I have an existing ipython kernel, with a communication file 'path/comm_file.json' and I want to execute code in this kernel using the Kernel Client API (actually I'm not picky, any method will do..). I understood that this is the best way to do things from the jupyter documentation. So I write the following code:
from jupyter_client import KernelClient
client = KernelClient(connection_file='path/comm_file.json')
client.execute('a = 10')
But the execute method leads to the following error:
File "C:\Python27\lib\site-packages\jupyter_client\client.py", line 249, in execute
self.shell_channel.send(msg)
File "C:\Python27\lib\site-packages\jupyter_client\client.py", line 143, in shell_channel
socket, self.session, self.ioloop
TypeError: object.__new__() takes no parameters
What am I doing wrong here??
I am also trying to figure out how the client works. Here's a place to start:
For a simple blocking client you can have a look a how jupyter_test_client and jupyter_console works.
from pprint import pprint
from jupyter_client.consoleapp import JupyterConsoleApp
class MyKernelApp(JupyterConsoleApp):
def __init__(self, connection_file, runtime_dir):
self._dispatching = False
self.existing = connection_file
self.runtime_dir = runtime_dir
self.initialize()
app = MyKernelApp("connection.json", "/tmp")
kc = app.kernel_client
kc.execute("print 'hello'")
msg = kc.iopub_channel.get_msg(block=True, timeout=1)
pprint(msg)
You will need helper functions to handle properly the zmq channels and json messages.
I was able to make a simple and bare KernelClient
work for me with this:
from jupyter_client.blocking import BlockingKernelClient
kc = BlockingKernelClient(connection_file='path/comm_file.json')
kc.load_connection_file()
kc.start_channels()
msgid = kc.execute('a = 10')
reply = kc.get_shell_msg(timeout=5)
That's indeed how JupyterConsoleApp
(used by jupyter_console) initializes its client when an existing kernel file is given.