Bug 14697

Summary: Hang due to missing asyncio loop argument: Initialising tasks at 44%
Product: [Build System, Metadata & Runtime] BitBake Reporter: Jate Sujjavanich <jatedev>
Component: bitbakeAssignee: Jate Sujjavanich <jatedev>
Status: RESOLVED FIXED QA Contact:
Severity: normal    
Priority: Medium+ CC: devjate, jatedev, poky.bs.watcher, poky.watcher, randy.macleod, steve
Version: 3.1.12   
Target Milestone: 3.1.14   
Hardware: x86   
OS: Multiple   
Whiteboard:
OS type for building Yocto: --- Type of Regression: ---
Verified: Documentation change: Don't know

Description Jate Sujjavanich 2022-01-18 15:48:10 UTC 
Commit 74a1e71b1e677a482fdedc685a71a1798ad63920 removed the loop argument to calls to start_server and start_unix_server from asyncio. This is to fix a bug with that library in python 3.10.

However, this introduces a regression with python 3.5 and 3.6 which still require that argument. This results in a hang at "Initialising tasks" at 44%. The code cannot automatically detect the loop because that functionality was not implemented until 3.7.

The Reference Manual specifies that dunfell and hardknott require python versions greater than or equal to 3.5 and 3.6, respectively.


Preconditions/Environment
-------------------------
Bitbake 1.46 under poky dunfell on Ubuntu 16.04

Triggering Action/Cause
-----------------------
Run bitbake for anything that requires tasks.

Expectation
-----------
Bitbake executes tasks and exits

Actual Result
-------------
Bitbake hangs indefinitely at 44%

Reproducibility
---------------
Reproduced on 3.1.13, 3.1.12
A git bisect showed that commit 74a1e71b1e677a482fdedc685a71a1798ad63920 introduced the regression.
Comment 1 Randy MacLeod 2022-01-20 15:57:24 UTC 
We suspect but have yet to confirm that this only affects dunfell releases.
We do need to support python-3.5 up to python-3.10 (F35 and more?) for dunfell.

The buildtools has python-3.6 so that may be a work-around but dunfell did originally support python-3.5 so this is a bug.
Comment 2 Jate Sujjavanich 2022-01-21 14:42:46 UTC 
I have tested dunfell's bitbake with python 3.6, and the bug does not occur there. Given the minimum python requirement in hardknott, I think the bug will not occur there.

It appears that the python maintainers added implied loop detection behavior to 3.6 and then added the argument deprecation in 3.7.

I am updating my patch to only add loop to the call on 3.5. Also, I'm adding a reference to this bug.
Comment 3 Randy MacLeod 2022-01-21 17:01:56 UTC 
Jate, thanks for the explanatory comment.

FYI, I reset the target milestone to 3.1.14. 

Some browsers seem to force it to 0.0.0. I used to have the problem but it hasn't happened for so long that I don't remember what fixed it!