[Ksummit-discuss] [TECH TOPIC] Building Stable Kernel Trees with Machine Learning

[Ksummit-discuss] [TECH TOPIC] Building Stable Kernel Trees with Machine Learning

[Sasha Levin]

Hi all,

Julia Lawall and myself would like to propose this topic for the
technical track.

An overview:

Building stable trees is difficult; we are required to find only commits
that fix bugs (needle) in the massive flow of commits that go upstream
(haystack).

Currently the process is based on authors and maintainers tagging their
commits properly and helping stable maintainers to know that they should
be picking up these patches.

However, this doesn't always happen right. Commits get lost, forgotten,
or never looked at to begin with. This means that important fixes are
being left out of stable trees and not reaching the users who rely on
stable trees for fixes.

This talk with go over a new approach to detect bug fixing commits in
the kernel tree using machine learning, and demonstrate how it was used
to submit over a thousand commits to various stable trees.

--
Thanks,
Sasha