Check valid relation-tuple characters on insert

[relvacode]

Is your feature request related to a problem? Please describe.

With a postgres backend it's possible to insert a relation-tuple such that certain read operations always fail due to malformed string input when the server encounters a tuple in the database which contains a subject as a subject set with a colon : in the set's object.

After the problem relation-tuple is created it's no longer possible to check, or list relation-tuples that encounter it.

Now, this is described in the limitations page https://www.ory.sh/keto/docs/concepts/subjects and the document does heavily recommend using UUID mappings so this isn't a bug per-se. However, the server will still accept these on insert over gRPC causing a very vague error when attempting any read operation (I was only able to trace this down by looking at Keto's code).

Related code

keto/internal/relationtuple/definitions.go

Line 199 in 0ac4fe1

func (s *SubjectSet) FromString(str string) (Subject, error) {

For example,

members of group:B are members of group:A

NAMESPACE	OBJECT   RELATION NAME   SUBJECT
default		group:B	 member          default:group:A#member

&acl.RelationTupleDelta{
    Action: acl.RelationTupleDelta_INSERT,
    RelationTuple: &acl.RelationTuple{
        Namespace: "default",
        Object:    "group:B",
        Relation: "member",
        Subject:  &acl.Subject{
            Ref: &acl.Subject_Set{
                Set: &acl.SubjectSet{
                    Namespace: "default",
                    Object:    "group:A",
                    Relation:  "member",
                },
            },
        },
    },
}

Describe the solution you'd like

Relation tuples should be validated on insert such that they can be parsed back out of the database. In other words, anything that goes in should be able to come back out.

Describe alternatives you've considered

I've changed my application to not use any of the disallowed characters but this may help those that encounter this in future.

[zepatrik]

This will be resolved as part of #638 because we don't string en- and de-code anymore with that.
Thanks for the detailed report 👍

[zepatrik]

Btw, in your case you should consider having a "group" namespace, as that is really what namespaces are for.
See https://www.ory.sh/keto/docs/concepts/namespaces#scoping-of-objects for some details.