Conversation
remove hardcoded comment refs
…) and don't bump lixbml2 yet, need to port code
[bot] New upstream conan recipes
…_BINARY_FUNCTION / Wno-enum-conversion
…center-index#29907 vendored to nrel-v2 remote...
jmarrec
added
Developer Issue
Pull Request - Ready for CI
🧪 Test Results DashboardSummary
|
| Run | XML File | Status |
|---|---|---|
| run1 | results.xml |
✅ Found |
| run3 | results.xml |
✅ Found |
| run2 | results.xml |
✅ Found |
…. Didn't clean options passed, most of them useless
… workflows are using that antiquated way to configure (directly, vs via cmake --preset)
prompted by, for context: boost upgrade from 1.79.0 to 1.86.0 seem to have fixed a bug/changed the behavior of boost::filesystem::canonical which now will actually resolve it to the /mnt/build and not /__w/build, hence a few WorkflowJSON tests failing because oswPath() uses canonical, and resourcesPath() is /__w/build, from CMake --- This looks pwned from something similar https://github.com/marketplace/actions/maximize-build-disk-space#how-it-works I see you’re creating a 8GB swapfile on it, which at least seem to make sense. The Prepare workspace step I’m quite puzzled by it. https://github.com/NatLabRockies/OpenStudio/blob/8008ef767fdc0f9d3dd3fabd383da15d009aef76/.github/workflows/full-build.yml#L229-L243 If I follow, the step by step for prepare_dir “/__w/OpenStudio/OpenStudio/build” “/mnt/build”: 1. Create /mnt/build if it doesn’t exist 2. If /__w/.../build already exists as a real directory (not already a symlink), copy its contents to /mnt/build and delete it 3. Create the symlink: /__w/.../build → /mnt/build result: anything that writes to /__w/.../build is transparently redirected to /mnt/build. CMake thinks it’s building in /__w/.../build but the files actually land on /mnt. I don’t think that’s what we should be doing
|
The RemoteBCL_BCLSearchResult is failing, and it is not due to this branch, I think this is a bug introduced while updating the BCL to use NLR instead of NREL. I filed it at |
…ll platforms from conan_recipes in GHA
|
I uploaded bumped deps to the nrel-v2 artifactory from github actions from commit d91c69e https://github.com/jmarrec/conan-recipes/actions/runs/25434976827 And redid windows in https://github.com/jmarrec/conan-recipes/actions/runs/25443826236/job/74642271487 I also had to manually rebuild swig on centos-9 because I was hitting GLIBC / GLIBCXX ABI problems, it seems it was built on Ubuntu-24.04 originally so it wouldn't work on either ubuntu-22.04 nor centos-9. I rebuilt on Centos-9 because it has GLIBC 2.34 while Ubuntu has GLIBC 2.35, so older = better |
|
I know how to fix this, just waiting for centos to finish before I push |
…so they use _WIN32_WINNT >= 0x0602 `#define BOOST_WINAPI_VERSION_WIN8 0x0602` https://www.boost.org/doc/libs/1_86_0/boost/winapi/config.hpp ``` dumpbin /symbols "C:\Users\julien\.conan2\p\boostebc618dba587a\p\lib\libboost_log.lib" | Select-String "v2s_mt" | Select-Object -First 1 39B 00000000 UNDEF notype () External | ?throw_@system_error@v2s_mt_nt62@log@boost@@SAXPEBD_K0H@Z (public: static void __cdecl boost::log::v2s_mt_nt62::system_error::throw_(char const *,unsigned __int64,char const *,int)) ``` Fixed #5606 (comment)
|
There's a workflow issue: https://github.com/NatLabRockies/OpenStudio/actions/runs/25445076680/job/74646824192?pr=5606#step:18:8419 It's rerunning the wrong test... I think the bug is that Here is a demo: (py312)(3.2.2)julien@OS-build-release3$ ctest --show-only | rg RemoteBCL_BCLSearchResult
Test #678: BCLFixture.RemoteBCL_BCLSearchResult
(py312)(3.2.2)julien@OS-build-release3$ ctest -E 'IddFixture' --show-only | rg RemoteBCL_BCLSearchResult
Test #640: BCLFixture.RemoteBCL_BCLSearchResult |
|
I still have a windows issue locally when building the CLI (openstudiolib builds fine) I think I tracked down the root cause:
if(WIN32)
add_compile_options(/FI winnls.h) # force-include so LCTYPE is always available
endif()
|
CLI/App.hpp must be included after ScriptEngine.hpp so that windows.h (and winnls.h) is pulled in before CLI11's windef.h, avoiding a missing LCTYPE in boost/regex's w32_regex_traits.hpp. see #5606 (comment) I also forward declare the CLI::App class.
|
Everything appears to be building. The 22.04 arm64 is 1) a fluke probably and 2) not a huge concern. This is blocking a lot of things, so I'm going to merge this as-is. The full-build.yml workflow needs to be completely revamped IMHO anyways, so it's fine if If we're fixing any issues later on develop. The ability to build with an up-to-date mac trumps it by far. |
1 participant
Pull request overview
I just upgraded to Tahoe 26.4 and got XCode 26.4, which bumped Apple Clang from 17 to 21 (quite a jump).
-Wno-enum-constexpr-conversion no longer exists, so we can't build with boost 1.79 even with workarounds anymore
I therefore bumped all deps quite aggressively, then reverted a few:
Assuming CI (after building some packages) is still happy, we would vendor the pinned recipes and packages for all platforms we build, as I have historically done. (I have a helper workflow to do so for github actions machines at https://github.com/jmarrec/conan-recipes/blob/main/.github/workflows/upload_deps.yml, I usually also build some niche os/compiler combos myself, the centos stuff)
Pull Request Author
src/model/test)src/energyplus/Test)src/osversion/VersionTranslator.cpp)Labels:
IDDChangeAPIChangePull Request - Ready for CIso that CI builds your PRReview Checklist
This will not be exhaustively relevant to every PR.