fix: edge case where component would sometimes be marked a string

[erunion]

🎟 Asana

🆘 Problem

This resolves an edge case that was introduced in some logic I added to repair schemas without a type that was sometimes marking the schema object as a string. This would result in the following "broken" UI:

You can see it in action here: http://bin.readme.com/s/5eaa651315dd18002423bd32

Now if we remove the InputErrorResponse, we'll see that it's resolved:

Now what if you move the InputErrorResponse schema below NewPreauthorization?

😬

🚧 Resolution

So the reason this was happening was that since our cleanupSchemaDefaults method for requestBody and component schema objects is recursive, and we were invoking it against the components object, and not individual schemas, it was difficult for us to ascertain when we were first processing a component schema and when we should add a missing type at the top level of a schema if it didn't have one.

For this specific case, what was tripping up our logic was a few things:

• Both InputErrorResponse and NewPreauthorization schemas didn't have a defined type.
• Our code would correctly identify InputErrorResponse as an object because we hadn't seen a top-level properties property yet until that point, and when we did we immediately knew that we were processing an object, and could mark it as such.
• Once we hit the NewPreauthorization schema, and after we had already processed InputErrorResponse, we had the following data in our internal recursive prevProps memory state:

[
  "schemas", 
  "InputErrorResponse", 
  "properties", 
  "code", 
  "message", 
  "NewPreauthorization"
]

Since we retained knowledge of parsing properties, our code misinterpreted this as having already seen this schema before, is it not parsing properties again for it and since it's missing a type to add string so it's still valid JSON Schema.

Resolving this involved a couple things:

• Shift our recursive processing of component schemas to start at the schema itself, not the components level. This way we'll know exactly when we're first seeing a component schema.
• Update our internal prevProps memory state to clean itself out if we don't have a prevProp. This will resolve the issue of prevProps retaining data from previously processed schemas.

[rafegoldberg]

this stuff is crazy

[erunion]

All this just to remove this kind of stuff because RJSF renders it as an orphan block of text: