snappy 1.2.2, a transitive dependency of arrow, includes <immintrin.h> only
when SNAPPY_HAVE_BMI2 or SNAPPY_HAVE_X86_CRC32 is set, but MemCopy64() selects
an AVX code path on a separate condition, `defined(__x86_64__) &&
defined(__AVX__)`. The vcpkg port passes -DSNAPPY_HAVE_BMI2=OFF and
-DSNAPPY_HAVE_X86_CRC32=OFF to keep the binaries portable, so on a toolchain
whose default target enables AVX that branch is compiled with no intrinsics
header and snappy.cc fails with "'__m256i' was not declared in this scope".
This broke the clang-tidy and clazy-report workflows the first night after they
moved to ubuntu-26.04. The runner switch also changed the vcpkg compiler hash,
so the binary cache went from "Restored 103 package(s)" to "Restored 0
package(s)" and every port was rebuilt from source, which is what exposed
snappy in the first place. The bug is not GCC 15 specific: microsoft/vcpkg#51047
reports it against GCC 14.3.1. It is still present in snappy 1.3.1 and main, and
vcpkg master still ships 1.2.2 with the SNAPPY_HAVE_* options off, so neither a
baseline bump nor a port update fixes it.
Vendor the port under ThirdParty/vcpkg-overlay-ports and add
avx-immintrin.patch, which includes <immintrin.h> for the __AVX__ branch as
well. The patch cannot change the ISA baseline of the result: it only supplies
a header on a path the compiler was already going to take. Verified by
reproducing the failure locally with an overlay triplet setting
-march=x86-64-v3, confirming the patched port builds both with and without that
flag, and checking that objdump finds no AVX instructions in the library built
with the stock toolchain.
Reference the two overlay ports by their individual directories rather than by
the parent directory. The two configurations resolve different versions of both
ports (default: snappy 1.2.2#1, thrift 0.22.0; rhel8: snappy 1.2.1,
thrift 0.20.0#1), so an overlay of the whole directory would downgrade thrift
for the default builds and bump snappy for RHEL8. Dry runs of both manifests
confirm each profile now picks up only its own overlay.
The bumped port-version misses the existing vcpkg binary caches, but those
caches are already cold for the ubuntu-26.04 workflows.
* Refactor: add MACOSX_BUNDLE option to qt_add_executable in multiple CMakeLists.txt files
* #14045 macOS: Guard std::stacktrace usage and skip ucontext on Apple platforms
* #14045 macOS: Guard std::ispanstream with std::istringstream fallback
* #14045 macOS: Add vcpkg overlay triplets for osx and exclude grpc on osx
* #14045 macOS: CMake macOS build support and openzgy OpenMP-disable patch
* #14045 CI: Add macOS build workflow
* #14045 macOS: Use feature-test macros to detect <stacktrace> and <spanstream>
* #14045 macOS: Make MAX_STIMPLAN_LAYERS constexpr to fix arm64 link error
* #14045 CI: Ship README-macOS.txt with artifact to document Gatekeeper workaround
* #14045 CI: Pin macOS runner to macos-14 and set deployment target to 14.0
* #14045 CI: Move macOS runner to macos-15 and fix README step on configure failure
* Update vcpkg to 2024.02.14 Release
A blank install of vcpkg did not work anymore due to changes in upstream repos. Had to update vcpkg.exe to latest to ensure download of libraries works as expected.
Had to update baseline to 2024.02.14 to be able to compile on VS 17.12. A bug related to naming of boost was fixed in boost 1.84
https://github.com/microsoft/vcpkg/issues/38980
* Bump to a more recent vcpkg due to a build error in arrow
https://github.com/apache/arrow/issues/42027
* Avoid fmt 11 due to a bug for MSVC
https://github.com/gabime/spdlog/issues/3251#issue-2648376977
* Remove obsolete GRPC config in top-level CMake
* Avoid pinning of Protobuf
Magne Sjaastad