Is passing NULL to Node::set_external valid behavior? about conduit HOT 4 OPEN

@rhornung67 yes, that is a case I was thinking about.

I don't know the exact order of the steps, we may not use NULLs, but that type of case is one where if folks are careful it makes sense for this to be allowed.

Error checking for null in verify and in other places is something we should add.

from conduit.

@BradWhitlock sorry this caused you frustration.

I don't recommend it, but I think there are folks that will want to use NULL w/ set external.

I agree that blueprint verify should flag invalid if pointers are NULL, and that the to_string methods (json / yaml) should error instead of crashing if they encounter null pointers.

from conduit.

I can possibly see allowing NULL to act as a placeholder for data that is not created yet. Still, it makes a Blueprint tree that is not valid and cannot be used with relay, partitioning, or other algorithms that do not expect to get NULL data in field values. It's easier just to not allow in the first place.

from conduit.

@cyrush don't we do this sort of thing (set up a BP tree with null external data ptrs) in Sidre when we read in a restart dump for a Fortran code, for example? IIRC, the BP tree is defined first, then Fortran app allocates the data arrays, then Conduit reads the data into those arrays from an HDF5 file.

@BradWhitlock maybe still allow this process, but don't allow using relay, etc. on an invalid BP tree?

from conduit.