[boost] Fix test package on macOS when *:shared=True #16348
Conversation
Signed-off-by: Uilian Ries <uilianries@gmail.com>
ghost
mentioned this pull request
|
I detected other pull requests that are modifying boost/all recipe:
This message is automatically generated by https://github.com/ericLemanissier/conan-center-conflicting-prs so don't hesitate to report issues/improvements there. |
The problem I suspect is that system installed ctest depends on system installed libarchive which depends on macOS libiconv which is slightly different than GNU libiconv. Since libiconv is a dependency of boost (at least default configuration), runtime loader tries to find symbols of libiconv in conan package (because VirtualRunEnv prepends lib paths of conan package into DYLD_LIBRARY_PATH, which is expected). Maybe we should always consider libiconv as a system lib on macOS? (or maybe not, not sure what's the best approach) https://stackoverflow.com/questions/57734434/libiconv-or-iconv-undefined-symbol-on-mac-osx |
This comment has been minimized.
This comment has been minimized.
| self.run(f"ctest --output-on-failure -C {self.settings.build_type}", env="conanrun") | ||
| if can_run(self): | ||
| cmake = CMake(self) | ||
| cmake.test() |
There was a problem hiding this comment.
bear in mind that one of the reasons we used ctest rather than cmake.test() is to ensure that in the event of the tools.build:skip_test config being abled, this test still runs unconditionally, as cmake.test() generally applies to library tests rather than executables in the test package
|
Bear in mind that the runtime dependencies of the
The version of I think the biggest issue here is that we are invoking Thanks @uilianries and @SpaceIm - at least we know the cause now :) |
|
New error, nothing related to the latest, Window long path on test package: |
It's an internal bug, it should be fixed soon. |
This reverts commit 3e3f74a.
Signed-off-by: Uilian Ries <uilianries@gmail.com>
This comment has been minimized.
This comment has been minimized.
|
I think the "fix" was actually a bit easier - the test executables, as generated by CMake, have embedded RPATH entries pointing to the conan-provided lib directories (libiconv, etc) - so on platforms where CMake does this, we don't strictly need to be in the conanrun context. On Windows, on the other hand, if |
This comment has been minimized.
This comment has been minimized.
It may not work on Linux because RPATH in executable is not transitive. |
This comment has been minimized.
This comment has been minimized.
Conan v1 pipeline ✔️All green in build 6 (
Conan v2 pipeline (informative, not required for merge) ✔️
All green in build 5 (
|
jcar87
changed the title
|
Boost is working with both Conan v1 and v2 🚀 |
Specify library name and version: boost/1.79.0
Some runtime error occurred when trying to run Boost test package on the CI:
https://c3i.jfrog.io/c3i/misc-v2/logs/prod-v2/TapaholesRepo/106-configs/macos-clang/boost/1.79.0//ff15167be75d82eb8c4445406fa2bff3f15f4cdb-test.txt
However, I can't reproduce that error, instead, it works for me, but I only have Clang Apple 14 installed. So my intention here is trying to use cmake build --target test, instead of invoking ctest directly, my guess is something related to DYLIB missing path.
My build log: