x86_64-unknown-redox-gcc (compiled from source, using relibc)
hello world in C works
gmp
binutils-unwrapped
openssl (not yet tested)
curl (compiles but aborts during runtime)
sed (functional)
Original Post:
In stdenv.mkDerivation, where does the $CC environment variable come from?
I’m trying to fork nixpkgs to add support for cross-compiling to Redox, which has its own toolchain. (My fork’s commit: b3be327)
When I cross-compile to aarch64, $CC is aarch64-unknown-linux-gnu-gcc. However, when I try to cross-compile to redox, $CC is cc instead of x86_64-unknown-redox-gcc. Any ideas how to fix this?
Please loop in us Sign in to GitHub · GitHub on your work! We are always especially excited to try non-unix non-embedded platforms, as they challenge our abstractions to leak like no other.
I would look at how LLVM/Clang toolchains work as it is the gold standard (standard libraries and tools always built separately, multi-target everything). In particular WASI might be a good guide: all we really had to do we swap in the wasi libc (which is sort of a musl cloudabi mashup). If you can just drop in relibc intsead and get something basic like zlib building, that would be an excellent first step.
Conversely, it looks like you are following the android pre-built stuff as a guide right now, which is fine if that’s what’s easiest for you to get something building, but not really a path we’d like to mere things. Indeed, we ought to be building bionic libc from source and avoiding the prebuilt android sdk/ndk too.
I finally figured out how to build relibc (libc for Redox+Linux) from source. However, I couldn’t get it to build without using a specific version of rust, so for now, the derivation isn’t mergable into nixpkgs.
I’ve run into a “Configuration x86_64-unknown-redox not supported” error while building gcc (related github issue for arm). It seems like the solution to that issue was to make changes to upstream gnu-config. I’m going to research how the Redox community is currently building gcc using relibc.
Cheers!
EDIT: I’ve managed to compile semi-functioning gcc using relibc
I’ve gotten a “hello world” to compile and run on Redox via a gcc built by Nix
However, my relibc derivation produces a broken output, so I’m currently using a prebuilt relibc from static.redox-os.org.
A lot of this still feels like black magic. I think I need to spend the next few evenings (or weeks) reading about the cross-compiling and the C build system in general. Are there any recommended resources for learning about these topics? If I find any, I’ll edit them into this post for the sake of future readers.
I’m also having trouble understanding the stdenv bootstrapping process in Nixpkgs [1]. @Ericson2314 has written some excellent comments in pkgs/stdenv (and all across nixpkgs in general). Maybe I should spend the next few evenings reading through them?
Thanks!
[1] For example, nix-shell -A pkgsCross.x86_64-unknown-redox.mkShell works, but nix-build -A pkgsCross.x86_64-unknown-redox.gcc fails. This makes zero sense to me now, but I imagine it would make more sense if I understood how stdenv is bootstapped in Nixpkgs Ohh, I think this has to do with cross-compiler packages versus packages meant to be run in Redox
I put most of my knowledge into the mentioned wiki article. If you have any other discoveries beyond that, you are welcome to put them their as well. Maybe I can add some more stuff about how c++ libraries are linked into the compiler wrapper. I know that @ehmry also works on alternative operating systems.
Thank you! And yes, I plan to document what I’ve learned once I reach a few more milestones.
I know that @ehmry also works on alternative operating systems.
Oh wow, that repo’s a very helpful reference resource. Maybe that’s how Redox support should be added—as a new target in Nixpkgs but with the numerous package fixes in a separate repo.
P.S. Status update for the curious: I think I’m close to building the Rust cross-compiler via Nix. EDIT: I did it! I just cross-compiled a functioning Hexyl (a rust package) using Nix!
Yes, I think that working with an overlay until you’ve reached a beta level of support is the way to go, but I can only speculate. So far its been a good way to track all the fixes that need to be made, and it allows for low-stress experimentation. Feel free to contact me if you have questions on bootstrapping or cross-compiling.