It appears that only HDF5 version 1.10.6 works for windows builds at the moment. The latest version, 1.14.4, doesn't include essential files like libhdf5.lib, among others, which might explain the issue. When I attempted to use version 1.12.2, I encountered the following error:

LINK : warning LNK4286: symbol '?find@Range@moab@@QEBA?AVconst_iterator@12@_K@Z (public: class moab::Range::const_iterator __cdecl moab::Range::find(unsigned __int64)const )' defined in 'Range.obj' is imported by 'Intx2Mesh.obj' [D:\a\dagmc-ci\moab_build\src\MOAB.vcxproj]
... (truncated warnings) ...
D:\a\dagmc-ci\moab_build\lib\Release\MOAB.dll : fatal error LNK1169: one or more multiply defined symbols found [D:\a\dagmc-ci\moab_build\src\MOAB.vcxproj]

For now, I’ve set the HDF5 version to 1.10.6. I assume, the MOAB library needs updates to properly support newer versions of HDF5 for Windows builds.

@gonuke, it seems yaml-cpp is a core dependency now. Please let me know how to address the gtest error.

I think this is failing for Windows on a very specific issue: CMake found the HDF5 libraries successfully during the configuration step, but they are not being found at the linking step. I am not sure how best to address this in Windows. I know there has been some substantial overhaul for Windows build in MOAB on the master branch, but switching to master didn't help.

As @ahnaf-tahmid-chowdhury points out, none of the 1.14.x conda-forge builds include libhdf5.lib files, but that appears to be intentional, so I wonder if we need to direct CMake to use different libraries?

I am not sure about this. I found that HDF5 1.14 does not include libhdf5.lib files. The HDF5 file found by CMake is actually hdf5.lib, which is smaller in size. For example, if you download and extract version 1.10.6, you will find both hdf5.lib (610KB) and libhdf5.lib (10.5MB), and in bin dir you will find hdf5.dll (3MB). However, in version 1.14.4, hdf5.lib (758 KB) and hdf5.dll (3.4MB) is available. I wonder if this is a bug on HDF5's side.

Is there any specific reason for using the conda version of HDF5 here? If we build it from source, I believe we may be able to overcome all the issues we are currently facing here.

Also, we should set our CMake version to at least 3.12 to enable <package>_ROOT. I noticed that these variables are being ignored during the configuration step because we have set the CMake version policy to 3.1. I recommend setting it to 3.18, which is also supported by ubuntu 20.04.

Is there any specific reason for using the conda version of HDF5 here? If we build it from source, I believe we may be able to overcome all the issues we are currently facing here.

If it's straightforward to build HDF5, we could certainly do that. What I'm concerned about is that building HDF5 for Windows launches us on a path for a complete overhaul of our build infrastructure.

I wonder if this is a bug on HDF5's side.

Seems unlikely since it's been a year since they included libhdf5.lib - so either no-one is using this for windows, or there's a way to only rely on hdf5.lib

Perhaps building hdf5 from source is the only option then

I wonder if this is a bug on HDF5's side.

Seems unlikely since it's been a year since they included libhdf5.lib - so either no-one is using this for windows, or there's a way to only rely on hdf5.lib

could we temporary fix this with a simple copy ? and have both name for the lib ?

This is not a good solution, but we could that way move forward :)

On older libraries, they are not the same file - they at least have different sizes.