Prevent H5Pset* routines from modifying default plists

[mattjala]

This prevents the connector in H5VL_def_conn_s and the one used by H5P_DEFAULT/H5P_FILE_ACCESS_DEFAULT from diverging.

[qkoziol]

In principal, yes, but rejecting the default property list of each type should apply to all the 'set' routines, for all types of property lists.

And, you don't need to reject H5P_DEFAULT, since H5P_object_verify() will already fail for that value.

[qkoziol]

I suggest that you put the appropriate checking inside H5P_object_verify()

[jhendersonHDF]

In principal, yes, but rejecting the default property list of each type should apply to all the 'set' routines, for all types of property lists.

In this case I'd call it a "not letting perfect be the enemy of good enough" situation. I asked Matt to create this PR due to the issue I encountered specifically with changing the VOL on the default FAPL. Fixing all the cases (and testing them) would be a decently larger effort.

And, you don't need to reject H5P_DEFAULT, since H5P_object_verify() will already fail for that value.

I'd actually prefer to explicitly reject H5P_DEFAULT and H5P_FILE_ACCESS_DEFAULT first as an invalid argument check and not rely on H5P_object_verify() doing that, in case someone updates it in the future to accept H5P_DEFAULT and another parameter to allow it to translate to the correct property list.

[qkoziol]

I understand your perspective, but I think the problem should be easy to address in a few hours of work: add a new 'is_lib_default' flag to the property list structure, which is only set to true for the actual default property lists, and is set to false when one of those property lists is copied.

Then add a parameter to H5P_object_verify(), or someplace else that's central to the 'set' calls, that indicates that H5P_object_verify() is being called from a 'set' routine. Then fail in H5P_object_verify() when both are true.

Probably only ~50 lines of actual code. (and 600 lines of trivial changes to calls to H5P_object_verify :-)

[qkoziol]

Please merge the new and old tests together, so there's only one property list test.

Otherwise, looks good - thanks for taking the time to implement a holistic approach that addresses the problem at its root.