Would a gnullvm environment be acceptable?

[PieroV]

Hi, I'm using the *-pc-windows-gnullvm target in Rust, and after #78655, I'm getting a few errors (e.g., one was solved in rust-lang/cc-rs#1176).
I am wondering: rather than adding maps from gnullvm to gnu on all Clang users, would it be possible to add a gnullvm environment in llvm/include/llvm/TargetParser/Triple.h/llvm/lib/TargetParser/Triple.cpp that behaves like a normal gnu environment?

[llvmbot]

@llvm/issue-subscribers-clang-driver

Author: Pier Angelo Vendrame (PieroV)

Hi, I'm using the `*-pc-windows-gnullvm` target in Rust, and after https://github.com//pull/78655, I'm getting a few errors (e.g., one was solved in https://github.com/rust-lang/cc-rs/pull/1176). I am wondering: rather than adding maps from gnullvm to gnu on all Clang users, would it be possible to add a `gnullvm` environment in `llvm/include/llvm/TargetParser/Triple.h`/`llvm/lib/TargetParser/Triple.cpp` that behaves like a normal `gnu` environment?

[PieroV]

CC'ing @mati865 @thomcc as maintainers of the *-pc-windows-gnullvm targets on Rust

[mati865]

I'm getting a few errors (e.g., one was solved in rust-lang/cc-rs#1176).

There were some teething issues, that's true but kinda expected with any new target. If something is still missing please ping me there.

---

cc @mstorsjo as LLVM's mingw-w64 maintainer

I'm not sure if that would bring any benefit to the LLVM project. Clang is quite flexible in its runtime configuration: you can switch between different TLS, std C++ lib, unwinder, etc. And perhaps most importantly, doesn't provide official binaries that have to work with both GNU and LLVM toolchains.

Rust, on the other hand, depends on static configuration of these things and gives no easy way to change them (well, you can use custom target with JSON spec, but that's not convenient). And also ships prebuilt tools that are the usual way the users use it (instead of distro package manager). To achieve that, it needs two distinct targets.

Or what out are proposing is just to accept gnullvm as gnu to make life easier for Rust?

[PieroV]

Or what out are proposing is just to accept gnullvm as gnu to make life easier for Rust?

Yes, for this. Not only for the compiler, but also for the Rust ecosystem in general.
For example, I had a similar failure withbindgen.

I would be happy to try to write a patch for a gnullvm environment, but I'd like to understand if it's a good idea from people who understand these topics better than me first 😄.

Edit: bindgen's failure details:

thread 'main' panicked at .../pkcs11-bindings/build.rs:51:10:                                
Unable to generate bindings: ClangDiagnostic("error: version 'llvm' in target triple 'x86_64-pc-windows-gnullvm' is invalid\n")

[mati865]

I don't know how to reproduce the issue with bindgen.
https://github.com/rust-lang/rust-bindgen builds fine, and https://github.com/rust-lang/rust-bindgen tests do pass.

Tested with:

$ rustc -vV
rustc 1.86.0 (05f9846f8 2025-03-31) (Rev2, Built by MSYS2 project)
binary: rustc
commit-hash: 05f9846f893b09a1be1fc8560e33fc3c815cfecb
commit-date: 2025-03-31
host: x86_64-pc-windows-gnullvm
release: 1.86.0
LLVM version: 20.1.1

If something doesn't use https://github.com/rust-lang/rust/blob/420ca7104f4db107837cba4d8e48a82c53fcb76c/compiler/rustc_target/src/spec/targets/x86_64_pc_windows_gnullvm.rs#L13, then it's probably an issue with that tool.

[PieroV]

I see, thanks.
The failing crate doesn't explicitly set a target (here is its build.rs), however it's being built inside the Firefox build system.
So, I will have to investigate a little bit more why it fails, to see how it's being called exactly.
Probably, my patch for using the gnullvm target isn't enough.

[mati865]

Sorry, I didn't see I pasted the same link twice.
I meant I built http://github.com/mozilla/pkcs11-bindings using gnullvm as the host toolchain.

however it's being built inside the Firefox build system

Likely something inside that build system contributes (or even causes?) this issue.