feat: add custom strfmt.Registry support

[eddycharly]

This PR adds custom strfmt.Registry support.

Format byte does not allow empty strings (see comes from https://github.com/asaskevich/govalidator/blob/a9d515a09cc289c60d55064edec5ef189859f172/patterns.go#L27).

This makes the secret below invalid:

apiVersion: v1
kind: Secret
metadata:
  name: secret
data:
  key: ''

Now, one could supply it's own registry like this to workaround the problem:

	registry := strfmt.NewFormats()
	b64 := strfmt.Base64([]byte(nil))
	registry.Add("byte", &b64, func(s string) bool {
		if len(s) == 0 {
			return true
		}
		return govalidator.IsBase64(s)
	})

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is NOT APPROVED

This pull-request has been approved by: eddycharly
Once this PR has been reviewed and has the lgtm label, please assign alexzielenski for approval. For more information see the Kubernetes Code Review Process.

The full list of commands accepted by this bot can be found here.

Needs approval from an approver in each of these files:

• OWNERS

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[alexzielenski]

This is an interesting bug. In upstream apiserver, native k8s types do not undergo schema validation so the empty string was allowed. CRDs on the other hand do, and use strfmt.Default.

I just tested this by making a CRD with a field with type: string and format: byte like Secret. K8s apiserver rejected it with:

* bytesField: Invalid value: "": bytesField in body must be of type byte: ""

So to fix this for native types causes CRD validation to be incorrect (unless we hardcoded some logic to decide which strfmt to use depending on the GVK, but thats not nice).

I wonder if this is a bug that should be fixed upstream, so that in k8s CRDs are allowed to have empty byte values

[eddycharly]

I wonder if this is a bug that should be fixed upstream, so that in k8s CRDs are allowed to have empty byte values

I wondered the same. Dunno maybe it was done intentionally.

[eddycharly]

I thought it was acceptable to allow providing a custom registry, WDYT ?

[eddycharly]

Or maybe a registry provider that can return a registry depending on the resource being validated ?

[alexzielenski]

I thought it was acceptable to allow providing a custom registry, WDYT ?

I dont think so, at least for a single registry, since the custom registry will either be wrong for native types, or wrong for CRDs. It would never be correct for both unless you provided a registry depending on the type.

Or maybe a registry provider that can return a registry depending on the resource being validated ?

Yeah, In the worst case scenario we can use the registry conditionally for native types, but I'd like to avoid that if we can. Unfortunately patching upstream seems the solution I can think of.

[eddycharly]

Unfortunately patching upstream seems the solution I can think of.

You mean changing the format validator for byte upstream ?

[k8s-ci-robot]

PR needs rebase.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[eddycharly]

Closing, became useless with recent changes.