dabele
commented
2 years ago related to #66 since it is planned to merge epidemiology and epidemiology_io libraries.
I think it is realistic for people to want to build their own IO to have their own data format or integrate just our simulation into their framework. also, right now we only require hdf5 for output, of course that could change.
Here's how I think it should work:
- flags should be specific to the dependency, e.g. EPI_USE_HDF5 (or s.th. like that). That way it can be properly configured exactly how you need it (i.e. use json for input, but use something other than hdf5 for output) especially since hdf5 is not a "light" dependency. Of course some dependencies will be required.
- flags should be autodetected, i.e. use HDF5 if you can find it using cmake
find_package - there should be a very limited amount of tests for these config options, i.e. not add the full set of windows, linux, release, debug, min, latest, etc, because that explodes quickly; maybe one windows, one linux test, because platform has a bigger impact on finding dependencies than the compiler
So far our standard setting for EPI_IO is off since hdf5 is needed as additional dependency. However, all our CI jobs use EPI_IO=ON. In my opinion, we should also test with EPI_IO=OFF or make it an obligatory/standard "setting" (that you maybe even can't turn off). Disabling to the possibility to set it off is surely not the best option, but the current handling isn't either. The question is: Will there be a use case where you really do not need the IO. I rather doubt it since you will probably always also have a detailed look into the results, otherwise the runs of the software are rather useless.
@dabele @joergbrech What do you think?