LWN: Comments on "Dueling memory-management performance regressions"

Dueling memory-management performance regressions

rweikusat2 — Wed, 19 Jun 2019 17:21:46 +0000

Sounds like squid :->.

Dueling memory-management performance regressions

epa — Mon, 17 Jun 2019 18:32:30 +0000

This kind of double standard is pretty common. You may file a bug report, with patch, fixing a recently introduced bug. Then you have to fight through a thicket of queries and objections -- scripts might be relying on the existing behaviour, it might be a bug fix but I think we should redesign this whole module from scratch before changing anything, we can't change anything until the 3.0 release. When quite obviously none of these tests applied at the point the bug was introduced.

Dueling memory-management performance regressions

nivedita76 — Mon, 17 Jun 2019 11:21:01 +0000

This situation does seem a bit remarkable reading the threads. It’s been over half a year now with David refusing (I don’t think that’s an uncharitable summary) to provide a reproducer while insisting that it’s trivial to create one.

This also shows the limitations of a strict no-regression policy: the original change obviously caused severe regressions, but because it sneaked into the kernel without anyone noticing at the time, Linus now thinks we should just be stuck with it unless we can fix David’s non-public use case.