Requirements for the new Validator

Requirements for the Result structure

No assumptions about kind of output
- E.g. operationoutcome, list of strings
- Must support conversion into different kinds of output by user-written converters
Must contain enough detail to support interpretation in different languages, different kinds of users (technical, end-user), or even (non-human) interpretation by the software that calls the validator
Must support ‘undetermined’ results
Must support ‘incomplete’ results and continuations
Detail level must be configurable
- reduce gc pressure
- long living refs to external objects (like source instance or conformance structures)
Detailed result reporting is not a by-product of validation it’s the whole point of doing validation
- its worth spending time not just on reporting success or failure, it’s the why that is the core result.
Tracks instance node and definition the result is about

The validator is supposed to find "errors", so these should not throw exceptions
In fact, all the validator does is determine whether an instance conforms to a profile - if it does not, that does not mean it is incorrect, it just doesn't conform. This is particularly the case when using profiles to look for instances in, say, a database -> instances are a "match" or not.
There can be errors caused by misconfiguration and failure of basic internal logic -> exceptions
There can be errors in the source StructureDefinition -> Reported using IErrorSource on the SD->schema converter.
There can be issues arising from invoking a failing subsystem => Undetermined result + report using IErrorSource (not dependent on the location in the instance often)
There can be logical mistakes (trying a binding validation on an integer) -> fail validation(?)
There can be true validation results (above).