Closed miroi closed 8 years ago
Can we do something about it @robertodr? We could have a minimum test set and an extended test set because it really takes long at the moment.
I guess it's so slow because it downloads the Boost archive. A possible solution would be to have a copy of the archive available on the repo.
If it's just the downloading we could try to group the tests with a common fixture.
The problem is we download it 4 times. Do we actually compile it? Can the tests be collapsed into one? It is ultra slow. Also there is come CMake code for BUILD_CUSTOM_BOOST which is never used. Also there is some code complication of the chrono stuff. But also that is not really used inside the tests so could be removed.
They can definitely be collapsed into one. I had honestly not thought that it could be such a showstopper ;) Can you point me to the unused BUILD_CUSTOM_BOOST
code? Might be a leftover or a bug. I think it is enough to just test the Boost.MPI and Boost.Python libs, possibly together. They are the trickier to build. I would however leave the header-only test as an autonomous test.
I think it would be good to test Python and MPI together for the sake of total test time.
So perhaps we can have one boost
test and one boost_headers
test.
The leftovers are for instance here:
https://github.com/scisoft/autocmake/blob/master/test/boost_mpi_libs/src/CMakeLists.txt#L3-L5
OK. I'll take care of streamlining the tests on my fork and file a pull request. Those lines are not IMO a leftover, a build with -jN would fail without.
OK then I misunderstood the lines - sorry.
I'll add a comment in the Boost module documentation, to clarify it also for the users.
The test/test.py::test_boost_libs takes too long. Isn't there a way to make it faster ?