This change removes reflection from validator providers, and instead
relies on cached metadata in in the modelmetadata.
In general this means that our MVPs don't need to cache anything, they
just look at the metadata and create what they need.
In the case of data-annotations, we update the model details provider to
add validation attributes to the modelmetadata. This would allow someone
to replace the DataAnnotationsValidatorProvider, but still use the
metadata in these attributes.
The change to the IModelValidatorProvider api (to use a context) is
intended to minimize allocations. Currently each validator provider needs
to return a list so you end up with N+1 lists (N validators + a final list
to compine them all). This change will let us just create the final list
(and a small context object). This is a very very high traffic API so it
seemed worth doing.
There's also some general massaging of namespaces and file locations.
Separates the MMP into two phases:
1). Creation of the ModelMetadata, discovery of properties and attributes
(reflection) is part of the MMP
2). Lookup of details based on attributes is now part of another phase,
and has its results cached.
Users can now implements and register an IFooMetadataProvider to customize
a single aspect of metadata (see how data annotations does it).
This also fixes#1503.
Currently all model binders except mutable object binder are independent of validation code. The mutable object binder which needs to do some validation ( for scenarios involving [BindRequired] and [BindNever]).
We would be going with an approach where required validaiton happens in input formatters and model binders.
This is needed as validation for value types can best be done at creation time.
Followup PRs:
Introduce support for skipping validation (and not binding) for a particular property/type etc.
Ajay Bhargav Baaskaran
Harsh Gupta
Ryan Nowak
N. Taylor Mullen