Core rewrite

[Lonami]

Summary

This branch is an attempt to rewrite the heart of the library under asyncio with the lessons learnt. The following is a rough UML diagram that depicts the current state:

The basic idea is that the MTProtoSender should not be concerned about Session files and just be able to handle arbitrary connections to arbitrary datacenters, as well as handling automatic reconnection in case of failure (but not reconnection for the first connect() call- that should be users' responsability).

Rationale

By sending all requests and receiving all responses from a separate loop, it makes it much easier to reason about the flow of the program, which is really important:

1. A request will be enqueued for sending, and its Future returned.
2. The request will be sent eventually, going from the sender, to the connection, to the TCP socket.
3. The future will be set with a response eventually, coming from the TCP socket, passing through the connection.

Using the standard way to handle Future gives much more power than handrolling a custom solution, and forces a send in order to get a handle you can await for receival. The futures can be cancelled, and if they have not been sent yet, they will be popped from the queue, which means timeouts can occur at this level.

The MTProtoSender is perfectly usable to test the performance at its lowest level to possibly spot new bottlenecks.

Bugs that this probably fixes

• Close pending loops properly in asyncio #825 - Close pending loops properly in asyncio, since disconnect() will await for the started send and receive pseudo-loops to finish.
• Multiple workers have problem in download #809 - Multiple workers have problem in download, there will be no "workers". They will just be ensured futures.
• ConnectionResetError #373 - ConnectionResetError , for the same reasons listed below.
• Getting a timeout error when trying to answer to Inline Bot #326 - Getting a timeout error when trying to answer to Inline Bot, these are normal but now handled correctly.
• Exception: Number of retries reached 0. #270 - Exception: Number of retries reached 0, due to the simplification of the code flow and removal of some old poorly structured code.
• Reconnect on lost network interface #178 - Reconnect on lost network interface, due to the way sockets are handled in the asyncio version.

These issues will be closed upon merging, and should be reopened if they still occur.

Pending things

• Probably double check everything to make sure that, at least, the basic functionality is implemented, such as enough async/await and exporting clients to different datacenters to download media.
• Make sure the new telethon.client package is properly documented in readthedocs.
• Turn master into the asyncio version, and create a backward-compatible "threaded" variant where you don't need to worry about await'ing anything. This is very important.
• Rename asyncio into asyncio-stale or similar since this will be its new variant, but the history should be preserved.
• Test the automatic sending of MessageContainer when a future may be cancelled. May be a good time to start writing proper tests.