ci: change runner to solx-linux-amd64-self-hosted

[nebasuke]

• Switch the runner from solx-linux-amd64 (per-run) to solx-linux-amd64-self-hosted (persistent) in the three workflows that target it: integration-tests.yaml, coverage.yaml, sanitizer.yaml.
• In integration-tests.yaml, share the solc build between the PR checkout and the temp-solx-main checkout via symlink when their solc submodule SHAs match — same pattern as the existing LLVM-sharing step.
• Refactor the LLVM share step to use the same output+skip pattern (steps.share-llvm.outputs.shared) as the new solc step instead of building inline, so the two share/build pairs are symmetric.
• Clear stale dirs/symlinks (rm -rf before ln -snf) in both share steps so a previous run's main-build output on the persistent self-hosted workspace can't be reused.

Runs

Integration tests:

• https://github.com/NomicFoundation/solx/actions/runs/25048377992
• https://github.com/NomicFoundation/solx/actions/runs/25053126423

[github-actions]

📊 solx Tester Report

➡️ Download

[github-actions]

📊 Hardhat Projects Report

➡️ Download

[github-actions]

📊 Foundry Projects Report

➡️ Download

[github-actions]

Coverage Summary

Crate	Line Coverage	Function Coverage
solx	🟢 94.7%	🟡 68.0%
solx-benchmark-converter	🔴 0.0%	🔴 0.0%
solx-codegen-evm	🟢 88.2%	🟢 80.4%
solx-compiler-downloader	🔴 0.0%	🔴 0.0%
solx-core	🟢 96.0%	🟢 88.0%
solx-dev	🔴 2.5%	🔴 3.4%
solx-evm-assembly	🟡 73.8%	🟡 64.8%
solx-solc-test-adapter	🔴 1.7%	🔴 2.1%
solx-standard-json	🟢 95.1%	🟢 89.4%
solx-tester	🔴 37.8%	🔴 35.7%
solx-utils	🟢 83.0%	🟢 80.6%
solx-yul	🟡 78.9%	🟡 71.3%
Total	🟡 56.2%	🔴 48.7%

Codecov Report | HTML Report | Workflow Run

[Copilot]

Pull request overview

This PR updates CI workflows that run on the solx Linux AMD64 runner to use a persistent self-hosted runner, and adds/refactors build-artifact sharing between the PR checkout and a secondary temp-solx-main checkout to reduce redundant builds while avoiding stale reuse.

Changes:

• Switch runs-on from solx-linux-amd64 to solx-linux-amd64-self-hosted in the sanitizer, coverage, and integration workflows.
• In integration-tests.yaml, refactor LLVM sharing to an output-driven “share then conditional build” pattern, and add an equivalent solc sharing step via symlinks when submodule SHAs match.
• Add cleanup (rm -rf before ln -snf) in sharing steps to prevent stale paths/symlinks from being reused on a persistent workspace.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
.github/workflows/sanitizer.yaml	Switch sanitizer workflow to the persistent self-hosted runner label.
.github/workflows/coverage.yaml	Switch coverage workflow to the persistent self-hosted runner label.
.github/workflows/integration-tests.yaml	Switch to self-hosted runner; refactor LLVM sharing/build flow; add solc sharing/build flow with stale-path cleanup.

[hedgar2017]

@nebasuke — stepping back from the per-line comments for a moment, the three problems flagged here (the two stale-symlink fallback paths, plus the build-script-hash dimension that share-solc misses) all come from the same place: the workflow is reimplementing a SHA-keyed mirror in bash, in parallel to the cache-key logic the build-llvm and build-solc composite actions already own.

Worth considering folding the mirror in as a mirror-from input on those actions instead. The workflow then collapses to two ordinary uses: calls per artifact:

- name: Build LLVM (PR)
  uses: ./.github/actions/build-llvm
  with: { build-type: Release, enable-assertions: 'false' }

- name: Build LLVM (main)
  uses: ./.github/actions/build-llvm
  continue-on-error: true
  with:
    working-dir: temp-solx-main
    mirror-from: '.'
    build-type: Release
    enable-assertions: 'false'

- name: Build solc (PR)
  uses: ./.github/actions/build-solc

- name: Build solc (main)
  uses: ./.github/actions/build-solc
  continue-on-error: true
  with:
    working-dir: temp-solx-main/solx-solidity
    mirror-from: solx-solidity

Inside the action, a single new step gated on inputs.mirror-from != '', falling through to the existing cache-restore + build chain when it can't mirror:

- name: Mirror artifact from sibling checkout
  id: mirror
  if: inputs.mirror-from != ''
  shell: bash
  run: |
    SRC="${{ inputs.mirror-from }}/target-llvm/target-final"
    DST="${{ inputs.working-dir }}/target-llvm/target-final"
    SRC_KEY=$(<compute the same cache key, with mirror-from's SHAs>)
    [ "$SRC_KEY" = "${{ steps.llvm-cache.outputs.key }}" ] && [ -d "$SRC" ] || {
      echo "mirrored=false" >> "$GITHUB_OUTPUT"; exit 0
    }
    rm -rf "$DST"
    mkdir -p "$(dirname "$DST")"
    cp -al "$SRC" "$DST"
    echo "mirrored=true" >> "$GITHUB_OUTPUT"

Existing build steps' if: extends to ... && steps.mirror.outputs.mirrored != 'true'.

Two things this gets that the surgical fixes won't:

• Cleanup is unconditional and adjacent to the write — rm -rf "$DST" sits one line above cp -al, so it can't drift out of the SHA-match branch the way the current rm -rf did. Submodule-internal stale state and Copilot's broader "stale temp-solx-main" concern collapse into the same fix.
• The mirror predicate inherits the action's full cache-key dimensions for free — boost version, MLIR flag, build-script hash, sanitizer — so the share decision can't drift from the cache decision as keys grow.

Also cp -al instead of ln -snf. Hardlinks under cmake/cargo's rename-on-write conventions don't have the through-write surface symlinks do, and the consumers here are read-only.

Bigger than a surgical fix on this PR — happy to take it as a follow-up if you'd rather close #375 with the rm -rf cleanup and the script-hash check in the share predicate (per the inline comments).

[nebasuke]

Thanks — going with the surgical path on this PR. Filed the mirror-from refactor as #448 to track separately.

What landed here (path B):

• aa3b16d wipes temp-solx-main wholesale before the optional Checkout main. That's your "delete temp-solx-main before the optional checkout" variant, and it closes both stale-symlink concerns (LLVM and solc submodule-internal) plus Copilot's broader stale-checkout flag in one step.
• 484673c + da1d918 add the build-script-hash dimension to both share predicates, mirroring the cache-key composition in build-{llvm,solc}/action.yml.
• 0be720f extracts the find | sort -z | xargs -0 cat | sha256sum | cut pipeline into a named hash_tree shell function — it's computed twice per step (PR vs main) and the inline one-liner obscured what was being compared.

Why not path A on this PR: the architectural win (mirror predicate inheriting the action's cache-key dimensions for free, hardlinks instead of symlinks, cleanup adjacent to the write) is real and worth doing, but it rewrites the composite actions — bigger blast radius than the three defects this PR is fixing. Easier to review the surgical patch against the inline comments and ship that, then refactor cleanly under #448.