Add MVP LLVM based mingw-w64 targets

[mati865]

Fixes #72241

Those are rustc side changes to create working x86_64 and AArch64 Rustc hosts and targets.
Apart from this PR changes to various crates are required which I'll do once this is accepted.

I'm expecting more changes on rustc side later on as I cannot even run full testsuite at this moment because passing JSON spec breaks paths in various tests.

Tier 3 policy:

A tier 3 target must have a designated developer or developers (the "target maintainers") on record to be CCed when issues arise regarding the target. (The mechanism to track and CC such developers may evolve over time.)

I pledge to do my best maintaining it, MSYS2 is one of interested consumers so it should have enough testing (after the releases).

Targets must use naming consistent with any existing targets; for instance, a target for the same CPU or OS as an existing Rust target should use the same name for that CPU or OS. Targets should normally use the same names and naming conventions as used elsewhere in the broader ecosystem beyond Rust (such as in other toolchains), unless they have a very good reason to diverge. Changing the name of a target can be highly disruptive, especially once the target reaches a higher tier, so getting the name right is important even for a tier 3 target.

This triple name was discussed at t-compiler/LLVM+mingw-w64 Windows targets

Target names should not introduce undue confusion or ambiguity unless absolutely necessary to maintain ecosystem compatibility. For example, if the name of the target makes people extremely likely to form incorrect beliefs about what it targets, the name should be changed or augmented to disambiguate it.

I think the explanation in platform support doc is enough to make this aspect clear.

Tier 3 targets may have unusual requirements to build or use, but must not create legal issues or impose onerous legal terms for the Rust project or for Rust developers or users.

It's using open source tools only.

The target must not introduce license incompatibilities.

It's even more liberal than already existing *-pc-windows-gnu.

Anything added to the Rust repository must be under the standard Rust license (MIT OR Apache-2.0).

Understood.

The target must not cause the Rust tools or libraries built for any other host (even when supporting cross-compilation to the target) to depend on any new dependency less permissive than the Rust licensing policy. This applies whether the dependency is a Rust crate that would require adding new license exceptions (as specified by the tidy tool in the rust-lang/rust repository), or whether the dependency is a native library or binary. In other words, the introduction of the target must not cause a user installing or running a version of Rust or the Rust tools to be subject to any new license requirements.

There are no new dependencies/features required.

Compiling, linking, and emitting functional binaries, libraries, or other code for the target (whether hosted on the target itself or cross-compiling from another target) must not depend on proprietary (non-FOSS) libraries. Host tools built for the target itself may depend on the ordinary runtime libraries supplied by the platform and commonly used by other applications built for the target, but those libraries must not be required for code generation for the target; cross-compilation to the target must not require such libraries at all. For instance, rustc built for the target may depend on a common proprietary C runtime library or console output library, but must not depend on a proprietary code generation library or code optimization library. Rust's license permits such combinations, but the Rust project has no interest in maintaining such combinations within the scope of Rust itself, even at tier 3.

As previously said it's using open source tools only.

"onerous" here is an intentionally subjective term. At a minimum, "onerous" legal/licensing terms include but are not limited to: non-disclosure requirements, non-compete requirements, contributor license agreements (CLAs) or equivalent, "non-commercial"/"research-only"/etc terms, requirements conditional on the employer or employment of any particular Rust developers, revocable terms, any requirements that create liability for the Rust project or its developers or users, or any requirements that adversely affect the livelihood or prospects of the Rust project or its developers or users.

There are no such terms present/

Neither this policy nor any decisions made regarding targets shall create any binding agreement or estoppel by any party. If any member of an approving Rust team serves as one of the maintainers of a target, or has any legal or employment requirement (explicit or implicit) that might affect their decisions regarding a target, they must recuse themselves from any approval decisions regarding the target's tier status, though they may otherwise participate in discussions.

I'm not the reviewer here.

This requirement does not prevent part or all of this policy from being cited in an explicit contract or work agreement (e.g. to implement or maintain support for a target). This requirement exists to ensure that a developer or team responsible for reviewing and approving a target does not face any legal threats or obligations that would prevent them from freely exercising their judgment in such approval, even if such judgment involves subjective matters or goes beyond the letter of these requirements.

Again I'm not the reviewer here.

Tier 3 targets should attempt to implement as much of the standard libraries as possible and appropriate (core for most targets, alloc for targets that can support dynamic memory allocation, std for targets with an operating system or equivalent layer of system-provided functionality), but may leave some code unimplemented (either unavailable or stubbed out as appropriate), whether because the target makes it impossible to implement or challenging to implement. The authors of pull requests are not obligated to avoid calling any portions of the standard library on the basis of a tier 3 target not implementing those portions.

The target must provide documentation for the Rust community explaining how to build for the target, using cross-compilation if possible. If the target supports running binaries, or running tests (even if they do not pass), the documentation must explain how to run such binaries or tests for the target, using emulation if possible or dedicated hardware if necessary.

Building is described in platform support doc, running tests doesn't work right now (without hacks) because Rust's build system doesn't seem to support testing targets built from .json.
Docs will be updated once this lands in beta allowing master branch to build and run tests without .json files.

Tier 3 targets must not impose burden on the authors of pull requests, or other developers in the community, to maintain the target. In particular, do not post comments (automated or manual) on a PR that derail or suggest a block on the PR based on a tier 3 target. Do not send automated messages or notifications (via any medium, including via @) to a PR author or others involved with a PR regarding a tier 3 target, unless they have opted into such messages.

Understood.

Backlinks such as those generated by the issue/PR tracker when linking to an issue or PR are not considered a violation of this policy, within reason. However, such messages (even on a separate repository) must not generate notifications to anyone involved with a PR who has not requested such notifications.

Understood.

Patches adding or updating tier 3 targets must not break any existing tier 2 or tier 1 target, and must not knowingly break another tier 3 target without approval of either the compiler team or the maintainers of the other tier 3 target.

I believe I didn't break any other target.

In particular, this may come up when working on closely related targets, such as variations of the same architecture with different features. Avoid introducing unconditional uses of features that another variation of the target may not have; use conditional compilation or runtime detection, as appropriate, to let each target run code supported by that target.

I think there are no such problems in this PR.

[rust-highfive]

r? @oli-obk

(rust-highfive has picked a reviewer for you, use r? to override)

[rust-highfive]

⚠️ Warning ⚠️

• These commits modify compiler targets. (See the Target Tier Policy.)

[mati865]

Open for discussion: this is carried over from windows_gnu_base.rs but C/C++ libraries usually prepend lib like they do when targeting Linux.

[bjorn3]

Note that if this is changed, std::env::consts::DLL_PREFIX needs to be changed too.

[joshtriplett]

This implementation looks reasonable to me, and I'm glad to see these targets added.

Could you please add to the PR description a link to an issue where you fill out the target tier policy requirements?

[oli-obk]

r? @joshtriplett

[mati865]

Found a way to simplify spec (tested and it still works).

Could you please add to the PR description a link to an issue where you fill out the target tier policy requirements?

@joshtriplett it's not clear to me what issue you are talking about.
Couldn't find it in https://rustc-dev-guide.rust-lang.org/building/new-target.html or https://doc.rust-lang.org/nightly/rustc/target-tier-policy.html
Could you elaborate? You can ping me on Zulip if you prefer that channel.

[mati865]

Found and fixed the problem, sorry it took so long.
I'll squash commits later. Done

[dot-asm]

Not that my word weighs a lot here and I apologize, but I fail to see the need for an additional target. Is there a programming interface that clang uses/requires that gcc and/or Microsoft cl don't? What I'm trying to say is that it's rather the programming interface that defines any particular target platform, not the compiler. Compilers are supposed to be and actually are interchangeable. For example, on Windows you can use clang-cl instead of vendor cl with *-msvc targets. On Linux you can freely switch between clang and gcc. And I see no reason for why it would be impossible to have the similar arrangement for *-windows-gnu targets.

[driver1998]

Is there a programming interface that clang uses/requires that gcc and/or Microsoft cl don't?

MinGW+LLVM (and libc++, compiler-rt, LLVM binutils) is the only usable MinGW toolchain for Windows on ARM right now. GCC doesn't even start their Windows ARM support.

Also I guess the reason is, both llvm-mingw and MSYS2 CLANG* environments use the Universal CRT, instead of the traditional msvcrt, hence the incompatibility.

MinGW+LLVM+MSVCRT is possible, and I think it should be usable in *-windows-gnu targets, but that would also break compatibility with MSYS2.

I guess we should name it as *-windows-gnuucrt? How does rust handle musl vs glibc on Linux?

[mati865]

@dot-asm this not only the "how the linker is called" issue.
Whole toolchain is changed here by which I mean: ABI, development and runtime libraries. As opposed to using only Clang + GNU/MSVC libraries.

If you make Rust call clang as the linker with current windows-gnu targets the first thing you will face are errors about missing libraries with LLVM toolchains.
This could be fixed by setting no_default_libraries: false in windows-gnu spec files and letting linker do what it wants instead of specifying libs manually (which is unlikely to happen).

There are further issues though. GNU and LLVM toolchains use different runtime libraries: libgcc_*.dll, libstdc++.dll and libwinpthread-1.dll vs libunwind.dll and libc++.dll. There is no way to easily and reliably distinguish which set to use and mixing them is asking for problems.
32-bit libunwind is especially problematic here, simply linking https://github.com/rust-lang/rust/tree/master/library/rtstartup (which are specifically made to work with 32-bit libgcc) totally breaks exception handling by libunwind.

Another issue directly affecting Rust is ld.bfd inability to correctly use short import libraries created by LLVM (the same format as MSVC uses for years). This is the reason why Rust cannot use LLD by default with windows-gnu and in turn provide sanitizers, code coverage or PGO for this target.

There are more incompatibilities that likely won't affect pure Rust crates that I didn't mention here but with C/C++ code the cause issues.

[mati865]

Had to bring back -nolibc workaround because newer Clang versions skip compiler-rt when passed -nodefaultlibs.

[dot-asm]

First of all, C++ matters only when you want to compile C++ code and link it with Rust as FFI. Since this is rust-lang, we should concentrate on Rust itself. [Though since C++ is mentioned, I'll make a remark later on.]

@dot-asm this not only the "how the linker is called" issue. Whole toolchain is changed here by which I mean: ABI, development and runtime libraries. As opposed to using only Clang + GNU/MSVC libraries.

If it is so "bad" as you describe, then how would you explain following? I can

1. create empty libunwind.a,
2. build a Rust project with MinGW gcc with RUSTFLAGS set to -L native=$PWD -l static=unwind,
3. remove the final binary,
4. create empty libgcc.a and libgcc_eh.a and copy clang's libunwind.a,
5. re-build the project with MinGW clang, which is limited to just re-link,
6. and finally execute the resulting binary.

And I can even go the other way around. This is with unmodified stable cargo/rust. The switch between MinGW gcc and clang is performed by readjusting the PATH. In other words, as far as Rust itself is concerned, MinGW gcc and clang appear interchangeable as linkers. And the difference in libraries to link with appears very limited and manageable, you either link with libgcc.a/libgcc_eh.a or libunwind.a. Hence my question is what is more sustainable, to query availability of the libraries and make the choice at link time, or to maintain separate platforms? Again, from Rust's perspective alone!

Now, I'm not claiming that the above test accounts for everything. But at least it's a non-trivial multi-threaded[!] test and benchmark, and I can verify that clang is in fact invoking ld.lld. You mention 32-bit exception handling. Yeah, I can imagine it being incompatible. But it's 32-bit x86-specific thing, right? Is it possible that this is the only thing that gets actually broken? Given that 32-bit x86 is a disappearing breed, maybe it would be more appropriate to single out specifically that target?

As for C++. It's cc-rs's business anyway and it's perfectly possible to resolve it dynamically by querying whether it's gcc or clang, and make corresponding choices there. In cc-rs that is.

As already mentioned in first remark, it's not like you have to listen to me. I would find it saddening if it's resolved as disjoint targets, but it might be just me:-)

[mati865]

And I can even go the other way around. This is with unmodified stable cargo/rust.

Still you'd need to somehow detect which dynamic library (libgcc or libunwind) has been linked to vendor it:

rust/src/bootstrap/dist.rs

Line 132 in 0677edc

fn make_win_dist(

This is because name of import library doesn't affect name of dynamic library that will be linked.

Example of such vendored library as the dependency:

$ ntldd ~/.rustup/toolchains/nightly-x86_64-pc-windows-gnu/bin/std-afb6c3c717907344.dll
        ADVAPI32.dll => C:\WINDOWS\SYSTEM32\ADVAPI32.dll (0x00000247c75d0000)
        bcrypt.dll => C:\WINDOWS\SYSTEM32\bcrypt.dll (0x00000247c7a60000)
        libgcc_s_seh-1.dll => C:\Users\mateusz\.rustup\toolchains\nightly-x86_64-pc-windows-gnu\bin\libgcc_s_seh-1.dll (0x00000247c26f0000)
        KERNEL32.dll => C:\WINDOWS\SYSTEM32\KERNEL32.dll (0x00000247c7680000)
        msvcrt.dll => C:\WINDOWS\SYSTEM32\msvcrt.dll (0x00000247c7680000)
        USERENV.dll => C:\WINDOWS\SYSTEM32\USERENV.dll (0x00000247c2720000)
        WS2_32.dll => C:\WINDOWS\SYSTEM32\WS2_32.dll (0x00000247c7860000)

So let's move on to ABI incompatibilities. Unlike Linux where LLVM wants to be drop in replacement and closely follows GCC, with mingw-w64 different path has been chosen. I'd say its ABI is somewhere between MSVC and GCC, taking GCCas the base but instead of copying its bugs it borrows from MSVC.
So it has better interoperability with MSVC DLLs than GCC but is not 1:1 compatible with GCC. The most common issue here is using TLS across DLLs built by different compilers. LLVM uses native TLS (just like MSVC) and GCC uses emuTLS (which is can of worms on Windows), naturally they are not compatible.
There were more such cases that I cannot recall right now.

There are more reasons to create new spec:

• code coverage
• sanitizers
• PGO
• enabling function_sections in spec file
• enabling has_thread_local in spec file

All of them used to crash with GCC + ld.bfd, right now I think first three cause build failure and other two cause resulting binaries to crash.
They work fine only when using LLVM toolchain (although I need to do more tests for has_thread_local).

I still stand by my comment at #72241 (comment)

Adding new target would do the trick but it might be an overkill if things can work with existing target.

Except two years forward I don't think it's possible.

As already mentioned in first remark, it's not like you have to listen to me. I would find it saddening if it's resolved as disjoint targets, but it might be just me:-)

It's up to @joshtriplett to decide.

[mati865]

Thank you for the explanation, updated the description.

I have one more question about ucrt.
This target uses it as a C library, but it's not visible from the target's name.

Any plans to introduce windows-gnullvm + msvcrt and/or windows-gnu + ucrt targets later?
What naming will they use if added?

dot-asm created proof of concept repo showing how to not depend on any *CRT, maybe that could be used to make stdlib safe with any *CRT.
As for mingw-w64 libmsvcrt.a is just a copy of CRT that was configured as the default, so it's either copy of libmsvcrt-os.a or libucrt.a.

I see the mingw64 (classic), ucrt64, and clang64 "distributions" in MSYS2.
If ucrt64 is windows-gnu + ucrt, and clang64 is windows-gnullvm + ucrt, then I guess that nobody cares about windows-gnullvm + msvcrt?

I don't think I've had any requests for it.

And that we may add a windows-gnu + ucrt target in the future?

Possibly but ideally windows-gnu would support either CRT.

[ChrisDenton]

msvcrt.dll is essentially a holdover from the 90's and is not intended to be used by third party applications. At this point it's only kept around for backwards compatibility. There is not really any good reason to still use it now that the UCRT exists and is well maintained.

[mati865]

msvcrt.dll is essentially a holdover from the 90's and is not intended to be used by third party applications. At this point it's only kept around for backwards compatibility. There is not really any good reason to still use it now that the UCRT exists and is well maintained.

If it was up to me I'd deprecate mingw-w64 toolchains with MSVCRT long time ago.
Too many users prefer to use it though.

[dot-asm]

dot-asm created proof of concept repo showing how to not depend on any *CRT, maybe that could be used to make stdlib safe with any *CRT.

Just in case for reference. I've summarized all the relevant points discussed here in the README at https://github.com/dot-asm/min-crt-poc.

[mati865]

Sorry it took so long.

@rustbot ready

[petrochenkov]

@bors r+

[bors]

📌 Commit 60361f2 has been approved by petrochenkov

[bors]

⌛ Testing commit 60361f2 with merge 17180f4...

[bors]

☀️ Test successful - checks-actions
Approved by: petrochenkov
Pushing 17180f4 to master...

[rust-timer]

Finished benchmarking commit (17180f4): comparison url.

Summary: This benchmark run did not return any relevant results.

If you disagree with this performance assessment, please file an issue in rust-lang/rustc-perf.

@rustbot label: -perf-regression