Upgrade windows docker images base to Ubuntu 24.04

[Raniz85]

Newer versions of Rust require a newer version of MingW.

Fixes #1754, #1737

[Emilgardis]

/ci try --target *indows-gnu

[github-actions]

Try run for comment

• try / conclusion - ❌ Result

Failed Jobs

• x86_64-pc-windows-gnu - ❌ Test Image
• i686-pc-windows-gnu - ❌ Test Image

[Raniz85]

Error is about XDG_RUNTIME_DIR, I'll have a look tomorrow

…

-- Raniz On Wednesday 11 March 2026 at 6:20 pm, github-actions[bot] ***@***.***> wrote:

 GITHUB-ACTIONS[BOT] left a comment (cross-rs/cross#1755)  [#1755 (comment)]      TRY  [https://github.com/cross-rs/cross/actions/runs/22961580714/job/66653409000#step:4:18]  RUN FOR COMMENT  [#1755 (comment)]         * try / conclusion  [https://github.com/cross-rs/cross/actions/runs/22961580714/job/66667154258?pr=1755#]  - ❌ Result    FAILED JOBS         * x86_64-pc-windows-gnu  [https://github.com/cross-rs/cross/actions/runs/22961580714/job/66653812685?pr=1755#]  - ❌ Test Image       * i686-pc-windows-gnu  [https://github.com/cross-rs/cross/actions/runs/22961580714/job/66653812854?pr=1755#]  - ❌ Test Image    —   Reply to this email directly, view it on GitHub  [#1755 (comment)],  or unsubscribe  [https://github.com/notifications/unsubscribe-auth/ABF54R66G4WZD2BEBPHZX634QGN4LAVCNFSM6AAAAACWOOL5Y6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DANBQHAYTAMBYGU].    You are receiving this because you authored the thread.Message ID: ***@***.***>

[Emilgardis]

have a look at #1580 also, where we've been trying to bump the image

[Raniz85]

This turned out to be quite the rabbit hole. Made deeper because I haven't worked with C++ for some 20 years :S

I've spent considerable time trying to figure out what goes wrong - with a lot of help from pushing Claude in various directions.

The issue seems to be that the rust-cmake-hello-world crate is incompatible with newer GCC.
The reason that Claude gave me is that re2 uses call_once which is reliant on thread locals and this doesn't work with MingW-GCC-13 across DLL-boundaries. I found some info online that seems to support this (among others, this: llvm/llvm-project#161039).

The solution is to either force libstdc++ to be linked statically (I don't like this solution) or to upgrade RE2 (newer versions doesn't seem to use call_once, which turned out to be a bit of a rabbit hole itself :)

Either way, the fix needs to be done in https://github.com/cross-rs/rust-cmake-hello-world.

I think that upgrading RE2 is the better option, but that will break the test for most (all?) other images right now since they don't support C++17, which Abseil (a dependency of RE2 since ~May 2024) requires - it could help the effort in #1850 though since 24.04 has toolchains that support C++17.

There's a PR open on rust-cmake-hello-world to upgrade RE2, once that is merged, the CI run should pass for these images again (though it will break for pretty much all the other images that use GCC-8 or older).

One option would be to point test.sh to specific revisions of rust-cmake-hello-world instead of main until all images have been updated.

[Raniz85]

Two questions:

1. Should I change test.sh to use different refs for rust-cmake-hello-world depending on which image we're testing?
2. Should we combine this PR with Updated Docker image to Ubuntu-24.04 and Cargo dependencies #1580?

[Emilgardis]

I think using different refs makes sense, I switched the PR to be put on branch 24 so we can make the test script use that branch instead when needed

[Emilgardis]

We can combine the prs, but we can also just rebase the other one on top of it :) small incremental changes are nice though also

[Raniz85]

I agree :)

I'll update the test script to use the new ref and then we can retry the image tests

[MichaIng]

Since Debian/Ubuntu dropped sjlj for dwarf already (which you see in your updated patch as well), this isn't a reason anymore to compile the mingw compilers from source. Using the distro packages also makes assuring the POSIX variant obsolete, since there are explicit packages for this now: g++-mingw-w64-x86-64-posix gfortran-mingw-w64-x86-64-posix

So that way, the mingw.sh could be skipped and removed.

I am working on this on my fork as a continuation of #1580: MichaIng#2

But for cargo tests with --features=re2, I get a null pointer dereference:

wine: Unhandled page fault on execute access to 0000000000000000 at address 0000000000000000

Hence I removed cpp = true from the targets.toml for the two windows-gnu targets.

Or am I missing something, and own mingw compiler builds are still needed, and probably solve this particular issue I see with re2?

---

What I am still fiddling with on my fork is Solaris: #1599 #1730
I recognized a little late that I am working on issues unrelated to Ubuntu 20.04 => 24.04, but as I am on it ... DilOS 2 libs are too old for recent Rust, it seems DilOS 3 is inevitable (not released, just saw the APT suite, but it is empty). But I am still trying to do some linker mapping from new/missing to old/existing symbols. But this is another topic. I'll try to get everything green on my fork, and open a new PR then.