Notices by Steve Canon (steve@discuss.systems)

@airspeedswift did you try splitting the blame up into subexpressions?

@inthehands Gonna have the Porsche subreddit real upset with that last one. 😂

Tried applying the transform matrix explicitly, seems to have gotten worse??? Do I have to unpremultiply instead?

Dog appears to be rendered in the wrong coordinate system. Any suggestions?

Tried doing translation after rotation and now I don’t know where she went…

Sad news for fans of Atlas: he is now chasing voles and offering floating-point tidbits in another realm.

He had been declining for a few months, and hadn't been able to move at all for a couple days, so we had a vet friend come put him down at home today. She came in the evening, which was nice because Riley could be with him for it. Very sad, but I'm really glad that she was able to witness it, since they've been together since forever.

pet death

~hey girl~

@dotstdy @luna iostream formatting is the most clown-shoes misfeature in the whole clown-shoes language.

@irene the unreasonable effectiveness of DARVO will never stop surprising me

@inthehands This question confuses me. The content is crucial, but these messages are likely not a problem, depending on context. The main thing I want in a commit message is a link back to the issue/bug report/forums discussion/whatever that led to a change, if there is one.

If there isn't one, and this is just an editorial change, then "edited bar.txt" is fine (superfluous, because the change itself has that information, but fine). If there's a substantive change and no discussion to link to, I'd like to have rationale in the message. But if the rationale is already somewhere else, all I need is a link to it (and when you paraphrase it in your commit message, you'll probably get some details wrong, so I'd rather have the link).

@inthehands @joe bold to assume that Joe didn’t mean it to begin with

@dalias @jrose because velvel wanted to use it for debugging purposes, roughly

@jrose because then copysign/abs/negate could produce a sNaN result from qNaN input, contra clause 5 norms.

(Or else wouldn’t just be bitwise ops, also undesirable)

Jen’s parents jammed one of our utensil drawers shut with a knife in such a way that I had to drill a hole through the side of the cabinet to perform laparoscopic surgery.

Honestly, children are easier than parents.

@cr1901 @pkhuong @mcc there’s just a special fast-path for them

@cr1901 @joe @pkhuong dependency-breaking xor/movzx are "free" (they have to be decoded, but they don't need to execute).

@joe @cr1901 @pkhuong Intel's dedication to making that particular mistake _over and over again_ will never cease to amaze me.

@joe @cr1901 @pkhuong (They did it for 8->16, 16->32, and SSE->AVX/AVX512. And each time they managed to find a way to make it worse.)