gh-90467: StreamReaderProtocol - add strong reference to created task

[python273]

gh-90467: StreamReaderProtocol - add strong reference to created task

[gvanrossum]

Oh now I have to research whether it’s possible to have more than one task (like the other PR #30505 assumed) and if so whether it’s necessary to keep all of them alive, or just the most recent one.

[python273]

hmm, StreamReaderProtocolis created once per connection

cpython/Lib/asyncio/streams.py

Lines 77 to 83 in ccf94a6

	def factory():
	reader = StreamReader(limit=limit, loop=loop)
	protocol = StreamReaderProtocol(reader, client_connected_cb,
	loop=loop)
	return protocol
	return await loop.create_server(factory, host, port, **kwds)

and

     State machine of calls: 
  
       start -> CM [-> DR*] [-> ER?] -> CL -> end 
  
     * CM: connection_made() 
     * DR: data_received() 
     * ER: eof_received() 
     * CL: connection_lost() 

cpython/Lib/asyncio/protocols.py

Lines 66 to 90 in ccf94a6

	class Protocol(BaseProtocol):
	"""Interface for stream protocol.
	The user should implement this interface. They can inherit from
	this class but don't need to. The implementations here do
	nothing (they don't raise exceptions).
	When the user wants to requests a transport, they pass a protocol
	factory to a utility function (e.g., EventLoop.create_connection()).
	When the connection is made successfully, connection_made() is
	called with a suitable transport object. Then data_received()
	will be called 0 or more times with data (bytes) received from the
	transport; finally, connection_lost() will be called exactly once
	with either an exception object or None as an argument.
	State machine of calls:
	start -> CM [-> DR*] [-> ER?] -> CL -> end
	* CM: connection_made()
	* DR: data_received()
	* ER: eof_received()
	* CL: connection_lost()
	"""

[gvanrossum]

Yeah, that sounds about right. I'll give the author and reviewers of the other PR some time to find a flaw in this argument. In the meantime I'm going to press the Update branch button.

[gvanrossum]

Also you still need to add a blurb -- see the failing test for instructions.

[bharel]

Don't see a flaw.

I'm not entirely sure why I thought it might be called multiple times for concurrent connections. A protocol should be used for only one connection and the state machine is not per connection but per protocol.

As long as the end user plays according to the book it seems like I made a mistake.

The test simply leaks until the cleanup function removes the loop and the GC activates. You can choose whether to include it or not.

Please add Simon to the acks :-) His original PR was the same until I modified it.

[miss-islington]

Thanks @python273 for the PR, and @gvanrossum for merging it 🌮🎉.. I'm working now to backport this PR to: 3.10, 3.11.
🐍🍒⛏🤖

[bedevere-bot]

GH-96344 is a backport of this pull request to the 3.11 branch.

[bedevere-bot]

GH-96345 is a backport of this pull request to the 3.10 branch.