Impressive trolling.
> deepSeq is basically useless because of how almost every datastructure is ciruclar
Circular references will cause "infinite recursion", and these will cause evaluation errors. So valid nix code will not contain circular references. Compositions of derivations create merkel DAGs. There are also fixed points, but deepSeq can still handle those scenarios.
> trace is almost useless because you can never know if a statement is a value or a computation.
use of trace creates a thunk. So it will always be a thunk.
> Derivations actually put something in your store when you evaluate them.
No. Instantiation will create store derivations, realizations will perform a build, and successful builds create store paths. `nix-instantiate '' --eval -A hello.version` Evaluates the expression, but doesn't produce a store derivation.
> Except that I don't need to know this for apt or any other package manager.
You're also probably not creating your own .deb's. You would seek other options.
> Nix is the only one that forces me to learn endless minutia about every single language.
Eventually you will to learn your problem domain. Just because you do python dev on ubuntu, doesn't eventually you will eventually need to know how python finds modules.
> Haha. HAHA. HAHHAHAHAHHA. Oh sorry. That wasn't laughter. That was me trying to hide my tears.
Hmm, I can't see your face. So don't worry about saving it.
> show-trace is about as useless of an addition as I can imagine.
Pretty useful for me, but I also use for work, and free time.
> Nix totally creates a file on your disk in this case. Yes, it doesn't "build" the package as in, it doesn't compile it or fetch it. But it actually does do something on your machine.
Sure?
> The terminology around nix is just disastrously bad.
Because the terms don't align exactly with other existing terms. Haskell has similar issues, where the terminology is foreign for many, but accurate.
> And conda works perfectly without this hack.
Well, conda tries to solve different things. It's definitely not a generic package manager. For supporting python use cases, conda does what it does.
Also, using escalated privileges to install packages is the norm for almost all other package managers. user-level installation is definitely the exception.
> I get it, nix people love nix and love to make excuses for its horrible failures. But seriously, other package managers do this beautifully. Can't we just sometimes accept reality?
Do what? FHS? How many borked distribution upgrades have occurred because of FHS incoherence.
> And that was 20 years ago. The 2.0 CLI is still a mess.
`nix-*` commands makes sense in context of a phd thesis. The `nix ` 2.0 cli, is definitely more ergonomic. Most package manager cli's until about 5 years ago were also pretty bad. Still remember having to use dpkg on ubuntu to fix some issues.
> Oh boy... how I wish this was true!
It is true, that's why I said it.
> Haskell is such a mess in Nix that there are two entire incompatible toolchains.
nixpkgs' haskellPackages is reflective of all hackage packages, and works well for consumption. If you're eluding to haskell.nix, that's a bit more complex and magical.
> Both of which are incredibly hard to use by the way.
This makes sense that it's hard for you.
> And Python packages which are pure are also a disaster!
The python package ecosystem is a hot mess for all distros [0] [1]. The "I can selectively choose small version ranges of dependencies to satisfy my singular use case" doesn't integrate well to distro's trying to present a coherent package set.
[0]: https://drewdevault.com/2021/11/16/Python-stop-screwing-dist...
[1]: https://blogs.gentoo.org/mgorny/2021/11/07/the-future-of-pyt...