# Known Issues¶

• The submodule directories are empty. To clone the code including all submodules, use

git clone --recursive https://github.com/SeisSol/SeisSol.git


If you already cloned the repository without --recursive, you can use the following command to get the submodules.

git submodule update --init --recursive


## Installation¶

• The build process fails because of missing files. SeisSol requires some external libraries which are integrated via git submodules. For some of the libraries it is important to have the correct version (e.g. the XMDF Writer) because of changes in the interface of the library. If you clone the submodules with git, you should always get the correct version. However, downloading the SeisSol and its submodules directly from the Github homepage might result in an incorrect combination of versions.

• The parallel SCons build fails when Scalasca is enabled. Scorep can not compile two files in parallel if they have the same name even if they are in different subdirectories. As a workaround when building SeisSol with Scalasca, omit the -j option.

• The SCons configuration step fails, although all dependencies are installed and all variables are correct.
SCons maintains a cache in the root directory. When you build different versions of SeisSol or your environment changes (e.g. compiler update), the cache might not be evicted correctly. You can use the option --config=force in when running SCons or delete the hidden cache folder .sconf_temp in SeisSol root directory.

## Asynchronous output¶

Some MPI versions have issues with threads and I/O. The MPI-IO implementation is either not threadsafe or too strict when it comes to locking. This can lead to crashes or deadlocks when using the asynchronous output. A simple workaround is to use the POSIX back-end of the XDMF writer, this is compiling SeisSol without HDF5. For checkpointing, one should also switch to POSIX or SIONlib.

## easi and Intel/16.0¶

Using Intel/16.0 in a local cluster, compile error occurs in PUMLreader.cpp Using Intel/17.0 will solve this.

Using Intel/16.0 on supermuc, we observed some bugs in the fault stress initialization (you can see the imprint of some partitions in the initial stress). The bugs are not showing off with latest Intel/17.0 module.

## “Holes” in the fault output¶

“Holes” in the fault output may appear if the reference point or reference vector, defined by (Xref, Yref, and Zref) is wrongly located (point exactly on the fault, or null vector).