go-python / gpython

gpython is a python interpreter written in go "batteries not included"
BSD 3-Clause "New" or "Revised" License
869 stars 95 forks source link

Getting started on asyncio #178

Open Tatskaari opened 2 years ago

Tatskaari commented 2 years ago

I've been trying to understand how this project works, and also how we can support newer language versions. I've done some reading into how asyncio works, and I thought I would share that here. Sorry if this is not news to anybody, I'm not much of a python developer.

The co-routine introduced in python3.5 seem to be the biggest language feature we need to support to become current. The op codes can be found here: https://docs.python.org/3/library/dis.html#opcode-GET_AWAITABLE

A basic example program looks like:

import asyncio

async def foo():
    return "test"

async def bar():
    return await foo()

print(asyncio.run(bar()))

And produces the following bytecode:

  1           0 LOAD_CONST               0 (0)
              2 LOAD_CONST               1 (None)
              4 IMPORT_NAME              0 (asyncio)
              6 STORE_NAME               0 (asyncio)

  3           8 LOAD_CONST               2 (<code object foo at 0x0000021C72C52130, file "test.py", line 3>)
             10 LOAD_CONST               3 ('foo')
             12 MAKE_FUNCTION            0
             14 STORE_NAME               1 (foo)

  6          16 LOAD_CONST               4 (<code object bar at 0x0000021C72C521E0, file "test.py", line 6>)
             18 LOAD_CONST               5 ('bar')
             20 MAKE_FUNCTION            0
             22 STORE_NAME               2 (bar)

 10          24 LOAD_NAME                3 (print)
             26 LOAD_NAME                0 (asyncio)
             28 LOAD_METHOD              4 (run)
             30 LOAD_NAME                2 (bar)
             32 CALL_FUNCTION            0
             34 CALL_METHOD              1
             36 CALL_FUNCTION            1
             38 POP_TOP
             40 LOAD_CONST               1 (None)
             42 RETURN_VALUE

Disassembly of <code object foo at 0x0000021C72C52130, file "test.py", line 3>:
              0 GEN_START                1

  4           2 LOAD_CONST               1 ('test')
              4 RETURN_VALUE

Disassembly of <code object bar at 0x0000021C72C521E0, file "test.py", line 6>:
              0 GEN_START                1

  7           2 LOAD_GLOBAL              0 (foo)
              4 CALL_FUNCTION            0
              6 GET_AWAITABLE
              8 LOAD_CONST               0 (None)
             10 YIELD_FROM
             12 RETURN_VALUE

The byte code of calling the functions is actually unchanged. The functions marked with async just have to return a co-routine object. The GET_AWAITABLE opcode seems to cast the other await-able types e.g. generators or classes that implement self.__await__.

We might be able to implement this fairly easily by just wrapping the Function type in a co-routine type. To get any actual concurrency out of this implementation though, we need to implement the event loop API.

https://docs.python.org/3/library/asyncio-eventloop.html#:~:text=The%20event%20loop%20is%20the,asyncio%20functions%2C%20such%20as%20asyncio.

This is the hard part which actually allows creation of futures and tasks. I'm not sure if the await for byte code that was already implemented in 3.4 has all the features we need, but it looks like we can get pretty far with the byte code we have already.

The last piece of the puzzle is select() which allows us to wait on a socket while executing another co-routine.

There's a good write-up here: https://stackoverflow.com/questions/49005651/how-does-asyncio-actually-work

qgbcs commented 4 months ago

any progress ?