No way to make a new thread

[mikrosk]

I've started to investigate this issue here: signal11/hidapi#93 -- as it turned out, it was a simple mistake. So definitely, python + hidapi have some trouble together.

I see both 'while 1' and 'while 2' but then the first thread (the main one) is frozen. raw_input() doesn't work anymore. If I put everything in the same thread, it works.

from threading import Thread
import hid

def receive_handler():
    print 'while 2'

    while True:
        report = handle.read(64)
        if report:
            #print HID_COMMAND(report[0]).toString()
            #if report[0] == HID_COMMAND.USB_HIDCMD_GET_DEVICETYPE:
            #    print "Version: {0}.{1}.{2}".format(report[49], report[50], report[51])
            print '{0} bytes read, first byte: {1}'.format(len(report), hex(report[0]))
        else:
            print 'no bytes read'

handle = hid.device(0x0911, 0x251c)

handle.set_nonblocking(False)

receive_thread = Thread(target=receive_handler)
receive_thread.start()

print 'while 1'

while True:
    key = raw_input()
    print 'key ' + key + 'pressed'
    if key == 'w':
        report = bytearray(65)
        report[1] = 0x82 # command which issues a reply from my device
        handle.write(report)

    elif key == 'q':
        break

handle.close()

[qdot]

Is HIDAPI even threadsafe like that? I thought the handle initialization had to live on the same thread?

[mikrosk]

According to issues/questions in the original signal11 tracker, it is. Specially when C implementation does work (see the linked issue above). How else you would poll for events?

[qdot]

There's many ways to poll for events without threads. That said, like you mentioned, according to signal11/hidapi#56 hidapi is threadsafe across platforms, so yeah, should probably work in python too. Could be that the cython structure wrapping hidapi isn't threadsafe or something. I currently know very little of cython (I forked because cython-hidapi was missing some functions I needed and the main repo is defunct), but I guess now is as good a time as any to learn. :)

[mikrosk]

I think its not caused by cython. I implemented also a C api version with direct calls to hidapi and it behaves same. It seems more like a collision of python and hidapi threading system but I can't imagine how that could happen.

[mikrosk]

After a lot of headache I managed to solve it. By a workaround, of course. I replaced hid_read() with hid_read_timeout + added a little sleep in receive_handler(). I still don't understand how this can work in plain C and cannot work in C Api and/or Cython but at least I've got a solution. If you find the real cause, please post it.