update SuiteSparse easyblock to only install SuiteSparse libraries with make install

[appolloford]

(created using eb --new-pr)

[boegel]

It looks like these changes are not backwards-compatible at all...

@appolloford Can you provide some more info on why these changes are required?

[appolloford]

This PR is trying to make the installation of SuiteSparse as close to normal ConfigureMake as possible and keep backwards-compatibility as much as possible. I have tested 3.6.0, 4.0.0, 4.5.0, 5.1.2, 5.7.1, 5.8.1, 5.13.0, and 7.1.0. Although the older versions (< v5.1.2) are not supported anymore (which needs metis-4.x), I still keep the related codes there for reference.

The current behaviour of this easyblock is (1) revise the UConfig or SuiteSparse_Config for the older versions (< 4.5.0) which does not support setting options from environment (this is almost the same as the original configuration step) (2) Set proper environment variables for newer versions (< 6.0.0) (3) Left easyconfig to handle the CMAKE_OPTIONS for the recent versions

A few major changes:

1. If METIS is not provided in the dependencies, it will compile its own metis lib. After v6.0.0, users are not allowed to use preinstalled METIS either. My understanding is that it does not install metis lib either but only use its object file when compiling some libs.
2. By using make; make install, only c libraries will be installed (ideally static libs for versions < 4.5.0 and shared libs for else). No object files or subfolders are copied to the installdir (e.g. KLU/Lib).
3. Since its matlab tools need be compiled with mex and need all dependencies in relative path (e.g. KLU/Lib). The related files are not copied to the installdir anymore.
4. I think suitesparse does not need mpi and remove the code looking for MPICC in the original code. Suitesparse easyconfigs could move to gfbf instead of foss in the future.

Some concerns:

1. I tested SuiteSparse-5.10.1-foss-2021a-METIS-5.1.0-CUDA-11.3.1.eb and it passed, but there is no other versions I can test.
2. Suitesparse starts to use CMake from v5 in GraphBLAS, but lacks of proper options to set BLAS/LAPACK/INSTALL_PREFIX. Since there is no such easyconfig file and they are old, I did not do more work to support them.

Details:

1. Before v4.5.0, suitesparse uses INSTALL_LIB and INSTALL_INCLUDE to control installdir and use METIS_PATH and METIS to know the path of metis lib
2. From v4.5.0 to v6.0.0, suitesparse use INSTALL to control installdir and GraphBLAS need its own CMAKE_INSTALL_PREFIX
3. From v4.5.0 to v6.0.0, metis path is set by MY_METIS_LIB and MY_METIS_INC

Please let me know if you find any mistakes. I may forget some details and write something wrong here.

[boegel]

@boegelbot please test @ generoso
EB_ARGS="SuiteSparse-5.7.1-foss-2020a-METIS-5.1.0.eb SuiteSparse-5.7.1-intel-2020a-METIS-5.1.0.eb SuiteSparse-5.8.1-foss-2020b-METIS-5.1.0.eb SuiteSparse-5.8.1-intel-2020b-METIS-5.1.0.eb SuiteSparse-5.10.1-foss-2021a-METIS-5.1.0.eb SuiteSparse-5.10.1-foss-2021b-METIS-5.1.0.eb SuiteSparse-5.10.1-intel-2021a-METIS-5.1.0.eb SuiteSparse-5.10.1-intel-2021b-METIS-5.1.0.eb SuiteSparse-5.13.0-foss-2022a-METIS-5.1.0.eb SuiteSparse-5.13.0-foss-2022b-METIS-5.1.0.eb --installpath /tmp/$USER/pr3004"

[boegelbot]

@boegel: Request for testing this PR well received on login1

PR test command 'EB_PR=3004 EB_ARGS="SuiteSparse-5.7.1-foss-2020a-METIS-5.1.0.eb SuiteSparse-5.7.1-intel-2020a-METIS-5.1.0.eb SuiteSparse-5.8.1-foss-2020b-METIS-5.1.0.eb SuiteSparse-5.8.1-intel-2020b-METIS-5.1.0.eb SuiteSparse-5.10.1-foss-2021a-METIS-5.1.0.eb SuiteSparse-5.10.1-foss-2021b-METIS-5.1.0.eb SuiteSparse-5.10.1-intel-2021a-METIS-5.1.0.eb SuiteSparse-5.10.1-intel-2021b-METIS-5.1.0.eb SuiteSparse-5.13.0-foss-2022a-METIS-5.1.0.eb SuiteSparse-5.13.0-foss-2022b-METIS-5.1.0.eb --installpath /tmp/$USER/pr3004" EB_CONTAINER= EB_REPO=easybuild-easyblocks /opt/software/slurm/bin/sbatch --job-name test_PR_3004 --ntasks=4 ~/boegelbot/eb_from_pr_upload_generoso.sh' executed!

• exit code: 0
• output:

Submitted batch job 11915

Test results coming soon (I hope)...

- notification for comment with ID 1756123421 processed

Message to humans: this is just bookkeeping information for me,
it is of no use to you (unless you think I have a bug, which I don't).

[boegelbot]

Test report by @boegelbot

Overview of tested easyconfigs (in order)

• SUCCESS SuiteSparse-5.7.1-foss-2020a-METIS-5.1.0.eb
• FAIL (build issue) SuiteSparse-5.7.1-intel-2020a-METIS-5.1.0.eb (partial log available at https://gist.github.com/boegelbot/6c49ebf1ee66b55f650f73f5d6a76f0d)
• SUCCESS SuiteSparse-5.8.1-foss-2020b-METIS-5.1.0.eb
• FAIL (build issue) SuiteSparse-5.8.1-intel-2020b-METIS-5.1.0.eb (partial log available at https://gist.github.com/boegelbot/345768cdb9777e0362882613b72b558a)
• SUCCESS SuiteSparse-5.10.1-foss-2021a-METIS-5.1.0.eb
• SUCCESS SuiteSparse-5.10.1-foss-2021b-METIS-5.1.0.eb
• SUCCESS SuiteSparse-5.10.1-intel-2021a-METIS-5.1.0.eb
• SUCCESS SuiteSparse-5.10.1-intel-2021b-METIS-5.1.0.eb
• SUCCESS SuiteSparse-5.13.0-foss-2022a-METIS-5.1.0.eb
• SUCCESS SuiteSparse-5.13.0-foss-2022b-METIS-5.1.0.eb

Build succeeded for 8 out of 10 (10 easyconfigs in total)
cns3 - Linux Rocky Linux 8.5, x86_64, Intel(R) Xeon(R) CPU E5-2667 v3 @ 3.20GHz (haswell), Python 3.6.8
See https://gist.github.com/boegelbot/a5c137fe87b240a59528842550b5aa1e for a full test report.

[appolloford]

These two tests failed for intel license.

• FAIL (build issue) SuiteSparse-5.7.1-intel-2020a-METIS-5.1.0.eb (partial log available at https://gist.github.com/boegelbot/6c49ebf1ee66b55f650f73f5d6a76f0d)
• FAIL (build issue) SuiteSparse-5.8.1-intel-2020b-METIS-5.1.0.eb (partial log available at https://gist.github.com/boegelbot/345768cdb9777e0362882613b72b558a)

Error: A license for Comp-CL is not available now (-15,570,115).

[akesandgren]

Test report by @akesandgren

Overview of tested easyconfigs (in order)

• SUCCESS SuiteSparse-5.7.1-intel-2020a-METIS-5.1.0.eb
• SUCCESS SuiteSparse-5.8.1-intel-2020b-METIS-5.1.0.eb

Build succeeded for 2 out of 2 (2 easyconfigs in total)
b-an02.hpc2n.umu.se - Linux Ubuntu 20.04, x86_64, Intel(R) Xeon(R) CPU E5-2690 v4 @ 2.60GHz, Python 3.8.10
See https://gist.github.com/akesandgren/23e356e8f9553d3735859db5bb4e0321 for a full test report.

[akesandgren]

Also since SuiteSparse uses CMake from v6 we should take care of that in the easyblock.
It won't be the first time we do CMakeMake/ConfigureMake compatible easyblocks.

This can however be done in a separate PR

[appolloford]

Actually SuiteSparse does not follow the normal cmake; make; make install steps. It uses CMake in several subfolders (I am not sure whether all subfolders have been moved to cmake projects) but the top level is still managed by a makefile. The official guide says the whole suite should be built in this command:

CMAKE_OPTIONS="-DNPARTITION=1 -DNSTATIC=1 -DCMAKE_BUILD_TYPE=Debug" make

I think this does not match with the CMakeMake easyblock and I leave the CMAKE_OPTIONS to be controlled by each easyconfig file, like

easybuilders/easybuild-easyconfigs#18857

[akesandgren]

According to the python-linting test you should also remove these imports
shutil, stat, easybuild.tools.filetools.adjust_permissions, easybuild.tools.filetools.copy_dir
since they aren't used

[akesandgren]

LGTM

[akesandgren]

Going in, thanks @appolloford!