Validate chdb-core main artifacts in Rust CI

.github/workflows/core-main-artifact-check.yml

@@ -0,0 +1,130 @@
+name: Check chdb-core main artifacts
+
+# A one-off, pre-release consumer check. The normal Rust workflow continues to
+# test its release pin; this one tests four CI artifacts from the same core SHA.
+on:
+  pull_request:
+    branches: [main]
+    paths:
+      - .github/workflows/core-main-artifact-check.yml
+      - tests/bundled_runtime_hidden.rs
+      - tests/runtime_faces.rs
+
+permissions:
+  contents: read
+
+jobs:
+  main_artifacts:
+    if: github.head_ref == 'codex/chdb-core-26.9.2-rc.2'
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - os: ubuntu-latest
+            run_id: '38427250828'
+            artifact: chdb-artifacts-linux-x86_64
+            asset_prefix: linux-x86_64
+          - os: ubuntu-24.04-arm
+            run_id: '38429301339'
+            artifact: chdb-artifacts-linux-aarch64
+            asset_prefix: linux-aarch64
+          - os: macos-latest
+            run_id: '38427008431'
+            artifact: chdb-artifacts-macos-arm64
+            asset_prefix: macos-arm64
+            test_min_macos: '12.0'
+          - os: macos-15-intel
+            run_id: '38428913578'
+            artifact: chdb-artifacts-macos-x86_64
+            asset_prefix: macos-x86_64
+            test_min_macos: '12.0'
+    runs-on: ${{ matrix.os }}
+    env:
+      CARGO_TERM_COLOR: always
+      CORE_SHA: dec391dcdf788cbecede338a0f305b399d8495e1
+      CORE_RUN_ID: ${{ matrix.run_id }}
+      CORE_ARTIFACT: ${{ matrix.artifact }}
+      ASSET_PREFIX: ${{ matrix.asset_prefix }}
+      CHDB_TEST_MIN_MACOS: ${{ matrix.test_min_macos }}
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Verify source and fetch the exact CI artifact
+        env:
+          # A read-only fine-grained token for chdb-core Actions can be added
+          # as this secret if the Rust repository's token cannot read it.
+          GH_TOKEN: ${{ secrets.CHDB_CORE_ARTIFACT_READ_TOKEN || github.token }}
+        run: |
+          set -euo pipefail
+          actual=$(gh run view "$CORE_RUN_ID" -R chdb-io/chdb-core \
+            --json headSha,conclusion,headBranch \
+            --jq '[.headSha,.conclusion,.headBranch] | join(" ")')
+          if [ "$actual" != "$CORE_SHA success main" ]; then
+            echo "::error::core run $CORE_RUN_ID is '$actual', expected '$CORE_SHA success main'"
+            exit 1
+          fi
+          gh run download "$CORE_RUN_ID" -R chdb-io/chdb-core \
+            -n "$CORE_ARTIFACT" -D "$RUNNER_TEMP/core-artifact"
+
+      - name: Unpack matching native libraries and headers
+        run: |
+          set -euo pipefail
+          DYNAMIC_DIR="$RUNNER_TEMP/core-dynamic"
+          STATIC_DIR="$RUNNER_TEMP/core-static"
+          mkdir -p "$DYNAMIC_DIR" "$STATIC_DIR"
+          tar -xzf "$RUNNER_TEMP/core-artifact/$ASSET_PREFIX-libchdb.tar.gz" -C "$DYNAMIC_DIR"
+          tar -xzf "$RUNNER_TEMP/core-artifact/$ASSET_PREFIX-libchdb-static.tar.gz" -C "$STATIC_DIR"
+          test -s "$DYNAMIC_DIR/libchdb.so"
+          test -s "$STATIC_DIR/libchdb.a"
+          cmp "$DYNAMIC_DIR/chdb.h" "$STATIC_DIR/chdb.h"
+          echo "Using chdb-core $CORE_SHA: $CORE_ARTIFACT from run $CORE_RUN_ID"
+
+      - name: Check formatting and dynamic Clippy
+        if: matrix.os == 'ubuntu-latest'
+        env:
+          CHDB_LIB_DIR: ${{ runner.temp }}/core-dynamic
+          CHDB_INCLUDE_DIR: ${{ runner.temp }}/core-dynamic
+        run: |
+          cargo fmt --check
+          cargo clippy --all-targets -- --deny warnings
+
+      - name: Build and test dynamic library
+        env:
+          CHDB_LIB_DIR: ${{ runner.temp }}/core-dynamic
+          CHDB_INCLUDE_DIR: ${{ runner.temp }}/core-dynamic
+          LD_LIBRARY_PATH: ${{ runner.temp }}/core-dynamic
+          DYLD_LIBRARY_PATH: ${{ runner.temp }}/core-dynamic
+          RUST_BACKTRACE: full
+        run: |
+          cargo build --verbose
+          cargo test -- --test-threads=1
+          cargo test --features durable -- --test-threads=1
+          if [ "$RUNNER_OS" = Linux ] && [ "$RUNNER_ARCH" = X64 ]; then
+            cargo test --features durable-s3 --lib -- --test-threads=1 durable::backends
+          fi
+
+      - name: Check all-feature Clippy against the static archive
+        if: matrix.os == 'ubuntu-latest'
+        env:
+          CHDB_LIB_DIR: ${{ runner.temp }}/core-static
+          CHDB_INCLUDE_DIR: ${{ runner.temp }}/core-static
+        run: cargo clippy --all-targets --all-features -- --deny warnings
+
+      - name: Build, inspect and run statically linked tests
+        env:
+          CHDB_LIB_DIR: ${{ runner.temp }}/core-static
+          CHDB_INCLUDE_DIR: ${{ runner.temp }}/core-static
+          CHDB_CHECK_HOST_UNWIND: '1'
+          RUST_BACKTRACE: full
+        run: |
+          set -euo pipefail
+          cargo test --features static --no-run --tests -j 2 --message-format=json > "$RUNNER_TEMP/artifacts.json"
+          jq -r 'select(.reason == "compiler-artifact" and .profile.test == true and .executable != null)
+                 | .executable' "$RUNNER_TEMP/artifacts.json" > "$RUNNER_TEMP/test-binaries.txt"
+          test -s "$RUNNER_TEMP/test-binaries.txt"
+          xargs .github/scripts/check-static-link.sh < "$RUNNER_TEMP/test-binaries.txt"
+          while read -r binary <&3; do
+            echo "== $binary"
+            "$binary" --test-threads=1 < /dev/null
+          done 3< "$RUNNER_TEMP/test-binaries.txt"
+          cargo test --features static --doc

tests/bundled_runtime_hidden.rs

@@ -1,15 +1,15 @@
-//! The bundled C++ runtime must not be visible from outside the archive.
+//! The bundled C++ runtime must not capture symbols used by the host runtime.
 //!
-//! This is the cause the runtime cases in `runtime_faces.rs` observe the effect
-//! of, checked directly and without linking anything.
+//! The runtime cases in `runtime_faces.rs` check the effect in a linked binary;
+//! this checks the archive's symbols directly.
 //!
 //! `libchdb.so` / `.dylib` gate their exports at link time, with a version
 //! script on Linux and an exported-symbols list on macOS, which is why dynamic
 //! linking never had this problem. An archive has no link step of its own: every
 //! symbol it carries reaches the final artifact and takes part in resolution. So
-//! for a static build the gate has to be compiled into the objects, and this
-//! asserts that it was — that the three bundled runtime targets have no
-//! externally visible definitions at all.
+//! for a static build the gate must be baked into the archive. Older archives
+//! hide the bundled runtime's definitions, while newer ones prelink and localize
+//! them on Linux or rename them on macOS.
 //!
 //! Two reasons to check the cause and not only the effect. It holds on Linux,
 //! where the visibility can regress with no symptom to observe, so no runtime
@@ -20,9 +20,8 @@
 //! over its own build output (`chdb/build/check_static_lib_hermetic.sh`, gate
 //! 2). That one guards the release; this one guards what actually arrived.
 //!
-//! Measured across the engines this crate has been pinned to: v26.7.0 shows
-//! 2365 externally visible definitions among 58 runtime members, and
-//! v26.7.2-rc.2 and v26.7.3 show 0 among the same 58.
+//! Measured across older engines: v26.7.0 shows 2365 externally visible
+//! definitions among 58 runtime members, and v26.7.2-rc.2/v26.7.3 show 0.
 
 mod common;
 
@@ -37,7 +36,7 @@ const RUNTIME_MEMBER_PREFIXES: [&str; 3] = ["libcxx__", "libcxxabi__", "libunwin
 const OFFENDERS_SHOWN: usize = 20;
 
 #[test]
-fn bundled_runtime_objects_export_nothing() {
+fn bundled_runtime_symbols_cannot_bind_host() {
     let artifact = Path::new(env!("CHDB_LINKED_ARTIFACT"));
     assert!(
         !artifact.as_os_str().is_empty(),
@@ -54,13 +53,33 @@ fn bundled_runtime_objects_export_nothing() {
     }
 
     let members = runtime_members(artifact);
-    // Zero would mean the naming changed and this test had quietly stopped
-    // testing anything, which is worse than a failure.
-    assert!(
-        !members.is_empty(),
-        "no {RUNTIME_MEMBER_PREFIXES:?} members among the archive's members: either the \
-         bundled runtime is no longer a separate set of objects, or chdb-core renamed them"
-    );
+    if members.is_empty() {
+        // New Linux archives prelink all objects into one member and localize
+        // everything outside the C API. The old per-runtime-member check no
+        // longer describes that layout; verify its exported surface here and
+        // exercise the host unwinder in runtime_faces.
+        let all_members = run("ar", &[OsStr::new("t"), artifact.as_os_str()], None);
+        let names: Vec<_> = all_members.lines().collect();
+        assert!(
+            cfg!(target_os = "linux")
+                && names.len() == 1
+                && names[0].starts_with("libchdb")
+                && names[0].ends_with(".o"),
+            "unknown static archive layout: {names:?}"
+        );
+        let (defined, visible) = symbols(&[artifact.to_path_buf()]);
+        assert!(defined > 0 && visible.iter().any(|name| name == "chdb_connect"));
+        assert!(
+            visible.len() <= 100
+                && !visible.iter().any(|name| {
+                    name.starts_with("_Unwind_")
+                        || name.starts_with("unw_")
+                        || name.starts_with("__cxa_")
+                }),
+            "prelinked archive exports unexpected runtime symbols: {visible:?}"
+        );
+        return;
+    }
 
     let scratch = common::tempdir();
     let objects = extract(artifact, &members, scratch.path());
@@ -91,15 +110,14 @@ fn bundled_runtime_objects_export_nothing() {
         }
     }
 
-    // Asserted at zero rather than "disjoint from this host's system runtime".
-    // These targets are built to have no externally visible definitions at all,
-    // and a symbol that merely happens to be absent from the running OS version
-    // would otherwise slip through.
+    // The old layout hides these definitions; the new macOS layout renames
+    // every one so a host can never resolve a runtime reference to it.
     assert!(
-        visible.is_empty(),
+        visible.is_empty()
+            || (cfg!(target_os = "macos") && visible.iter().all(|s| s.ends_with(".chdb"))),
         "{} of {} definitions in the bundled runtime are visible from outside the \
-         archive; a static build can then bind a call made inside the engine to \
-         another copy of the C++ runtime",
+         archive under an unrenamed name; a static build can then bind a host \
+         runtime call to the bundled C++ runtime",
         visible.len(),
         defined
     );

tests/runtime_faces.rs

@@ -166,6 +166,21 @@ fn cases() -> Vec<Case> {
         run: arrow_release_callback,
     });
 
+    // chdb-io/chdb-rust#53: an archive may keep the engine working while
+    // breaking the host Rust runtime's unwinder. Exercise the real consumer
+    // after a failing engine query, isolated so a SIGSEGV cannot hide the
+    // outcome of the other runtime cases.
+    #[cfg(feature = "static")]
+    if std::env::var_os("CHDB_CHECK_HOST_UNWIND").is_some() {
+        // The current pin predates chdb-core's fix. Opt in when validating a
+        // candidate archive, and enable by default only after adopting it.
+        cases.push(Case {
+            name: "host_backtrace_and_panic",
+            expect: "backtrace=ok|panic=ok\n",
+            run: host_backtrace_and_panic,
+        });
+    }
+
     cases
 }
 
@@ -489,6 +504,25 @@ fn signal_handlers_before_connect() {
     println!("ok");
 }
 
+#[cfg(feature = "static")]
+fn host_backtrace_and_panic() {
+    chdb_rust::runtime::signal_handlers(false);
+    let conn = Connection::open_in_memory().expect("open");
+    assert!(conn
+        .query("SELECT * FROM chdb_rust_missing_table", OutputFormat::CSV)
+        .is_err());
+
+    let trace = std::backtrace::Backtrace::force_capture();
+    assert_eq!(trace.status(), std::backtrace::BacktraceStatus::Captured);
+    assert!(!trace.to_string().trim().is_empty());
+    print!("backtrace=ok|");
+
+    std::panic::set_hook(Box::new(|_| {}));
+    assert!(std::panic::catch_unwind(|| panic!("expected panic")).is_err());
+    drop(conn);
+    println!("panic=ok");
+}
+
 /// The deadly-signal disposition for `sig`, read via `sigaction` with a null
 /// `act` — a query, not a change.
 fn signal_disposition(sig: libc::c_int) -> libc::sighandler_t {