Re https://www.reddit.com/r/Racket/comments/gm5vmv/a_review_of_the_racket_programming_language/:

Would it make sense to prioritize single collection package over multi collection package? That is, when multi collection packages collide with single ones, prefer to keep single ones and make multi collection packages failed instead of failing both.

Also, https://pkg-build.racket-lang.org/(indirect%20conflicts.txt) seems to be a 404

That seems a bit arbitrary to me. For example, the package that Sage used to “attack” mine was probably a single-collection package. It might be better to keep some persistent state on the package server between builds to avoid moving docs around between builds like this, although I don’t personally mind the current model either. In effect, his attack made my docs move so they are now less discoverable, which is not great, but they are still there and you can get to them by clicking the link in the package listing. Someone who’s installed the package locally wouldn’t have this problem because they won’t have installed the attacking package.

I do think it’s unfortunate that packages that have conflicts are marked with https://github.com/zyrolasting/koyo/issues/1#issue-622058782\|red, but that’s just a UI/UX thing that could be improved.

It’s also worth pointing out that neither PyPI nor NPM (as far as I’m aware) build your packages and host their docs.

Oh yeah, I just recall Rosette used to have a similar(?) problem too: https://github.com/emina/rosette/blob/master/rosette/info.rkt#L10

Yup, I’ve definitely run into that as well.

I think we should fix the fact that the docs moved, and go to a first pkg wins approach

Also we should just namespace all documentation so that conflict doesn’t prevent both docs from appearing, but that’s harder