If I leave a field out of a partial update, I usually mean “leave it alone”. If I deliberately clear it, I expect the API to have a way to express that. Collapsing both requests into the same value is a small implementation shortcut that can make an edit form surprisingly destructive.
Consider a profile with an optional nickname. These payloads need a documented meaning:
{}
{"nickname": null}
For an endpoint using JSON Merge Patch, the first leaves the nickname unchanged and the second removes that member from the target JSON document. That format uses application/merge-patch+json. Its rules do not automatically apply to every endpoint called PATCH; choose and document the request format rather than expecting the HTTP method to explain the body.
There is a useful limitation here too. Merge Patch gives null a removal meaning, so it is awkward when storing an explicit JSON null is itself a meaningful operation. Arrays are replaced as a whole rather than edited item by item. If those rules do not fit the resource, choose a format that does. Calling an endpoint a partial update does not settle those decisions.
The implementation needs to preserve field presence from parsing through to persistence. A request mapper that fills every omitted property with null has already lost the information. So has validation that supplies a default before the update code can tell whether the client sent the field. Defaults that make sense when creating a record can be wrong when editing one.
I would make the contract explicit for the fields the endpoint allows people to change. Omitted means unchanged; a supplied value is validated and applied; an explicit clear is either supported or rejected with a useful explanation. An empty string should follow the field’s own rules too. It might be valid text, an invalid value or a documented way to clear something, but it should not acquire a meaning by accident.
This is worth a small test through the real request handling: change one field and check that an omitted neighbouring field survives, then explicitly clear the optional field and check the intended result. A unit test of the final update function can miss the mapper that erased the distinction several steps earlier.
