Validation Plugins kdbGet vs. kdbSet

[kodebach]

Some plugins perform validations during kdbGet (postgetstorage) and kdbSet (presetstorage), while others do it only during kdbSet. This is inconsistent and we should find a standard way to do things.

Examples of validation during kdbGet and kdbSet: type, macaddr
Examples of validation only during kdbSet: network, reference

Performing validation only during kdbSet can lead to problems, if the storage file is modified directly. Also some plugins also perform transformations in kdbGet, in which case validation during kdbGet is a must. Therefore, I think the standard way should be to validate during kdbGet and during kdbSet.

There is also the question of what to do, if validation fails. During kdbSet, it is pretty obvious that there should be an error to prevent creating invalid files. However, in kdbGet it is not so clear. If we produce an error and make kdbGet fail, we can never use kdbSet to fix the problem. But if we don't produce an error (and e.g. only produce a warning), applications may accidentally load invalid configurations.

IMO the best solution would be to validate both during kdbGet and kdbSet and always produce errors, unless a global flag (passed via the global keyset) is active. If the global flag is set, validation errors during kdbGet only produce a warning, but in kdbSet we still get an error. This allows us to fix broken configurations, but by default applications will not use invalid configs.

[markus2330]

I think it is easier to turn validation warnings to errors (e.g. by high-level APIs), then to implement such a flag.

And a validation plugin could remove non-validating keys that are not required. The only remaining case is if a key is required and invalid we should actually make an error, even on get.