Performance/Memory issues with rdl.Validate when FullValidation enabled

[evantorrie]

The current go implementation of rdl.Validate calls NewTypeRegistry(schema) on every invocation. For a complex schema, this is a costly operation.

ardielle-go/rdl/validator.go

Line 46 in 0e40fcc

checker.registry = NewTypeRegistry(schema)

This becomes apparent in go when unmarshalling large objects with nested structs containing a lot of "derived-from-String" types.

In tests, I've found that code which would just cache the TypeRegistry per schema when calling schema.Build() and then use that in a rdl.Validate2() style call which can accept a preconstructed TypeRegistry gives a close to 10x improvement on JSON unmarshalling speed.

[boynton]

Thanks, good find. Will fix that. Or... you can if you want :-)

[evantorrie]

I have a fix which involves exposing a Validator object -- and then having the code generator generate an instance of that at schema build time. I'll submit a PR, but I don't know enough details about the implementations in other languages to know whether it's the best way to solve this issue.

[boynton]

Well, Fixing it behind the scenes would be best, it is just a performance optimization for the Go library, other libraries have different ways to deal with it, so don't worry about those just yet. If you want to just send me a sketch (or pointer to your changes) I can integrate it, I just thought you were closer to it, having done the performance comparison already.

[evantorrie]

I changed it to use a transparent cache - no change to the exposed API.

Take a look at #29

[boynton]

Fixed in Release 1.4.13, PR #29