On Wed, 2017-07-19 at 15:58 +0200, Oliver Kowalke via Boost wrote:
Am 19.07.2017 3:12 nachm. schrieb "Peter Dimov via Boost" < boost@lists.boost.org>:
Sergei Nikulov wrote:
JFYI, same commands used on every system
1. generate> cmake .. 2. build> cmake --build . 3. test> ctest
Well not entirely, because for MS backends you can pick --config Release at build time and for the rest you need to pick -DCMAKE_BUILD_TYPE at configure time.
Not to mention the address-model, choosing which MSBuild supports at build time, but CMake not only doesn't, you have to use an entirely different generator for 64 bit.
What does this mean for boost's regression test? Especially for the fiber library I need a combination of: - address-model (32/64) - architecture (arm, arm64, mips, ....) - binary format (elf, ms, o32, mach-o, ...) - ABI (sysv, aapcs, ...) - implementation (assembler, ucontext, windows fibers) - segmented stacks (on/off) - valgrind (on/off) - transactional memory (on/off) ... Do I need to provide for each combination a different generator?
You can write a ctest script to iterate over each combination for testing. Its just a for loop over each one. Or a cmake script that can be written do it as well. Since its commmon with boost to test over many combinations, contributing a script for boost developers to use would be helpful.