Like client function generation, documentation generation amounts to inspecting the API type and extracting all the data we need to then present it in some format to users of your API.
This time however, we have to assist **servant**. While it is able to deduce a lot of things about our API, it can't magically come up with descriptions of the various pieces of our APIs that are human-friendly and explain what's going on "at the business-logic level". A good example to study for documentation generation is our webservice with the `/position`, `/hello` and `/marketing` endpoints from earlier:
While **servant** can see e.g. that there are 3 endpoints and that the response bodies will be in JSON, it doesn't know what influence the captures, parameters, request bodies and other combinators have on the webservice. This is where some manual work is required.
For every capture, request body, response body, query param, we have to give some explanations about how it influences the response, what values are possible and the likes. Here's how it looks like for the parameters we have above.
Types that are used as request or response bodies have to instantiate the `ToSample` typeclass which lets you specify one or more examples of values. `Capture`s and `QueryParam`s have to instantiate their respective `ToCapture` and `ToParam` classes and provide a name and some information about the concrete meaning of that argument, as illustrated in the code above.
`API` is a type provided by **servant-docs** that stores all the information one needs about a web API in order to generate documentation in some format. Out of the box, **servant-docs** only provides a pretty documentation printer that outputs [Markdown](http://en.wikipedia.org/wiki/Markdown), but the [**servant-pandoc**](http://hackage.haskell.org/package/servant-pandoc) package can be used to target many useful formats.
{"subject":"Hey Alp, we miss you!","body":"Hi Alp,\n\nSince you've recently turned 26, have you checked out our latest haskell, mathematics products? Give us a visit!","to":"alp@foo.com","from":"great@company.com"}
However, we can also add one or more introduction sections to the document. We just need to tweak the way we generate `apiDocs`. We will also convert the content to a lazy `ByteString` since this is what **wai** expects for `Raw` endpoints.
More customisation can be done with the `markdownWith` function, which allows customising some of the parameters used when generating Markdown. The most obvious of these is how to handle when a request or response body has multiple content types. For example, if we make a slight change to the `/marketing` endpoint of our API so that the request body can also be encoded as a form:
``` haskell
type ExampleAPI2 = "position" :> Capture "x" Int :> Capture "y" Int :> Get '[JSON] Position
:<|> "hello" :> QueryParam "name" String :> Get '[JSON] HelloMessage
If, however, you don't want the extra example encoding shown, then you can use `markdownWith (defRenderingOptions & requestExamples .~ FirstContentType)` to get behaviour identical to `markdown apiDocs`.
And if you spin up this server and request anything else than `/position`, `/hello` and `/marketing`, you will see the API docs in markdown. This is because `serveDocs` is attempted if the 3 other endpoints don't match and systematically succeeds since its definition is to just return some fixed bytestring with the `text/plain` content type.
