Tests SIGSEGV about lambrex

[otbrown]

Hi Rupert,

I'm not entirely surprised -- I've not tested with MPI turned on at all yet. This should work if you build AMReX with ENABLE_MPI=OFF, but it'll be useful to work out why this occurs regardless. Might provide some useful insight ahead of actually turning on MPI!

[otbrown]

Haven't checked yet, but this is almost certainly a consequence of the way Catch runs Test Cases. It reruns everything in TEST_CASE scope but each SECTION only once, currently amrex::Initialize and amrex::Finalize are called in TEST_CASE scope. Repeat calls to both are fine in serial but follow MPI initialization rules when MPI is included in AMReX.

Possible solutions:

• Move Initialize and Finalize to their own sections - as long as Catch always respects the order of sections (and I'm sure it does) this should be fine.
• Change to more traditional class based test fixtures with setup and teardown, which Catch also supports.

I'm inclined to solution 1 for its simplicity, will implement and test ASAP.

[otbrown]

Rupert pointed out to me that creating Initialize and Finalize sections required one TEST_CASE per executable. Luckily Catch provides a way of doing test run set-up and teardown through event listeners. Amrex (and MPI) are now initialised and finalised at the beginning and end of a test run. Will push and close.

Closed by 6599d81 and f82a0e0.