Implement async input()

[GoogleCodeExporter]

Currently, asyncio doesn't provide any helper to read asynchronously data from 
sys.stdin, like input(). Twisted implements twisted.internet.stdio:

https://twistedmatrix.com/trac/browser/trunk/twisted/internet/stdio.py
https://twistedmatrix.com/trac/browser/trunk/twisted/internet/_posixstdio.py
https://twistedmatrix.com/trac/browser/trunk/twisted/internet/process.py#L106
https://twistedmatrix.com/trac/browser/trunk/twisted/internet/_win32stdio.py

See also https://code.google.com/p/tulip/issues/detail?id=147

Original issue reported on code.google.com by [email protected] on 2 Nov 2014 at 10:55

[GoogleCodeExporter]

Hi,

This seems suspiciously similar to BaseSubprocessTransport and 
SubprocessProtocol. It seems like it could be implemented in a very similar 
matter using sys.stdin and sys.stdout. Thoughts?

Original comment by [email protected] on 17 Nov 2014 at 9:19

[ghost]

You can use loop.add_reader(fd, cb) and add a callback function to stdin file descriptor and change tty settings to cbreak. Or you can make a coroutine to read from curses screen getch method with nodelay option. I'm sure there is a more robust/complicated way to this, maybe including StreamReader.

[vxgmichel]

I implemented something similar as part of this project. It provides the following coroutines:

• get_standard_streams(*, use_stderr=False, loop=None): return two streams corresponding to stdin and stdout (or stderr)
• ainput(prompt=None, *, loop=None): asynchronous equivalent to ainput.

Everything is implemented in stream.py. It should work even if sys.stdin and sys.stdout don't have a file descriptor (inside IDLE for instance).

[socketpair]

https://gist.github.com/nathan-hoad/8966377 - is it good ? @vxgmichel

[socketpair]

I'm vote on implementing stream wrappers for generic file objects.

Something like simple and stupid

• (reader, writer) = asyncio.wrap_fileobject(fileobj) (use fileobj.read(), fileobj.write(), fileobj.flush() internally)
• (reader, writer) = asyncio.wrap_file_descriptor(fd) (use os.read(fd), os.write(fd) internally)
• (reader, writer) = asyncio.wrap_streaming_socket(socket_obj) (use socket_obj.send(), socket_obj.recv() internally)

[vxgmichel]

@socketpair There are a few differences between the example in your link and the way I wrote it:

• I use sys.stdout instead of os.fdopen(0, 'wb') (I'm not sure what is best though)
• I subclass StreamReader and StreamWriter to avoid closing the stream in the __del__ method
• I use an executor if stdin and stdout don't support the file interface (e.g. in IDLE)
• I use a caching mechanism to avoid creating new streams, though it's probably a bad idea :)

About the wrappers you described, I'm not sure it's a good idea to create high level streams from low level objects (file objets, descriptors, sockets). For instance, in order to open a new socket connection, you can either use:

• loop.create_connection: return (transport, protocol) from a socket (or host and port)
• asyncio.open_connection: return streams from host and port

Same thing for subprocesses:

• loop.connect_read_pipe, loop.connect_write_pipe: return (transport, protocol) from a pipe
• loop.subprocess_exec: return (transport, protocol) from a command
• asyncio.create_subprocess_exec: return a high level process object from a command

So I would expect file streams to work the same:

• loop.connect_read_file, loop.connect_write_file: return (transport, protocol) from a file descriptor
• asyncio.open_file: return streams from a file name

[socketpair]

@vxgmichel unfortunatelly, there are cases where file descriptor is pre-exist. For example, systemd's socket activation, or xinetd.

Also, if stdout/stdin is pipe, input and output may block easily, so wrapping them to asyncio stream is convenient.

[nhoad]

FWIW an issue with my gist is that it will break print() calls for sufficiently large output, because stdout is... surprise surprise, non-blocking. Even if you decide to only have a non-blocking stdin you'll hit issues, because stdin and stdout are actually the same object for TTYs, as per this issue: https://github.com/python/asyncio/issues/147.

Also, os.fdopen(1, 'wb') and sys.stdout are interchangeable and there's no reason to use one over the other.