I have rarely seen a more striking example of Betteridge's law.
Remote
Colin Watson
@cjwatson@mastodon.ie
mastodon 4.7.3Irish geek. Happily married, father of four, polyamorous, left-wing, Reform Jew by choice.
Debian and Ubuntu developer and freelancer, mainly working for Freexian. Formerly at Canonical on launchpad.net.
Tá mé ag foghlaim na Gaeilge. אני לומד עברית.
Previously @cjwatson@mastodon.social.
558 Followers
182 Following
31 Posts
Joined November 12, 2022
Pronouns:
he/him or they/them
Web (personal):
Web (professional):
Twitter (abandoned):
Replying to
Replying to
@katiefenn I found the biggest asshole I ever worked with on Mastodon the other day and preemptively blocked him (despite never having interacted here) just to reduce the chance of ever interacting in the future. That was quite cathartic.
Replying to
@gunstick @Ananas @InternetEh Probably because there _is_ such a thing as a scheduled test of the alarm machinery itself rather than of how people respond to it. It's a long time since I worked in an office, but in one such office there absolutely was a weekly alarm test in which people weren't expected to leave the building.
(Doesn't make the slop generator's behaviour any better of course, but it might explain how it got there.)
Free software activity in April 2026
groff, OpenSSH, and parted updates; helped with dput-ng data loss fix.
Blog post: https://www.chiark.greenend.org.uk/~cjwatson/blog/activity-2026-04.html
Replying to
@aspragg@ohai.social @jonny@neuromatch.social It was pretty much my first reaction too when I saw people being all bootlicky about LLMs on LWN. https://lwn.net/Articles/1075409/
Replying to
Replying to
@the_wub @whitequark @tael I have found that they can at least sometimes be persuaded if the argument is along the lines of "we intend to get the other one changed but there's an elapsed-time condition that hasn't been met yet".
Replying to
@corbet @WesternInfidels I definitely have (not in the kernel, but still). It's such an uncanny-valley feeling when you thought you were talking to a human but then realize you weren't really.
Replying to
While you can get big improvements here with relatively small changes, it's also easy to introduce problems like this, so ideally I'd have regression tests. Django has QuerySet.explain, so if I set things up right I could assert things about its output, like "no seq scans" or "no nested loops". I imagine I'd use the JSON output format to help with parsing. (2/3)
Replying to
But the results obviously won't be stable across PostgreSQL versions. I don't mind needing to make the odd tweak, but I'd like to understand the scale of changes that we'd be signing up for over time. Are there any good-quality Django applications (it's fine if they're PostgreSQL-specific) that have adopted this testing strategy? If so, I'd like to look at their git history. (3/3)
Replying to
@purple @jukkan @davidgerard desperately trying to vaguely resemble a LinkedIn junkie's idea of cool
This is, like, 4000x more petrolhead than my usual scene. But it's quite a fun weekend out #DoningtonPark
Replying to
@jvoisin@infosec.exchange Yep, it is indeed completely standard for Mark to nix strong candidates at the last stage after everyone else has left positive feedback on them. And roles stay open for years while everyone else tries to cover for them. This process is "scientific" and "data-driven" and no internal argument about it is brooked. Many other things about Canonical are great, but the hiring process is grim.
This is not why I left after nearly 20 years, but it's also not not why I left.
Replying to
@K4mpfie @brianbilston Surnames are often irregular so can be hard even for native speakers, but in this case, yes, "Yates" is the correct pronunciation.
Replying to
@stancarey gah, just read the article. I was only at David Abulafia's funeral recently! Absolute ghouls.
Replying to
Replying to
@trabex@newsie.social @fr0gger@infosec.exchange ultimately, to construct a back door in the SSH service that would let the attackers execute unauthorized code on systems with the compromised library installed
Replying to
@barubary@infosec.exchange @david_chisnall@infosec.exchange @linear@nya.social @davidgerard@circumstances.run It was definitely him, yes. He also didn't understand when people pointed and laughed, and IIRC somebody with more clue did in fact track down the bug in question to a clear error in the code being compiled.
Replying to
@nicemicro@fosstodon.org @Vordus@chitter.xyz The straight answer is that it's the number of letters between the "a" and the "y", not counting the endpoints. But I agree it's ridiculous.
Replying to
@kissane So glad to see my admins have preemptively suspended a huge pile of FB/Meta servers without me even having to ask
Replying to
@diegor I agree things like choice of seq scan are situational and data-dependent, of course; none of this is going to be universal, just like many other things one might want to test. Nevertheless, if I've gone through an optimization exercise involving looking at choices of query plans and finding ways to tweak them, that seems like something I want to record somewhere so that it doesn't regress accidentally.
Replying to
@sil My response was "this feels like more admin than the tax relief is worth" so I just decided to stop claiming, but if you find something I might be interested
Replying to
@adamshostack@infosec.exchange Conceivably https://bugzilla.mozilla.org/show_bug.cgi?id=1473933 / https://github.com/mozilla-firefox/firefox/commit/6fa8cead8c202eea4eb0a490fa97edd29616919f Would also need to check that there wasn't some other version of the delay after that change though - that's a bit too much git archaeology to do from my phone.
Replying to
@spycrab@kolektiva.social @simon@tutut.delire.party @hywan@floss.social Only Debian developers (formal project members) will be able to vote.
I wouldn't expect something relatively superficial like a hashtag campaign to make much difference at this point, as I suspect most eligible voters have fairly strong opinions of their own already. Though maybe there are more swing voters than I think ...
Replying to
@diegor I definitely don't want to use execution time testing. Our tests run in various contexts with wildly varying hardware capabilities, and this would mean a lot of noise - likely much worse for us than keeping up with changes in PostgreSQL.
Replying to
@liw@toot.liw.fi Lots of good advice in this thread. Other miscellaneous things I've used successfully:
- Fix something simple and debug it when it doesn't work. Repeat until things start to make sense.
- Often you need to know how something is used. Don't underestimate low-tech approaches. Set a breakpoint or print something loud and run tests until you hit it, or just deliberately break it and see what else breaks as a result.
- Use VCS history.
tig -Sthingis a superpower.