fix: improve handling of allOf / anyOf / oneOf #410
Conversation
|
|
||
| type IRTypeIntersection = {type: "type-intersection"; types: IRType[]} | ||
| type IRTypeUnion = {type: "type-union"; types: IRType[]} | ||
| type IRTypeOther = string |
There was a problem hiding this comment.
Eventually will probably aim to leave the typescript types (and schema code) in a "non-stringy" representation all the way through to the emit phase.
This will be particularly important for contextual alterations (eg; readonly / writeonly modifiers)
|
|
||
| // TODO: https://github.com/hapijs/joi/issues/3057 | ||
| it.skip("can intersect unions", async () => { | ||
| it("can intersect unions", async () => { |
There was a problem hiding this comment.
Was somewhat forced to fix this, as integration tests started outputting intersected unions due to the richer anyOf / oneOf / allOf support.
The "fix" is rather hacky, but seems to work well enough
| }) | ||
| }) | ||
|
|
||
| async function getActualFromModel( |
There was a problem hiding this comment.
Plan to later properly flesh out a set of test cases using in-line IRModel's directly, for cases where $ref isn't required.
| } | ||
|
|
||
| export interface IRModelObject extends IRModelBase { | ||
| allOf: MaybeIRModel[] |
There was a problem hiding this comment.
after this change, it's only the input normalization that has to deal with things being an object, composition or both at once.
intersectionandunion, making handling of compositions in schema/type builder easier to reason aboutoneOfis being treated as equivalent toanyOffor nowallOf/intersectionwhen an object schema is part of a composite, fixing previous information lossoneOf/anyOfto support "looking up" property definitions from the parent objectafter this change it appears
EmptyObjectis no-longer emitted by any of the integration-tests - which is generally a good sign