Fully Asyncio Support

[ethe]

I find #246 this issue so I wrote an asyncio support for thriftpy, and it can pass all the original test cases. Now you can use thriftpy like this:

######
# This is a thrift client
######
import thriftpy
import asyncio
from thriftpy.rpc import make_aio_client


echo_thrift = thriftpy.load("echo.thrift", module_name="echo_thrift")


async def request():
    client = await make_aio_client(
        echo_thrift.EchoService, '127.0.0.1', 6000)
    print(await client.echo('hello, world'))
    client.close()

######
# This is a thrift server
######
import asyncio
import thriftpy

from thriftpy.rpc import make_aio_server

echo_thrift = thriftpy.load("echo.thrift", module_name="echo_thrift")


class Dispatcher(object):
    async def echo(self, param):
        print(param)
        await asyncio.sleep(0.1)
        return param


def main():
    server = make_aio_server(
        echo_thrift.EchoService, Dispatcher(), '127.0.0.1', 6000)
    server.serve()


if __name__ == '__main__':
    main()

[wooparadog]

I'm not too familiar with how to write an asyncio library. But this pr tries to solve the same problem as #299. But #299 has a much smaller changeset.

Could you please provide more information?

[wooparadog]

This file contains too many copied code. At least some of them can be referenced instead of copied

[ethe]

@wooparadog The asyncio support of #299 based on TFramedTransport transport which is easier to be implemented. This transport tries to read the whole message out from socket at the begin and save it into buffer, it is easier because it just needs yield once from reading the whole message. My implementation based on another transport (TBinaryTransport) so it may be a little bit verbose.

[wooparadog]

If it's unavoidable for simpler abstraction, there should be a better code path like:

• contrib/aio/binary_transport
• contrib/aio/framed_transport

but it hurts so much for all those duplicate code...

[damnever]

thriftpy simply ignored seqid, but we better fix it.

[ethe]

I think it's not a problem about asyncio support, maybe we can create another pull request to solve it both sync and async module.

[thedrow]

This is awesome! Can we merge and release this?

[ethe]

Yeah I think it can be merged smoothly, I just resolved the last little conflict.

[ethe]

@wooparadog The different between two pull requests is they realized different transport (I support bufferd and #299 supports memory bufferd). In my opinion, the buffered transport realization is more difficult than framed transport on this feature. And I also find that there are some blocking I/O at the low level not be yielded out in #299 realization (read_frame, etc.), cause it still used blocking socket, and this pull request is fully non-blocking (based on non-blocking stream reader and writer).

[HIRANO-Satoshi]

@hit9 Could you merge this PR?

[hit9]

Sorry but I am removed from this organization.