Fails on Mac OS X about trezor-agent HOT 8 CLOSED

A bit of background:
I had an issue reported (#6) for Mac OS X last week.
It seemed that there was a problem with some socket system calls on Mac OS X, so I have rewritten the exit logic using polling (instead of calling shutdown() on the listening socket) at fb0d0a5.
However, I have tested this only on my Debian workstation, so I am not sure this fix works on Mac OS.

from trezor-agent.

Does the Resource temporarily unavailable error happens all the time?
Does it happen always after the first list_pubs() reply is sent?
What version of SSH client are you using on Mac OS X?

from trezor-agent.

An idea: can you tweak server.UNIX_SOCKET_TIMEOUT constant (at https://github.com/romanz/trezor-agent/blob/master/trezor_agent/server.py#L13)?
Currently, it is set 100 ms, but it may be increased to a few seconds to check if it affects the error.

from trezor-agent.

It seems that EAGAIN on Mac OS X has errno=35:
http://stackoverflow.com/questions/14595269/errno-35-eagain-returned-on-recv-call

from trezor-agent.

If the problem is that the recv() system call is failing with EAGAIN due to a timeout, could you please try the following patch (to make sure that the accepted connection is set to blocking mode) on Mac OS?

diff --git a/trezor_agent/server.py b/trezor_agent/server.py
index 38835fe..a04a3a4 100644
--- a/trezor_agent/server.py
+++ b/trezor_agent/server.py
@@ -63,6 +63,7 @@ def server_thread(server, handler, quit_event):

     def accept_connection():
         conn, _ = server.accept()
+        conn.settimeout(None)
         return conn

     while True:

from trezor-agent.

You can also now set the timeout via a command-line argument: b101281

from trezor-agent.

@romanz The conn.settimeout(None) fixed it. I tried with both KeepKey and Trezor and they both work.

from trezor-agent.

Great!
I will push the fix to master.

from trezor-agent.