More changes for conda packaging

[robertodr]

Changes motivated by packaging the library for Anaconda

[robertodr]

I can actually get a working MRCPP conda package:

conda create -n foo mrcpp=*=mpi* cmake -c nostress -c conda-forge --yes

I am able to compile the poisson example. Running works almost fine, I get:

and have been unable to solve it...

[stigrj]

Hmm, we might have a bug. Have you tried other examples?

[MinazoBot]

	2 Warnings
⚠️	There are code changes, but no corresponding tests. Please include tests if this PR introduces any modifications in behavior.
⚠️	Consider adding supporting documentation to this change. Documentation sources can be found in the doc directory.

Generated by 🚫 Danger

[codecov]

Codecov Report

Merging #112 into master will increase coverage by 0.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master     #112      +/-   ##
==========================================
+ Coverage   72.22%   72.24%   +0.01%     
==========================================
  Files         139      139              
  Lines        5955     5959       +4     
==========================================
+ Hits         4301     4305       +4     
  Misses       1654     1654

Impacted Files	Coverage Δ
src/core/MWFilter.h	100% <ø> (ø)	⬆️
src/core/CrossCorrelation.h	100% <ø> (ø)	⬆️
src/treebuilders/PHCalculator.cpp	100% <100%> (ø)	⬆️
src/core/CrossCorrelation.cpp	71.42% <100%> (+0.59%)	⬆️
src/core/MWFilter.cpp	50.84% <100%> (+0.42%)	⬆️
src/treebuilders/BSCalculator.cpp	98.5% <100%> (+0.02%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update dc2ec19...db4e10e. Read the comment docs.

[robertodr]

Nope, I was lazy. I'll try all the examples and see. Note that we have two variants of the package: OpenMP and OpenMP+MPI and it happens with both versions. I haven't understood how to make a serial version and haven't really investigated (as I thought it's rather pointless for distribution)

[robertodr]

conda-build compiles with -fstack-protector-strong which I think makes the (probable) bug appear.

[robertodr]

Update

The package still needs a little bit more work. Hold on merging!

• When compiling mpi_shared_tree.cpp I get:

CMakeFiles/mpi_shared_tree.dir/mpi_shared_tree.cpp.o: In function `main':
mpi_shared_tree.cpp:(.text+0x352): undefined reference to `mrcpp::SharedMemory::SharedMemory(int, int)'
mpi_shared_tree.cpp:(.text+0x40e): undefined reference to `void mrcpp::share_tree<3>(mrcpp::FunctionTree<3>&, int, int, int)'
mpi_shared_tree.cpp:(.text+0x648): undefined reference to `void mrcpp::share_tree<3>(mrcpp::FunctionTree<3>&, int, int, int)'

UPDATE I think this was due to the MRCPP::mrcpp target not exporting its dependency on MPI properly. I am now testing a fix.

• When compiling mpi_send_tree.cpp I get:

CMakeFiles/mpi_send_tree.dir/mpi_send_tree.cpp.o: In function `main':
mpi_send_tree.cpp:(.text+0x5d8): undefined reference to `void mrcpp::send_tree<3>(mrcpp::FunctionTree<3>&, int, int, int, int)'
mpi_send_tree.cpp:(.text+0x60d): undefined reference to `void mrcpp::recv_tree<3>(mrcpp::FunctionTree<3>&, int, int, int, int)'

UPDATE I think this was due to the MRCPP::mrcpp target not exporting its dependency on MPI properly. I am now testing a fix.

• When compiling mpi_matrix.cpp I get:

CMakeFiles/mpi_matrix.dir/mpi_matrix.cpp.o: In function `main':
mpi_matrix.cpp:(.text+0x570): undefined reference to `void mrcpp::send_tree<3>(mrcpp::FunctionTree<3>&, int, int, int, int)'
mpi_matrix.cpp:(.text+0x5a5): undefined reference to `void mrcpp::recv_tree<3>(mrcpp::FunctionTree<3>&, int, int, int, int)'

UPDATE I think this was due to the MRCPP::mrcpp target not exporting its dependency on MPI properly. I am now testing a fix.

Of the compiled examples, all but scf smash the stack. scf gives:

============================================================
                        Running SCF
------------------------------------------------------------
 Iter      E_np1          dE_n      ||phi_np1||   ||dPhi_n||
------------------------------------------------------------
  1  -3.5389437389e-01  1.5e-01   1.0341694208e+00  3.7e-01
free(): invalid pointer
Aborted (core dumped)

[stigrj]

I'm not able to reproduce the stack error on my laptop. Which compiler is used on conda?

[robertodr]

GNU 7.3.0 I wasn't able to reproduce either :(

…

On Tue, Dec 3, 2019 at 4:46 PM Stig Rune Jensen ***@***.***> wrote: I'm not able to reproduce the stack error on my laptop. Which compiler is used on conda? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#112?email_source=notifications&email_token=AA4JOEOSWFTMP3VEVDWVJF3QWZ5M5A5CNFSM4JUYSU3KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEFZ2KAA#issuecomment-561227008>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AA4JOEPMTBVD47HLRJ6ZACDQWZ5M5ANCNFSM4JUYSU3A> .

-- Roberto Di Remigio

[robertodr]

This is the full list of compiler flags used with conda build:

[1/91] $BUILD_PREFIX/bin/x86_64-conda_cos6-linux-gnu-c++  -DHAVE_OPENMP -Dmrcpp_EXPORTS -I../src -Iinclude -isystem $PREFIX/include/eigen3 -fvisibility-inlines-hidden -std=c++17 -fmessage-length=0 -march=nocona -mtune=haswell -ftree-vectorize -fPIC -fstack-protector-strong -fno-plt -O2 -ffunction-sections -pipe -isystem $PREFIX/include -fdebug-prefix-map=$SRC_DIR=/usr/local/src/conda/mrcpp-1.0.2 -fdebug-prefix-map=$PREFIX=/usr/local/src/conda-prefix -O3 -DNDEBUG -fPIC   -fopenmp -std=c++14 -MD -MT src/CMakeFiles/mrcpp.dir/core/CrossCorrelationCache.cpp.o -MF src/CMakeFiles/mrcpp.dir/core/CrossCorrelationCache.cpp.o.d -o src/CMakeFiles/mrcpp.dir/core/CrossCorrelationCache.cpp.o -c ../src/core/CrossCorrelationCache.cpp

[robertodr]

I think the stack smashing was a result of not correctly exporting the OpenMP dependency in the MRCPP::mrcpp CMake target.

[stigrj]

So, what should really be in the VERSION file at this point?

[robertodr]

I assume we only want to package tagged releases, so it should be the release number.

[stigrj]

But this is not a tagged release