Jiri Kosina: stable workflow

May 2, 2014

Participants: Ben Hutchings, Greg Kroah-Hartman, Guenter Roeck, Jan Kara, Josh Boyer, Mark Brown, and Steven Rostedt.

People tagged: Stable team, major distro kernel maintainers.

Jiri Kosina, responsible maintainer of kernels for SUSE enterprise products, suggests “for-stable” branches as a replacement for the current cherry-picking. [ Editor's note: This topic sparked substantial discussion last year. ]

Josh Boyer (Fedora) concurs with Jiri.

Steven Rostedt pointed out that the current -stable rules require that a patch be upstream before being backported, and that “for-stable” branches might tempt people to sneak in changes that are not yet in Linus's tree. Jiri Kosina argued that the same verification process should work for either cherry-picking or branch pulling, so that branch pulling should not be more vulnerable to sneaky people than cherry-picking is. Josh Boyer agreed with Jiri, noting that Dave Miller already does branch pulling for stable fixes in linux-net. Steven believes that the difference has to do with trust, and that it might be more difficult to verify pull requests. However, Steven did agree that it should be possible to automate checking that pull requests for -stable contained only commits already in mainline. On the other hand, Steven believes that the task of determining whether or not a given commit is a good candidate for -stable will resist automation. Jiri agreed that attempted sneaking of code into -stable does happen, whether on purpose or not. However, Jiri believes that such attempted sneaking is by random contributors rather than subsystem maintainers. He also notes that the standardized for-stable marking is already in place, which should make automation straightforward. Jan Kara pointed out that either the patch author or the maintainer can add “CC stable” in order to eventually cause the patch to land in the -stable queue, which led him to wonder what practical difference a pull request would represent.

Ben Hutchings (who is a stable-tree maintainer, but who will not be able to attend Kernel Summit this year) argued that pull requests must be opt-in, not mandatory. Mark Brown stated that in his ARM stable-kernel role, he would like to participate in any discussion, but in his upstream-developer role, he was not interested in doing extra work for each of the plethora of -stable trees. Guenter Roeck agreed with Mark.

Greg Kroah-Hartman said that he is happy to take -stable pull requests from maintainers that he trusts, and agreed that Dave Miller is a good example. That said, all that Greg asks for is that the maintainer CC -stable: Because maintainers are the Linux kernel's most limited resource right now, any extra work on the part of the maintainer, while welcome, is strictly voluntary. (Reiterated here.)