github-actions[bot] commented on code in PR #68562:
URL: https://github.com/apache/doris/pull/68562#discussion_r4122379903
##########
be/cmake/thirdparty.cmake:
##########
@@ -120,6 +120,13 @@ add_thirdparty(lance_c LIB64 NOTADD)
if (NOT GLIBC_COMPATIBILITY)
list(APPEND COMMON_THIRDPARTY m lance_c)
endif()
+# libpaimon_c.a (built from paimon-rust) brings in Rust compiler_builtins that
+# would otherwise steal libm symbols. Place libm before it so the final linker
+# resolves C math symbols from the system library first.
+add_thirdparty(paimon_c LIB64 NOTADD)
Review Comment:
[P1] Make the standard BE build refresh existing third-party installs.
Adding this imported target makes `libpaimon_c.a` mandatory for both link
modes, but `build.sh:482-499` skips the third-party build whenever the old
brotli/libhdfs sentinel exists. A checkout with a valid pre-PR installed tree
therefore has no Paimon archive and `./build.sh --be` fails when CMake uses
this target. Please include the new archive in the installation gate and
rebuild Lance and Paimon with the same Rust toolchain when needed.
##########
thirdparty/build-thirdparty.sh:
##########
@@ -2447,6 +2453,131 @@ build_lance_c() {
fi
}
+build_paimon_rust() {
+ check_if_source_exist "${PAIMON_RUST_SOURCE}"
+ cd "${TP_SOURCE_DIR}/${PAIMON_RUST_SOURCE}"
+
+ rm -rf "${BUILD_DIR}"
+ mkdir -p "${BUILD_DIR}"
+
+ local cargo_bin="${PAIMON_RUST_CARGO:-${CARGO:-cargo}}"
+ if ! command -v "${cargo_bin}" >/dev/null 2>&1; then
+ echo "cargo is required to build paimon-rust. Install Rust 1.94.0 or
set PAIMON_RUST_CARGO."
+ exit 1
+ fi
+
+ local required_rust_version="1.94.0"
+ local cargo_env=(
+ "CARGO_BUILD_JOBS=${PARALLEL}"
+ "CARGO_TARGET_DIR=${PWD}/${BUILD_DIR}"
+ )
+ if command -v rustup >/dev/null 2>&1 && [[ -z "${RUSTUP_TOOLCHAIN}" ]];
then
+ # The presence check must look for the toolchain the minimum actually
+ # requires, not a literal: with only an older toolchain installed the
+ # stale check would skip the install below and then force
+ # RUSTUP_TOOLCHAIN to a version rustup cannot dispatch, failing the
+ # build before any archive is produced.
+ local required_rust_regex="${required_rust_version//./\\.}"
+ if ! rustup toolchain list | grep -Eq
"^${required_rust_regex}([[:space:]-]|$)"; then
+ rustup toolchain install "${required_rust_version}" --profile
minimal
+ fi
+ cargo_env+=("RUSTUP_TOOLCHAIN=${required_rust_version}")
+ fi
+
+ local cargo_version
+ if ! cargo_version="$(env "${cargo_env[@]}" "${cargo_bin}" --version | awk
'{print $2}')"; then
+ echo "failed to get cargo version for paimon-rust. Install Rust
${required_rust_version} or set PAIMON_RUST_CARGO/RUSTUP_TOOLCHAIN."
+ exit 1
+ fi
+ # Rust 1.94.0 is the minimum supported version. Allow newer toolchains when
+ # callers explicitly select one or rustup is unavailable on the system.
+ # NOTE: paimon_c and lance_c are both Rust staticlibs linked into the same
+ # BE binary; they must be built with the SAME rustc toolchain so the linker
+ # resolves both crates' std references against a single std copy. Mixing
+ # toolchains makes the precompiled std hashes differ and the linker pulls
+ # both std copies in, colliding on the unmangled `rust_eh_personality`
+ # (duplicate symbol). Rebuild both lance_c and paimon_rust whenever the
+ # toolchain changes, using the same RUSTUP_TOOLCHAIN for both builds.
+ if ! awk -v required="${required_rust_version}" -v
actual="${cargo_version}" 'BEGIN {
+ split(required, r, ".");
+ split(actual, a, ".");
+ for (i = 1; i <= 3; i++) {
+ if ((a[i] + 0) > (r[i] + 0)) {
+ exit 0;
+ }
+ if ((a[i] + 0) < (r[i] + 0)) {
+ exit 1;
+ }
+ }
+ exit 0;
+ }'; then
+ echo "paimon-rust requires Rust/Cargo ${required_rust_version} or
newer, but found ${cargo_version}."
+ echo "Install Rust ${required_rust_version} or set
PAIMON_RUST_CARGO/RUSTUP_TOOLCHAIN."
+ exit 1
+ fi
+
+ if [[ "${KERNEL}" != 'Darwin' ]]; then
+ cargo_env+=("CFLAGS=${CFLAGS:-} -std=gnu17")
+ fi
+
+ local cargo_args=(build --release --locked -p paimon-c --features
paimon/storage-hdfs)
+ if [[ "$(echo "${PAIMON_RUST_CARGO_OFFLINE}" | tr '[:lower:]'
'[:upper:]')" == "ON" ]]; then
+ cargo_args+=(--offline)
+ fi
+ env "${cargo_env[@]}" "${cargo_bin}" "${cargo_args[@]}"
+
+ # Generate the C header from the Rust extern "C" surface via cbindgen.
+ # cbindgen is a pinned, Doris-controlled input: an unpinned "current"
+ # release would regenerate paimon.h differently between builds. The
+ # pinned version installs under a Doris-controlled --root and its
+ # resolved absolute path is invoked directly (a custom CARGO_HOME does
+ # not necessarily put cargo-installed binaries on PATH). Offline builds
+ # pass --offline to the install command, exactly like the fetch/build
+ # handling above — cargo fails on a missing local crate cache instead
+ # of reaching for the network.
+ local cbindgen_version="0.29.4"
+ local cbindgen_bin="${PAIMON_RUST_CBINDGEN:-}"
+ if [[ -z "${cbindgen_bin}" ]]; then
+ local
cbindgen_root="${TP_SOURCE_DIR}/.doris-cbindgen-${cbindgen_version}"
+ cbindgen_bin="${cbindgen_root}/bin/cbindgen"
+ if [[ ! -x "${cbindgen_bin}" ]]; then
+ local cbindgen_install_args=(install cbindgen
+ --version "${cbindgen_version}" --locked --root
"${cbindgen_root}")
+ if [[ "$(echo "${PAIMON_RUST_CARGO_OFFLINE}" | tr '[:lower:]'
'[:upper:]')" == "ON" ]]; then
+ cbindgen_install_args+=(--offline)
+ fi
+ echo "cbindgen not found; installing pinned ${cbindgen_version}
via cargo install ..."
+ env "${cargo_env[@]}" "${cargo_bin}" "${cbindgen_install_args[@]}"
+ fi
+ elif [[ ! -x "${cbindgen_bin}" ]]; then
+ echo "PAIMON_RUST_CBINDGEN=${cbindgen_bin} is not an executable file."
+ exit 1
+ fi
+ # Write a temporary cbindgen.toml so the generated header carries our
+ # include-guard / cpp-compat settings without touching the upstream tree.
+ local cbindgen_toml="${BUILD_DIR}/cbindgen.toml"
+ mkdir -p "${BUILD_DIR}"
+ cat >"${cbindgen_toml}" <<'EOF'
+language = "C"
+include_guard = "PAIMON_C_H"
+pragma_once = true
Review Comment:
[P2] Carry the selected Cargo and offline mode into header generation. For a
crate directory, [cbindgen 0.29.4 runs `cargo
metadata`](https://github.com/mozilla/cbindgen/blob/v0.29.4/src/bindgen/cargo/cargo_metadata.rs#L216-L226)
using `$CARGO` or `cargo` from `PATH`. This invocation supplies neither
`CARGO=${cargo_bin}` nor `CARGO_NET_OFFLINE`, so
`PAIMON_RUST_CARGO=/path/to/cargo` can complete the archive build and cbindgen
install, then fail when `cargo` is absent from `PATH`.
`PAIMON_RUST_CARGO_OFFLINE=ON` can also let metadata fetch uncached
dependencies. Pass the selected executable and ensure this metadata step can
complete without network access in offline mode.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
