Hacker News Reader

Comment by szpght | Hacker News Reader

The bug was finally fixed with the following commit: https://github.com/openzfs/zfs/pull/15579/commits/679738cc40...

It looks like they managed to exactly pin down what was happening.

Sidenote: What a great commit message. The expanded comment explains why both checks are necessary, but the commit message gives so much more context for anyone wondering and `git blame`ing that line.

Totally. Also: I've never seen Sponsored-by:

> Signed-off-by: Rob Norris <rob.norris@klarasystems.com>

> Sponsored-by: Klara, Inc.

> Sponsored-by: Wasabi Technology, Inc.

This is a longstanding tradition from FreeBSD. A list of our commit message trailers: https://docs.freebsd.org/en/articles/committers-guide/#_incl....

"Sponsored by:" search in FreeBSD commit messages: https://freshbsd.org/?q=%22Sponsored+by%3A%22

I like it! Seems like a great way to incentivise companies to contribute.

> I like it! Seems like a great way to incentivise companies to contribute.

It also helps with lawsuits, which was a major thing right at the beginning of history of BSD.

* https://en.wikipedia.org/wiki/UNIX_System_Laboratories,_Inc.....

Linux later had the lawsuit issue with SCO and IBM and tracking where certain things came from (which was not helped by the fact that Linus Torvalds refused to use source code tracking for the longest time (later taking up Bitkeeper, and later developing git)).

* https://en.wikipedia.org/wiki/SCO–Linux_disputes

The beginning of BSD was the 1978 shipment of 1BSD by Bill Joy.

You likely mean the end of the effort to replace all AT&T code, which was more than a decade later.

A more "correct" fix has been posted https://github.com/openzfs/zfs/pull/15615

Neither of these fixes are great because neither of them include tests. This is a dataloss bug in a filesystem, no existing test covered it, and no new tests were added to demonstrate the efficacy of the supposed fix or prevent regressions.

There's an ongoing work on a stress test for this. https://github.com/openzfs/zfs/pull/15608

There was, actually, a test, but it turns out to be rather hard to make a deterministic test to reproduce a rare race condition without sticking your fingers in at runtime and forcing the "wrong" ordering.

The same was true of the fix for the completely deterministic cp from unencrypted to encrypted with block cloning. There doesn’t seem to me much incentive to write tests.

Reply on news.ycombinator.com