I am trying build an api which conforms to the json:api spec.
My api has three resources /task, /item and /result. A task has the fields name, description and state. A item has the fields itemName. A count is kept server-site for the item and the count is returned when a user retrieves the item with a GET request. The count is incremented server-side when the item is updated. There is a one-to-many relationship between task and item. In a sense an item is appended to a task. When the tasks state changes a script runs server-side to do some processing on the associated items. Once the script finishes the output is available in the result resource.
Per the spec, I am using the POST verb to create a task and the PATCH to update a task. I just want one endpoint which handles both the create /update (appending) of an item. But, I'm not sure which verb to use? Can I use PATCH to update the item but also create an item if it doesn't exist?
I also thought that perhaps I should be using the PUT verb. But, my understanding here is that this verb is used to simply replace a resource rather than update it. I don't think this is right for my user-case as an items count is incremented when updated, so replacing it is not what I want todo. But, the count is handled server-side so a user doesn't have the option of "replacing" the count anyway.
my understanding here is that this verb is used to simply replace a resource rather than update it.
This is a common understanding - wrong, but common.
The IANA registry documents the authoritative reference for the semantics of http methods. Most of the common ones are defined by RFC 7231; PATCH is defined by RFC 5789.
PUT is an appropriate choice when the message body is a complete representation of what you want the resource to be. It may be easier to think about "saving a file"; PUT describes what the client expects the document to look like when it has been saved.
It's appropriate to use PUT for either updating a document or creating one, provided that the client knows the identifier for the document (just in the same way that we can use save to create a file, or replace a file, but we need to know the file name).
If you read the text of the specification, you'll see that - while the semantics of the request are to save the new representation "as-is", the server isn't required to do that -- the server, after all, is in control of its own documents -- so there is room to cover read only fields, or fields that should be updated by the server only. You need to have a little bit of care with the response headers to avoid implying that you saved the representation as is, but other than that you should be fine.