Bump rustc version

[osa1]

This PR bumps rustc version which enables cargo's build-std feature
(#2251) and also allows us to remove some feature flags that became
stable in the meantime (ptr_offset_from, maybe others too).

(build-std feature exists in the current version too, but it's too
buggy or too different than the currently documented version, so I
couldn't use it)

[osa1]

Confirmed that upgrading to LLVM 11 fixes the issue.

[osa1]

In general we need to make sure rustc and wasm-ld (and other LLVM tools if we use them) will be the same version, otherwise as we've just seen incompatibility can cause compile errors only in some cases. I think in theory it could even cause runtime bugs.

I'm trying to find out if there's a reliable way to find out what version of LLVM a given rustc binary is using.

[osa1]

I can run nix-build -A tests --max-jobs 12 --pure locally just fine but CI fails to build. @nomeata any ideas?

[nomeata]

If you follow the “details” link next to the evaluation job, you see the Hydra view with all the individual jobs. There, the tests have not run yet, and the failing jobs (at the moment) are related to darwin stuff.

So likely, the tests will also pass on CI, once they will be built.

[osa1]

So likely, the tests will also pass on CI, once they will be built.

Right, my question was whether the build failures look familiar.

[osa1]

Should I try to bump nixpkgs maybe?

[nomeata]

CMake Error at CMakeLists.txt:4 (message):
  libc++abi now requires being built in a monorepo layout with libcxx
  available
  
builder for '/nix/store/25020dfwps2ddzkh9g8464bxhi3n2wvd-libc++abi-11.0.0rc2.drv' failed with exit code 1

Maybe worth a try. Might also be worth seeing whether clang11 builds in plain nixpkgs master on darwin.

[dfinity-ci]

In terms of gas, 3 tests improved and the mean change is -1.1%.
In terms of size, 3 tests regressed and the mean change is +0.1%.

[osa1]

In terms of gas, 3 tests improved and the mean change is -1.1%.

Thanks, LLVM

[nomeata]

You changed the commit to a rev on master, without chanigng the branch field. Git doesn't tell you that, but it’ll be confusing for the next person to look at this. So before merging, make sure that the branch field correctly says which branch the revision is from.

[osa1]

Will do.

We may not want to change the branch, I mainly wanted to see if using the latest version will fix build issues on Darwin.

[nomeata]

Right, that’s what I gathered. Just leaving this comment so that this doesn't slip through in the end.

[osa1]

rustc PR that fixes showing LLVM version used in rustc -vV: rust-lang/rust#80981

[nomeata]

The rts job builds on linux! So that’s promising, isn’t it?

[osa1]

It was building before nixpkgs bump too. It even passed the tests on my system.

[osa1]

Unfortunately shell doesn't work for me anymore. niv doesn't build with recent GHCs, even the tip of the master branch. It's not clear what the supported version of GHC is, README doesn't mention it.

[nomeata]

Yes, that’s the problem with trying to upgrade nixpkgs to solve one issue, one suddenly has plenty of other issues.

What was the problem we were trying to solve originally by upgrading? LLVM 11 building on darwin? If so, what I’d try is to copy master’s LLVM files to release-20.09 and see if that works.

[osa1]

Original problem was some kind of build error on Darwin, but I don't remember if it was LLVM or something else.

what I’d try is to copy master’s LLVM files to release-20.09 and see if that works.

I need to clone nixpkgs, make the changes, and point to my fork in sources.json, right?

[nomeata]

I need to clone nixpkgs, make the changes, and point to my fork in sources.json, right?

For local experimentation, you can use NIV_OVERRIDE_nixpkgs=../path/to/your/nixpkgs, but on CI, yes. You can use a branch on https://github.com/dfinity-lab/nixpkgs for that

[nomeata]

I guess after #2542 we can try this again? Might be worth closing this rather than fixing the conflicts, though.

[osa1]

We updated rustc in #2761, closing.