ci: extract the standard-config toolchain prologue into build-toolchain

[nebasuke]

Share more of the toolchain between Tests and Slang tests, and fix the Free disk space (macOS) mixing for Slang while we're at it.

Claude summary

Why now

Run 29762518487 (Slang Tests on #554, macOS arm64) failed with ld: write() failed, errno=28 — ENOSPC at 98% disk during the test-binary link. test.yaml gained a free-disk-space-macos prologue step; slang-tests.yaml never did. That's the drift class this PR removes: the same six-step prologue existed in five near-identical copies, and each new prologue step had to be added five times.

What

New .github/actions/build-toolchain composite encoding the standard prologue exactly once:

checkout-submodules → setup-sfw (optional) → prepare-msys (Windows) → free-disk-space-macos → build-llvm → build-solc

with llvm:/solc: toggles. Converted call sites:

Call site	Config
test.yaml / cargo-checks	solc: 'false'
test.yaml / build-and-test	defaults
cache-warmup.yaml / warm-llvm	solc: 'false'
cache-warmup.yaml / warm-solc	llvm: 'false'
slang-tests.yaml	solc: 'false' (keeps its own Release+MLIR solc build for the lit tests)

The LLVM/solc build params in the composite define the standard-config cache keys — drift between copies was a silent cache miss; now changing them re-keys all consumers together.

Behavior changes (all deliberate)

• slang-tests + cache-warmup macOS legs now run free-disk-space-macos — previously missing exactly where the coldest LLVM builds happen. Fixes the fix(tester): resolve the Slang deploy target in source order #554 failure mode.
• Windows: fresh llvm-sys builds cannot link against the bundled static libstdc++.a under MSYS2 GCC 16 — CI is one cache-fingerprint change from breaking #550 libstdc++ workaround applies to every standard-config Windows consumer: the bundled static libstdc++.a is dropped from the restored LLVM tree (after build-llvm's mid-action cache save, so the cached tree keeps it). Defuses the landmine documented in Windows: fresh llvm-sys builds cannot link against the bundled static libstdc++.a under MSYS2 GCC 16 — CI is one cache-fingerprint change from breaking #550 for test.yaml's Windows leg, which was green only while the cargo cache kept llvm-sys Fresh. Release binaries keep static linking (release.yaml doesn't use this action).
• rust-cache steps move after the composite: prepare-msys switches the rustup host (part of the Swatinem key), so the post-msys position keeps hitting caches already saved on main. Late restore costs nothing on artifact-cache hits.
• cache-warmup warm jobs get SFW-first order (previously msys-before-SFW) — matching test.yaml's proven order.
• warm-solc drops explicit boost-version: '1.83.0' — it's the build-solc default; solc cache key unchanged.
• Inline submodule-checkout scripts → existing checkout-submodules action (same update, plus --init).

Not included

• Specialty configs (sanitizer, coverage, integration, release) stay on direct build-llvm/build-solc calls: one-producer/one-consumer pairs with distinct params; folding them in would make the composite a switch bag.
• The shared 5-leg matrix stays duplicated — composite actions can't carry a matrix, and a reusable workflow would drag job-level structure (container/volumes anchors, permissions, per-workflow cache config) into a shared file for little gain.

Relation to #553

Extracted from the cross-os-tester branch (#553), which will rebase onto this and add cross-os-tests.yaml as a sixth consumer. One deliberate difference: this version passes optional: 'true' to setup-sfw, preserving what every current call site does. The #553 copy omits it, and setup-sfw defaults to mandatory — that would make a socket.dev outage fail all five jobs, a change the extraction shouldn't smuggle in (worth a look on #553 whether that was intended).

Validation

• actionlint clean on all three touched workflows.
• Cache-key-neutral by construction for LLVM/solc artifact caches and Swatinem caches (params unchanged, post-msys cache position preserved for build-and-test; cargo-checks is Linux-container-only so its key has no msys sensitivity).

[Copilot]

Pull request overview

This PR reduces CI workflow drift by extracting the shared “standard toolchain prologue” (submodule init + optional Socket Firewall + platform prep + disk cleanup on macOS + standard LLVM/solc builds) into a single composite action, and then reusing it across the main test, slang test, and cache warmup workflows. This centralizes the standard-config cache-key-defining build parameters and ensures macOS Slang legs also run the disk cleanup step that previously drifted out.

Changes:

• Introduce a new composite action (build-toolchain) that encodes the standard CI toolchain prologue with llvm/solc toggles.
• Update test.yaml, slang-tests.yaml, and cache-warmup.yaml to use build-toolchain, aligning macOS disk cleanup and Windows MSYS prep across consumers.
• Standardize cache behavior by consolidating the LLVM/solc build parameter source of truth in the shared action.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
.github/workflows/test.yaml	Replaces duplicated prologue steps with build-toolchain, and positions rust-cache after toolchain setup.
.github/workflows/slang-tests.yaml	Switches to build-toolchain (LLVM only) while continuing to build a separate Release solc for lit tests.
.github/workflows/cache-warmup.yaml	Uses build-toolchain for warm-LLVM and warm-solc jobs to keep standard-config warming consistent.
.github/actions/build-toolchain/action.yml	Adds the new shared composite action implementing the standard CI prologue and standard LLVM/solc build parameters.

[github-actions]

Coverage Summary

Crate	Line Coverage	Function Coverage
solx	🟢 94.7%	🟡 68.0%
solx-benchmark-converter	🔴 0.0%	🔴 0.0%
solx-codegen-evm	🟢 88.9%	🟢 83.4%
solx-compiler-downloader	🔴 0.0%	🔴 0.0%
solx-core	🟢 96.6%	🟢 91.0%
solx-dev	🔴 2.4%	🔴 3.0%
solx-evm-assembly	🟡 70.6%	🟡 66.2%
solx-solc-test-adapter	🔴 1.7%	🔴 2.1%
solx-standard-json	🟢 95.4%	🟢 91.5%
solx-tester	🔴 38.2%	🔴 35.9%
solx-utils	🟢 83.8%	🟢 83.6%
solx-yul	🟡 78.9%	🟡 71.3%
Total	🟡 53.9%	🔴 47.0%

Codecov Report | HTML Report | Workflow Run