@Gankra_ Woah! I thought I was getting away with a lot by turning my dog into the HLSL mascot. Apparently I can go so much further. Challenge accepted.
https://t.co/9z0cSvkSdo
@slava_pestov LLVM is why I 80-col, but my zealotry for manual line wrapping is 100% JavaScript’s fault. Electron apps and Git PR web views make manual wrapping super relevant.
@slava_pestov I have two problems with this:
(1) Please, split your LaTeX source into multiple files.
(2) 80-col or GTFO. Come on man! Why doesn’t your text editor do this for you?!?!
Other than that… wow… just wow… _Love it_
Hot take: `unsigned` is not a type, it is a qualifier. When used without a type specifier, `int` is the type because of C’s default-int rule. Default-int is evil and should be eradicated from the world. `unsigned` should never be used without a type. </rant>
@tobiashieta I’ve heard enough stories of people who don’t host their code on GitHub, but someone else pushed it, and now it is in copilot. LLVM was unofficially on GitHub several times over before the project moved. The only way to stay out of copilot is to be closed source.
IMO, @Github copilot is a huge threat to OSS. I don’t care if my code gets cloned and used without attribution, but a lot of people do. This will drive people to make code closed source.
@disruptnhandlr I always thought of the best PGO data set as being something vaguely representative of your expected inputs, but as few cases as possible, so that you can re-generate the data often.
@disruptnhandlr The truly embarrassing thing is that one silly test, has almost as much impact as the copyright encumbered suite of cases that I used in the past… Performance is silly…
@disruptnhandlr Sadly I’ve never gotten around to finding a copyright unencumbered set of tests that could live in-tree. The in-tree training source is here: https://t.co/ywNk9KoFLl