installation of AmberTools v20 is broken due to switch to CMake in custom easyblock for Amber

[boegel]

In easybuilders/easybuild-easyblocks#2445 we updated the Amber easyblock to use CMake, but we failed to take into account that that same easyblock is also used for AmberTools.

Problems with AmberTools-20-intel-2020a-Python-3.8.2.eb:

• the runtime patching done on AmberTools/src/cpptraj/configure is failing, because the relative path is now run (building is done from a separate build dir due to the switch to CMakeMake), can be fixed by using %(builddir)s/AmberTools/src/cpptraj/configure (but not sure if that patching is still needed when using CMake)
• CMake is missing as a build dependency
• configuring with CMake fails with:

CMake Error at cmake/LibraryTracking.cmake:174 (message):
  Incorrect usage.  At least one LIBRARY should be provided.
Call Stack (most recent call first):
  cmake/MPIConfig.cmake:122 (import_libraries)
  CMakeLists.txt:118 (include)

Easy fix may be to tweak the Amber easyblock a bit more to stick to the old approach for AmberTools?

@branfosj @akesandgren Thoughts?

[branfosj]

I'd prefer switching to cmake for ambertools as well.

Adding https://github.com/easybuilders/easybuild-easyconfigs/blob/develop/easybuild/easyconfigs/a/Amber/AmberTools-20_fix_missing_MPI_LIBRARY_error.patch should fix the cmake fail.

[akesandgren]

Yeah that patch is vital, and any other AmberTools patches from Amber should be added to the AmberTools build.
I do try to make sure that the patches in Amber are correctly named, i.e. AmberTools vs Amber.
So these should be part of AmberTools:

AmberTools-20_cmake-locate-netcdf.patch
AmberTools-20_fix_missing_MPI_LIBRARY_error.patch
AmberTools-20_fix_xblas_missing_make_dependency.patch

After that AmberTools 20 should work correctly using CMakeMake too.

[boegel]

Thanks, that patch helps.

Now I'm hitting this:

CMake Error at cmake/PythonBuildConfig.cmake:67 (message):
  Missing required Python packages: numpy scipy matplotlib.  Please install
  these and try again.  If you cannot install them, you may set BUILD_PYTHON
  to FALSE to skip building Python packages, or set DOWNLOAD_MINICONDA to
  TRUE to create a python environment automatically.

So the CMake-based installation is building more stuff, leading to more dependencies? Ugh...

[akesandgren]

See Amber for deps :-) the diff in deps between them is basically zero.
The reason being that Amber is just AmberTools + pmemd

[boegel]

OK, but why does AmberTools require more dependencies when building with CMake compared to without it?

I had to add SciPy-bundle and matplotlib, but now also FFTW (since the existing easyconfig for AmberTools is built with intel toolchain):

CMake Error: The following variables are used in this project, but they are set to NOTFOUND.
Please set them or make sure they are set and tested correctly in the CMake files:
FFTW_LIBRARY_SERIAL
    linked by target "cmTC_da004" in directory /tmp/easybuild_build/AmberTools/20/intel-2020a-Python-3.8.2/easybuild_obj/CMakeFiles/CMakeTmp

CMake Error at cmake/TryLinkLibrary.cmake:135 (try_compile):
  Failed to configure test project build system.
Call Stack (most recent call first):
  cmake/jedbrown/FindFFTW.cmake:207 (try_link_library)
  cmake/3rdPartyTools.cmake:240 (find_package)
  CMakeLists.txt:194 (include)

[akesandgren]

Because the list wasn't complete before?
Haven't done an intel build of Amber lately so I don't know the deps for that.

And the CMake based build is much better at searching and finding external deps.

[boegel]

Made some progress in #14028, but not finished yet, running into a problem with sander, not sure how to replace the -nosanderapi configuration option with an equivalent that CMake listens to...

Any input welcome!

[branfosj]

Possibly BUILD_SANDER_API. If not then I'll look further when I am next attached to a system with the Amber tools source code.