[jtravs]

A draft, on main, independent of #520. Closes #151.

What changed

• Two new CI lanes compile the device targets: the cuda preset under CUDA 12.8, the floor, and the hip preset under ROCm 7.2, tests included, with no device. They run on a change to native/ or to ci.yml, and sit under ci-passed, so a broken device build blocks a merge. They're skipped on any other change, which ci-passed accepts.
• A build from source no longer stops on a compiler's warning. SOLPROP_NATIVE_WARNINGS_AS_ERRORS now follows BUILD_TESTING: on for every preset and for anything building the tests, off for a pip or uv build. CI's wheels and poe check-native's distribution wheel set it on explicitly, so nothing CI builds got less strict. This is the maintainer's decision.
• It fixes a real break on main, the first thing the CUDA lane found: native/tests/native/test_aperture.cpp compiled under GCC, but nvcc couldn't deduce a list of pairs whose later members were bare braces, so a CUDA build with its tests stopped there. Wheel builds don't compile the tests, so hand-run device builds never showed it.

Why

Five device-build breakages were found in 13 days, by #151's table, and each was found only when someone next rented or queued a GPU.

What it costs

A hosted runner, with no cache, in a run that passed:

	Toolkit install	Whole job
HIP, ROCm 7.2	about 2.4 min	11 min alone, 19 min in a full CI run
CUDA 12.8	about 2 min	25 min alone, 26 min in a full CI run

So #151's limit of 30 minutes holds without a Kokkos cache, which I haven't added. These were general.cpp's first compiles there, and it didn't run out of memory. The disk isn't a concern, with 94 GB free after a toolkit's install.

How it was verified

• Each job passes, in this PR's CI run on 4421d978, as lanes of ci.yml under ci-passed: HIP in 19 minutes and CUDA in 26, with the other lanes passing beside them.
• HIP fails when it should: an unused variable in a device-only branch stopped it, -Werror,-Wunused-variable.
• CUDA fails when it should, on the real test_aperture.cpp error above.
• CI builds no CUDA or HIP target, so device builds break unnoticed #151's other test, by a different route than it asks: my planted capture passed the CUDA job, because I'd put if constexpr outside the lambda, which is legal. I then held both forms to nvcc 12.8.1 directly: inside the lambda it stops with "An extended host device lambda cannot first-capture variable in constexpr-if context", and outside it compiles. I didn't spend another CI run on putting the right form through the job.

Not done

• A fault only a card shows still needs a card. The AMD Instinct MI325X fault on Stop the default boundary check doing work nothing uses #516 (Illegal opcode in command stream) compiled cleanly. Tests running on a device stay by hand (No scheduled lane runs the gpu-marked tests on a device #354).
• Native: the HIP build compiles again under -Werror #69's confirmation: CI builds no CUDA or HIP target, so device builds break unnoticed #151 asked the first HIP job to confirm it, and the job passing does.
• Newer toolkits. This uses the floors. CUDA 13 and ROCm 10 are the qualified stacks (The next device qualification: what CUDA and HIP have not run, and two public statements to correct #353), and a second pair of jobs could cover them if you want both.
• The ruleset is unchanged: ci-passed is already the required check, so nothing there needs changing.

Suggested labels: testing, area: native, area: gpu.

🤖 Generated with Claude Code

https://claude.ai/code/session_01XGENPtq9Gp7VCh5QmcDdzQ

[read-the-docs-community]

Documentation build overview

📚 solprop | 🛠️ Build #35656408 | 📁 Comparing 4421d97 against latest (8ed26dd)

  🔍 Preview build  

2 files changed

± dev/tooling.html
± guide/performance/building-native.html

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-05T10:38:39.766695Z	4421d97	Draft marked ready
🔒 Security Review	✅ Completed 2026-11-05T10:38:53.122506Z	4421d97	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.