fix(ci): force clang for Linux x86_64 native builds (#1059)

Verified locally: gcc 14.2.0 (Debian Trixie) builds a binary whose
tree-sitter-hcl extractor silently drops the 4 HCL fixture files in
tests/benchmarks/resolution/fixtures/hcl/. Clang 14 (Debian Bookworm)
builds a binary that parses them correctly. Same Rust source, same
Cargo.lock — only the C compiler differs (cc-rs uses CC for the
bundled parser.c).

The drop triggers an engine-parity WASM backfill on every full-build
call (#1054), and the v3.10.0 pre-publish gate flagged the resulting
~1.7s overhead per call as a regression vs the v3.9.6 published binary
(which evidently was built before this gcc/parser-c interaction
manifested).

Forces `CC=clang`/`CXX=clang++` in three places that build the host
x86_64-unknown-linux-gnu artifact:

- publish.yml `build-native` matrix entry for the gate's binary
- publish.yml `preflight` job (so npm test parity matches published)
- build-native.yml's matching matrix entry (host build for PR checks)
- ci.yml `native-host-build` ubuntu-latest leg (so PR test runs match)

Cross-compile targets (aarch64-gnu, x86_64-musl) are unaffected — they
already pin their own toolchains. macOS and Windows are unaffected.

Refs #1054

Author: carlos-alm

.github/workflows/build-native.yml

@@ -72,6 +72,17 @@ jobs:
         with:
           workspaces: crates/codegraph-core
 
+      # Force clang for the host x86_64-linux-gnu build. With gcc (the runner's
+      # default `cc`) the bundled C parsers in tree-sitter-hcl produce a binary
+      # whose HCL extractor silently drops files (#1054). Clang produces a
+      # working binary.
+      - name: Use clang for x86_64-linux-gnu
+        if: matrix.target == 'x86_64-unknown-linux-gnu'
+        run: |
+          clang --version
+          echo "CC=clang" >> "$GITHUB_ENV"
+          echo "CXX=clang++" >> "$GITHUB_ENV"
+
       - name: Install cross-compilation tools (aarch64-gnu)
         if: matrix.target == 'aarch64-unknown-linux-gnu'
         run: |

.github/workflows/ci.yml

@@ -61,6 +61,16 @@ jobs:
         with:
           workspaces: crates/codegraph-core
 
+      # Force clang on Linux runners. With gcc (the default cc) the bundled
+      # C parsers in tree-sitter-hcl produce a binary whose HCL extractor
+      # silently drops files (#1054). Clang produces a working binary.
+      - name: Use clang on Linux
+        if: matrix.os == 'ubuntu-latest'
+        run: |
+          clang --version
+          echo "CC=clang" >> "$GITHUB_ENV"
+          echo "CXX=clang++" >> "$GITHUB_ENV"
+
       - name: Install napi-rs CLI
         run: npm install -g @napi-rs/cli@3
 

.github/workflows/publish.yml

@@ -44,6 +44,15 @@ jobs:
         with:
           workspaces: crates/codegraph-core
 
+      # Match the build-native job: gcc-built tree-sitter parsers drop files
+      # on Linux x86_64 (#1054). Use clang here too so preflight parity tests
+      # exercise the same binary characteristics as the published artifact.
+      - name: Use clang for native build
+        run: |
+          clang --version
+          echo "CC=clang" >> "$GITHUB_ENV"
+          echo "CXX=clang++" >> "$GITHUB_ENV"
+
       - name: Install napi-rs CLI
         run: npm install -g @napi-rs/cli@3
 
@@ -178,6 +187,18 @@ jobs:
         with:
           workspaces: crates/codegraph-core
 
+      # Force clang for the host x86_64-linux-gnu build. With gcc (the runner's
+      # default `cc`) the bundled C parsers in tree-sitter-hcl produce a binary
+      # whose HCL extractor silently drops files (#1054), triggering an
+      # expensive WASM backfill on every build call. Clang produces a working
+      # binary. Verified locally: gcc 14.2.0 drops .tf files, clang does not.
+      - name: Use clang for x86_64-linux-gnu
+        if: matrix.target == 'x86_64-unknown-linux-gnu'
+        run: |
+          clang --version
+          echo "CC=clang" >> "$GITHUB_ENV"
+          echo "CXX=clang++" >> "$GITHUB_ENV"
+
       - name: Install cross-compilation tools (aarch64-gnu)
         if: matrix.target == 'aarch64-unknown-linux-gnu'
         run: |