ci: exercise the in-process LLVM backend (kill-policy arm for #7301)

.github/workflows/llvm-inprocess.yml

@@ -0,0 +1,111 @@
+# CI arm for the in-process LLVM backend (#7241, merged in #7301).
+#
+# The GC knob kill-policy (CLAUDE.md) demands every shipped mode be exercised
+# by CI or deleted; this job is the exercise for `PERRY_LLVM_INPROCESS`.
+# Each step asserts its subject was LIVE (liveness line, diff verdict) rather
+# than merely that nothing threw — see "Four ways a gate can be unable to
+# fail". NON-REQUIRED until it has run green once; promote afterwards
+# (a new gate has never been green, so promoting first blocks every PR).
+name: llvm-inprocess
+
+# No trigger-level `paths:` — deliberately. A required check whose workflow
+# is path-filtered at the trigger never CREATES a check run for PRs outside
+# those paths, so the required context sits "waiting" forever and blocks the
+# merge (the promotion trap CodeRabbit flagged on #7304). Filtering lives in
+# the `changes` job instead: a job skipped by `if:` still reports a check
+# run (conclusion: skipped), which branch protection accepts.
+on:
+  pull_request:
+  push:
+    branches: [main]
+
+concurrency:
+  group: llvm-inprocess-${{ github.ref }}
+  # Gate-trap 3: cancel superseded PR runs, but NEVER cancel main runs — a
+  # busy merge day would otherwise starve the gate to zero executions.
+  cancel-in-progress: ${{ github.event_name == 'pull_request' }}
+
+jobs:
+  # Cheap relevance filter via the PR files API (no checkout, no third-party
+  # action). Pushes to main always run — main executions are the gate's
+  # anchor and the promotion prerequisite.
+  changes:
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    permissions:
+      contents: read
+      pull-requests: read
+    outputs:
+      relevant: ${{ steps.filter.outputs.relevant }}
+    steps:
+      - id: filter
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          if [ "${{ github.event_name }}" = "push" ]; then
+            echo "relevant=true" >> "$GITHUB_OUTPUT"
+            exit 0
+          fi
+          files=$(gh api "repos/${{ github.repository }}/pulls/${{ github.event.pull_request.number }}/files"             --paginate --jq '.[].filename')
+          if echo "$files" | grep -qE '^(crates/perry-codegen/|crates/perry/src/commands/compile/|experiments/llvm-inprocess-spike/|benchmarks/app-patterns/kernels/batch\.ts$|\.github/workflows/llvm-inprocess\.yml$)'; then
+            echo "relevant=true" >> "$GITHUB_OUTPUT"
+          else
+            echo "relevant=false" >> "$GITHUB_OUTPUT"
+          fi
+
+  native-backend:
+    needs: changes
+    if: needs.changes.outputs.relevant == 'true'
+    runs-on: macos-15
+    timeout-minutes: 90
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install LLVM 22 (pinned major — fail loudly on drift)
+        run: |
+          set -euo pipefail
+          brew install llvm@22 2>/dev/null || brew install llvm
+          PREFIX="$(brew --prefix llvm@22 2>/dev/null || brew --prefix llvm)"
+          # llvm-sys 221 requires major 22. If the runner's formula moves on,
+          # this must go red, not quietly build something else.
+          "$PREFIX/bin/llvm-config" --version | grep -q '^22\.'
+          echo "LLVM_SYS_221_PREFIX=$PREFIX" >> "$GITHUB_ENV"
+
+      - name: Build with the llvm-inprocess feature
+        run: |
+          cargo build --profile perry-dev -p perry -p perry-runtime-static \
+            -p perry-stdlib-static --features perry/llvm-inprocess
+
+      - name: Unit gates (528 incl. corpus construction + RS4GC pin)
+        run: |
+          set -euo pipefail
+          out=$(cargo test --profile perry-dev -p perry-codegen \
+            --features llvm-inprocess --lib 2>&1) || { echo "$out"; exit 1; }
+          # The corpus gates must have RUN, not skipped: a checkout missing
+          # the tracked .ll corpora would otherwise green vacuously.
+          echo "$out" | grep -q "dialect::tests::corpus_spike ... ok"
+          echo "$out" | grep -q "dialect::tests::corpus_batch_kernel ... ok"
+          echo "$out" | grep -q "inprocess::tests::rs4gc_schedules_in_process ... ok"
+
+      - name: Native-mode smoke — liveness, behavior parity, object-byte verdicts
+        run: |
+          set -euo pipefail
+          export PERRY_RUNTIME_DIR="$PWD/target/perry-dev"
+          export PERRY_NO_AUTO_OPTIMIZE=1
+          BIN=target/perry-dev/perry
+          SRC=experiments/llvm-inprocess-spike/spike.ts
+
+          "$BIN" "$SRC" -o /tmp/spike_text
+          /tmp/spike_text > /tmp/text.out
+
+          PERRY_LLVM_INPROCESS=native "$BIN" "$SRC" -o /tmp/spike_native 2> /tmp/native.err
+          grep -q "in-process LLVM backend active" /tmp/native.err
+          /tmp/spike_native > /tmp/native.out
+          cmp /tmp/text.out /tmp/native.out
+
+          PERRY_LLVM_INPROCESS=diff "$BIN" "$SRC" -o /tmp/spike_diff 2> /tmp/diff.err
+          grep -q "ir-diff. OK" /tmp/diff.err
+
+          PERRY_LLVM_INPROCESS=diff PERRY_CODEGEN_UNITS=3 "$BIN" \
+            benchmarks/app-patterns/kernels/batch.ts -o /tmp/batch_diff 2> /tmp/diffu.err
+          grep -q "ir-diff. OK.*3 units" /tmp/diffu.err

changelog.d/7304-llvm-inprocess-ci-arm.md

@@ -0,0 +1,7 @@
+CI arm for the in-process LLVM backend (#7301 follow-up, kill-policy): a
+path-filtered macOS job builds the `llvm-inprocess` feature against a
+loudly-pinned LLVM 22, runs the perry-codegen suite asserting the corpus
+and RS4GC gates actually ran, and smokes `PERRY_LLVM_INPROCESS=native`
+end-to-end — liveness line, behavior parity with the text arm, and
+object-byte `=diff` verdicts for a single module and a forced 3-unit
+split. Lands non-required; promote after first green.