try yet, feel free to point it out, I’ll be happy to try it.

Many thanks in advance for your answers,

Best,

Yannick