sonarcloud: merge each shard's profiles before the scan downloads them (#210)

Each coverage shard now merges its own raw profiles into one sparse
.profdata with the archive toolchain's llvm-profdata and uploads only
that. The scan job merges the shard profiles instead of downloading
every raw profile beside the test archive, which on 11 Nov 2026 filled
the runner for vectorcraft (four ~7 GB zipped profraw artifacts).

A raw profile llvm-profdata cannot read, such as one cut short by a
test process that crashed while writing it (lightcraft, 11 Nov 2026),
now fails its shard with an error naming the file, instead of being
uploaded and failing the scan's merge.

Single-shard callers keep the cargo llvm-cov path unchanged.

.github/workflows/sonarcloud.yml

@@ -398,11 +398,44 @@ jobs:
             echo "::error::shard $SHARD produced no coverage profiles"
             exit 1
           fi
+      - name: Install LLVM tools for the archive's toolchain
+        run: rustup toolchain install "$(cat archive/rustc-release.txt)" --profile minimal --component llvm-tools-preview
+      - name: Merge this shard's profiles
+        # Raw profiles of a large workspace run to tens of gigabytes per shard; the scan job cannot
+        # hold every shard's at once. One sparse indexed profile per shard is a small fraction of that.
+        shell: bash
+        env:
+          SHARD: ${{ matrix.shard }}
+        run: |
+          set -euo pipefail
+          toolchain="$(cat archive/rustc-release.txt)"
+          llvm_bin="$(rustup run "$toolchain" rustc --print sysroot)/lib/rustlib/$(rustup run "$toolchain" rustc -vV | sed -n 's/^host: //p')/bin"
+          echo "Raw profiles: $(du -sh profraw | cut -f1) in $(find profraw -name '*.profraw' | wc -l) files"
+          df -h .
+          mkdir -p profdata
+          # failure-mode=any stops on any unreadable input. A profile cut short (a test process that
+          # crashed while writing it, or a full disk) fails this shard and names the file, instead of
+          # being uploaded and failing the scan's merge later with no shard to look at.
+          status=0
+          "$llvm_bin/llvm-profdata" merge -sparse --failure-mode=any profraw/*.profraw \
+            -o "profdata/shard-$SHARD.profdata" 2>merge-errors.txt || status=$?
+          if [ "$status" -ne 0 ]; then
+            cat merge-errors.txt >&2
+            sed -n 's/^warning: \(.*\.profraw\): \(.*\)$/\1: \2/p' merge-errors.txt | while IFS= read -r bad; do
+              echo "::error::shard $SHARD wrote a truncated or corrupt coverage profile: $bad"
+            done
+            echo "::error::llvm-profdata could not merge shard $SHARD's profiles (exit $status)"
+            rm -rf profdata
+            exit "$status"
+          fi
+          rm -rf profraw merge-errors.txt
+          echo "Merged profile: $(du -sh "profdata/shard-$SHARD.profdata" | cut -f1)"
+          df -h .
       - uses: actions/upload-artifact@cf430e030ddbb5b0abf93d22962f4752f3646cd9 # v7.0.2
         with:
-          name: profraw-${{ matrix.shard }}
-          path: profraw/
-          if-no-files-found: ignore
+          name: profdata-${{ matrix.shard }}
+          path: profdata/
+          if-no-files-found: error
           retention-days: 1
 
   sonarcloud:
@@ -577,8 +610,9 @@ jobs:
         if: steps.token.outputs.available == 'true' && inputs.shards > 1 && hashFiles('Cargo.toml') != ''
         uses: actions/download-artifact@9000827ccba6bdab643e8b6fd33ac0654aef8333 # v8.0.2
         with:
-          pattern: profraw-*
-          path: profraw
+          # One merged profile per shard (shard-N.profdata), never the raw profiles.
+          pattern: profdata-*
+          path: profdata
           merge-multiple: true
       - name: Download test archive
         if: steps.token.outputs.available == 'true' && inputs.shards > 1 && hashFiles('Cargo.toml') != ''
@@ -597,10 +631,10 @@ jobs:
           toolchain="$(cat archive/rustc-release.txt)"
           llvm_bin="$(rustup run "$toolchain" rustc --print sysroot)/lib/rustlib/$(rustup run "$toolchain" rustc -vV | sed -n 's/^host: //p')/bin"
           df -h . "$RUNNER_TEMP"
-          # Do not hold every raw profile beside the expanded archive. Retain the raw inputs
+          # Each shard merged its own raw profiles; combine those. Retain the shard profiles
           # until the merge succeeds, then release them before unpacking the test binaries.
-          "$llvm_bin/llvm-profdata" merge -sparse profraw/*.profraw -o coverage.profdata
-          rm -f profraw/*.profraw
+          "$llvm_bin/llvm-profdata" merge -sparse profdata/*.profdata -o coverage.profdata
+          rm -f profdata/*.profdata
           df -h . "$RUNNER_TEMP"
           # Unpack outside the checkout: the scanner analyses everything under the workspace, and build
           # output (generated sources) there would be counted as duplicated code.