Is passing NULL to Node::set_external valid behavior?
BradWhitlock opened this issue · comments
I am debugging some code in which a host code exposes some of its internal data as a Blueprint tree of data. The code makes a topology and some fields. Some of the fields were apparently allocated later. At the time the Blueprint fields are created, some of the arrays are NULL and the host makes nodes anyway and calls set_external(), passing NULL pointers! Back in my library, I get the host code's node (which contains some NULLs) and I passed this node to conduit::relay::io::save() to save a conduit_json representation of the host code's mesh and fields. This causes relay to die because some of the external nodes are NULL.
- I think Node::set_external(T*, len) should have immediately thrown an exception when given a NULL pointer.
- The conduit_json method in relay should have scanned its input nodes and thrown when it found an external node with NULL data.
- Various verify() methods should flag this case as invalid
This was hard to track down and it was all because the host code was sloppy about how it made the Blueprint mesh.
@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.
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.
@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?
@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.