Asyncio StreamReader fails to close Transport

[jdlh]

BPO	34176
Nosy	@asvetlov, @1st1, @JDLH

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2018-07-21.00:30:19.386>
labels = ['type-bug', '3.7', 'expert-asyncio']
title = 'Asyncio StreamReader fails to close Transport'
updated_at = <Date 2018-07-21.00:30:19.386>
user = 'https://github.com/JDLH'

bugs.python.org fields:

activity = <Date 2018-07-21.00:30:19.386>
actor = 'JDLH'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['asyncio']
creation = <Date 2018-07-21.00:30:19.386>
creator = 'JDLH'
dependencies = []
files = []
hgrepos = []
issue_num = 34176
keywords = []
message_count = 1.0
messages = ['322047']
nosy_count = 3.0
nosy_names = ['asvetlov', 'yselivanov', 'JDLH']
pr_nums = []
priority = 'normal'
resolution = None
stage = None
status = 'open'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue34176'
versions = ['Python 3.7']

[jdlh]

Asyncio's StreamReaderProtocol[1] often returns True from Protocol.eof_received(). This tells the Transport that "closing the transport is up to the protocol" [2]. However, StreamReaderProtocol does not call Transport.close(). More precisely, StreamReaderProtocol leaves the transport interaction to StreamReader. It calls StreamReader.feed_eof()[3]. But StreamReader does not call self._transport.close(). It only sets a self._eof flag, and awakens any of its functions which were waiting for data.

The only occurrence of self._transport.close() occurs in StreamWriter[4]. There is none in StreamReader or StreamReaderProtocol.

I believe that the fix should be for StreamReader to call self._transport.close(). However, it may need to avoid this if StreamReaderProtocol._over_ssl is True, and I don't immediately see how it can tell this. Also, it may need to wait for anything waiting for self._waiter to run, before closing the Transport.
Or, maybe StreamReaderProtocol should not return True from Protocol.eof_received().

I discovered this when using a Transport of my own devising, which reads from a file instead of from a socket, with asyncio.streams. The file did not close in the unit test. My transport duly called Protocol.eof_received(), but received a True in response, so did not close the file.

My workaround will probably be to not believe what Protocol.eof_received() tells me, and to close my transport regardless.

[1] Lib/asyncio.streams.StreamReaderProtocol.eof_received, line 259-266.
[2] 19.5.4.2.3. "Streaming protocols" <https://docs.python.org/3/library/asyncio-protocol.html\>
[3] Lib/Lib/asyncio.streams.StreamReader.feed_eof, lines 419-421.
[4] Lib/Lib/asyncio.streams.StreamWriter.close, lines 316-317.

All line numbers are as of tag v3.7.0, commit 1bf9cc5 .

Bug observed in v3.7.0. Not yet investigated in HEAD revision.

[kumaraditya303]

Python 3.7 is not supported anymore, moreover the issue does not has a reproducer to reproduce the issue and is out of date so closing. If this issue still exists, create a new issue with reproducer.

[kumaraditya303]

Also 3.11 has rewritten ssl implementation as of #31275.