pnorbert
pnorbert
hmm, really don't remember, I guess it's just an ancient test never updated to BP4, and now it fails with BP4.
@franzpoeschel How do you store chars in HDF5 in a portable way? In my tests H5T_NATIVE_CHAR becomes H5T_NATIVE_UCHAR/H5T_NATIVE_SCHAR depending on signedness, which in turn become H5T_STD_U8LE/H5T_STD_I8LE in the HDF5 file...
That's a pity. Now we can support char in ADIOS2 in a portable way as long as you write to BP. Once it goes to HDF5, you cannot read it...
This whole thing is up for discussion. I don't remember right now where vector allocation was a measurable issue.
It is so slow in Debug mode, that these test take a minute to run. Engine.BP.BPBufferSizeTest.SyncDeferredIdenticalUsage.BP3.Serial Engine.BP.BPBufferSizeTest.SyncDeferredIdenticalUsage.BP4.Serial
I have added a few BP3 and BP4 datasets for this backward compatibility tests. They are small so far so I guess they are okay to store here for now.
These tests can move to another repo entirely if we can make sure they are executed at least before every release. It is not urgent for this release, as it...
@chuckatkins As we are inching towards the next release, it is time to revisit this PR.
IMHO, it may be easier for an application, especially a middleware between application and ADIOS, to just ask the Engine itself if it is inside a Step to avoid calling...
Oh, same for the ADIOS class, the constructors with MPI are also missing, and the same problem for the C API.