mithril-vndb/docs/request.md

request(options)

• API
• How it works
• Typical usage
• Dynamic URLs
• Why JSON instead of HTML
• Why XMLHttpRequest instead of fetch
• Why return streams

---

API

stream = m.request(options)

Argument	Type	Required	Description
options.method	String	Yes	The HTTP method to use. This value should be one of the following: GET, POST, PUT, PATCH, DELETE, HEAD or OPTIONS.
options.url	String	Yes	The URL to send the request to. The URL may be either absolute or relative, and it may contain interpolations.
options.data	any	No	The data to be interpolated into the URL and serialized into the querystring (for GET requests) or body (for other types of requests).
options.async	Boolean	No	Whether the request should be asynchronous. Defaults to true.
options.user	String	No	A username for HTTP authorization. Defaults to undefined.
options.password	String	No	A password for HTTP authorization. Defaults to undefined. This option is provided for XMLHttpRequest compatibility, but you should avoid using it because it sends the password in plain text over the network.
options.config	xhr = Function(xhr)	No	Exposes the underlying XMLHttpRequest object for low-level configuration. Defaults to the identity function.
options.type	any = Function(any)	No	A constructor to be applied to each object in the response. Defaults to the identity function.
options.serialize	string = Function(any)	No	A serialization method to be applied to data. Defaults to JSON.stringify.
options.deserialize	any = Function(string)	No	A deserialization method to be applied to the response. Defaults to a small wrapper around JSON.parse that returns null for empty responses.
options.extract	string = Function(xhr, options)	No	A hook to specify how the XMLHttpRequest response should be read. Useful for reading response headers and cookies. Defaults to a function that returns xhr.responseText
options.initialValue	any	No	A value to populate the returned stream before the request completes
options.useBody	Boolean	No	Force the use of the HTTP body section for data in GET requests when set to true, or the use of querystring for other HTTP methods when set to false. Defaults to false for GET requests and true for other methods.
returns	Stream		A stream that resolves to the response data, after it has been piped through the extract, deserialize and type methods

How to read signatures

---

How it works

The m.request utility is a thin wrapper around XMLHttpRequest, and allows making HTTP requests to remote servers in order to save and/or retrieve data from a database.

m.request({
	method: "GET",
	url: "/api/v1/users",
}).run(function(users) {
	console.log(users)
})

Calls to m.request return a stream.

---

Typical usage

Here's an illustrative example of a self-contained component that uses m.request to retrieve some data from a server.

var SimpleExample = {
	oninit: function(vnode) {
		vnode.state.items = m.request({
			method: "GET",
			url: "/api/items",
			initialValue: []
		})
	},
	view: function(vnode) {
		return vnode.state.items().map(function(item) {
			return m("div", item.name)
		})
	}
}

m.route(document.body, "/", {
	"/": SimpleExample
})

Let's assume making a request to the server URL /api/items returns an array of objects in JSON format.

When m.route is called at the bottom, SimpleExample is initialized. oninit is called, which calls m.request and assigns its return value (a stream) to vnode.state.items. This stream contains the initialValue (i.e. an empty array), and this value can be retrieved by calling the stream as a function (i.e. value = vnode.state.items()). After the oninit method returns, the component is then rendered. Since vnode.state.items() returns an empty array, the component's view method also returns an empty array, so no DOM elements are created. When the request to the server completes, m.request parses the response data into a Javascript array of objects and sets the value of the stream to that array. Then, the component is rendered again. This time, vnode.state.items() returns a non-empty array, so the component's view method returns an array of vnodes, which in turn are rendered into div DOM elements.

Loading icons and error messages

Here's an expanded version of the example above that implements a loading indicator and an error message:

var RobustExample = {
	oninit: function(vnode) {
		var req = m.request({
			method: "GET",
			url: "/api/items",
		})
		vnode.state.items = req.catch(function() {
			return []
		})
		vnode.state.error = req.error.run(this.errorView)
	},
	view: function(vnode) {
		return [
			vnode.state.items() ? vnode.state.items().map(function(item) {
				return m("div", item.name)
			}) : m(".loading-icon"),
			vnode.state.error(),
		]
	},
	errorView: function(e) {
		return m(".error", "An error occurred")
	}
}

m.route(document.body, "/", {
	"/": RobustExample
})

When this component is initialized, m.request is called and its return value is assigned to req. Unlike the previous example, here there's no initialValue, so the req stream is in a pending state, and therefore has a value of undefined. req.error is the error stream for the request. Since req is pending, the req.error stream also remain in a pending state, and likewise, vnode.state.error stays pending and does not call this.errorView.

Then the component renders. Both vnode.state.items() and vnode.state.error() return undefined, so the component returns [m(".loading-icon"), undefined], which in turn creates a loading icon element in the DOM.

When the request to the server completes, req is populated with the response data, which is propagated to the vnode.state.items dependent stream. (Note that the function in catch is not called if there's no error). After the request completes, the component is re-rendered. vnode.state.error() is still undefined, but now view returns a list of vnodes containing item names, and therefore the loading icon is replaced by a list of div elements are created in the DOM.

If the request to the server fails, catch is called and vnode.state.items() is set to an empty array. Also, req.error is populated with the error, and vnode.state.error is populated with the vnode tree returned by errorView. Therefore, view returns [[], m(".error", "An error occurred")], which replaces the loading icon with the error message in the DOM.

---

Dynamic URLs

Request URLs may contain interpolations:

m.request({
	method: "GET",
	url: "/api/v1/users/:id",
	data: {id: 123},
}).run(function(user) {
	console.log(user.id) // logs 123
})

In the code above, :id is populated with the data from the {id: 123} object, and the request becomes GET /api/v1/users/123.

Interpolations are ignored if no matching data exists in the data property.

m.request({
	method: "GET",
	url: "/api/v1/users/foo:bar",
	data: {id: 123},
})

In the code above, the request becomes GET /api/v1/users/foo:bar

---

Why JSON instead of HTML

Many server-side frameworks provide a view engine that interpolates database data into a template before serving HTML (on page load or via AJAX) and then employ jQuery to handle user interactions.

By contrast, Mithril is framework designed for thick client applications, which typically download templates and data separately and combine them in the browser via Javascript. Doing the templating heavy-lifting in the browser can bring benefits like reducing operational costs by freeing server resources. Separating templates from data also allow template code to be cached more effectively and enables better code reusability across different types of clients (e.g. desktop, mobile). Another benefit is that Mithril enables a retained mode UI development paradigm, which greatly simplifies development and maintenance of complex user interactions.

By default, m.request expects response data to be in JSON format. In a typical Mithril application, that JSON data is then usually consumed by a view.

You should avoid trying to render server-generated dynamic HTML with Mithril. If you have an existing application that does use a server-side templating system, and you wish to re-architecture it, first decide whether the effort is feasible at all to begin with. Migrating from a thick server architecture to a thick client architecture is typically a somewhat large effort, and involves refactoring logic out of templates into logical data services (and the testing that goes with it).

Data services may be organized in many different ways depending on the nature of the application. RESTful architectures are popular with API providers, and service oriented architectures are often required where there are lots of highly transactional workflows.

---

Why XMLHttpRequest instead of fetch

fetch() is a newer Web API for fetching resources from servers, similar to XMLHttpRequest.

Mithril's m.request uses XMLHttpRequest instead of fetch() for a number of reasons:

• fetch is not fully standardized yet, and may be subject to specification changes.
• XMLHttpRequest calls can be aborted before they resolve (e.g. to avoid race conditions in for instant search UIs).
• XMLHttpRequest provides hooks for progress listeners for long running requests (e.g. file uploads).
• XMLHttpRequest is supported by all browsers, whereas fetch() is not supported by Internet Explorer and Safari.

Currently, due to lack of browser support, fetch() typically requires a polyfill, which is over 11kb uncompressed - nearly three times larger than Mithril's m.request.

Despite being much smaller, m.request supports many important and not-so-trivial-to-implement features like URL interpolation, querystring serialization and JSON-P requests. The fetch polyfill does not support any of those.

The fetch() API does have a few technical advantages over XMLHttpRequest in a few uncommon cases:

• it provides a streaming API (in the "video streaming" sense, not in the reactive programming sense), which enables better latency and memory consumption for very large responses (at the cost of code complexity).
• it integrates to the Service Worker API, which provides an extra layer of control over how and when network requests happen. This API also allows access to push notifications and background synchronization features.

In typical scenarios, streaming won't provide noticeable performance benefits because it's generally not advisable to download megabytes of data to begin with. Also, the memory gains from repeatedly reusing small buffers may be offset or nullified if they result in excessive browser repaints. For those reasons, choosing fetch() streaming instead of m.request is only recommended for extremely resource intensive applications.

---

Why return streams

Normally, XMLHttpRequest makes HTTP requests to a server asynchronously. This means that it cannot return the response data via a return statement, since the return statement runs synchronously long before the response data is actually available. Therefore, any library that makes requests must expose the response data using some other mechanism.

Some older libraries do so via callback functions, and newer ones (including the fetch API) return promises. m.request returns reactive streams.

Callback functions are the most basic mechanism for asynchronous flow control. They are not composable because they require the callback function to be passed at call time, and error handling mechanisms must similarly be declared upfront.

However, it's desirable to allow the callback function to be defined (and broken into subroutines) in different places than the call site, in order to achieve better separation of concerns. In addition, it's also desirable to wrap an abstraction around errors so that they can be thrown freely and handled safely from a single place, rather than requiring try/catch blocks in every callback function, or duplicating error handling code. The problems that arise from callbacks' lack of composability are infamous enough to earn nicknames such as "callback hell" and "pyramids of doom".

The Promise API is designed to address the shortcomings of callbacks. They are composable, which allows code to be refactored much more elegantly than using callbacks.

In the example below, the code is written in a naive style. It's highly procedural and does many different things: First it requests a project with id 123, then requests a user whose id is the value of project.ownerID, then proceeds to do something useful with the user. If there's an error, it is logged to console.

// AVOID
function doStuff() { /*...*/ }

function json(response) {
	return response.json()
}
function doStuffWithProjectOwner(projectID) {
	fetch("/api/v1/projects/" + projectID, {method: "GET"}).then(json).then(function(project) {
		fetch("/api/v1/users/" + project.ownerID, {method: "GET"}).then(json).then(function(user) {
			doStuff(user)
		})
		.catch(function(e) {
			console.log(e)
		})
	})
	.catch(function(e) {
		console.log(e)
	})
}
doStuffWithProjectOwner(123)

Here's a refactored version that defines composable, easy-to-reuse units, and that takes advantage of the error propagation feature of Promises to avoid repetitive error handling code:

// PREFER
function findProject(id) {
	return fetch("/api/v1/projects/" + id, {method: "GET"}).then(json)
}
function findUser(id) {
	return fetch("/api/v1/users/" + id, {method: "GET"}).then(json)
}
function getProjectOwnerID(project) {
	return project.ownerID
}

function doStuffWithProjectOwner(projectID) {
	return findProject(projectID)
		.then(getProjectOwnerID)
		.then(findUser)
		.then(doStuff)
		.catch(function(e) {
			console.log(e)
		})
}

doStuffWithProjectOwner(123)

The code above separates each request into a findProject and findUser functions which can be used in more use cases than only finding the user object that owns a project.

You can think of .then callbacks as pipes: the getProjectOwnerID receives the response of the findProject request as an input, and return an id, which is then passed as the input to findUser.

The feature of Promises that let us simplify error handling is that promises absorb other promises: in the .then(findUser) line, findUser returns a promise. Instead of a promise being passed as an input to the next callback, the promise chain waits for the findUser promise to complete, and only then continues down the chain of callbacks with the resolved value. If findUser throws an error, the .catch callback handles it, in addition to handling erros from findProject (and from getProjectOwnerID, for that matter).

Promises provide the machinery that facilitates writing small straightforward functions and composing them in flexible ways.

Mithril streams have many similarities to promises. The example above could be written like this:

// PREFER
function findProject(id) {
	return m.request({method: "GET", url: "/api/v1/projects/:id", data: {id: id}})
}
function findUser(id) {
	return m.request({method: "GET", url: "/api/v1/users/:id", data: {id: id}})
}
function getProjectOwnerID(project) {
	return project.ownerID
}

function doStuffWithProjectOwner(projectID) {
	return findProject(projectID)
		.run(getProjectOwnerID)
		.run(findUser)
		.run(doStuff)
		.catch(function(e) {
			console.log(e)
		})
}

doStuffWithProjectOwner(123)

Aside from the API signature difference between fetch and m.request, the only change required to achieve the same functionality was to replace all instances of .then with .run.

However, stream have some additional interesting properties. Let's suppose project objects have a team property that contains a list of user objects, and we wanted to display a list of designers and a list of developers in a project:

function getProjectTeam(project) {
	return project.team
}
function getTeamUsersByType(team, type) {
	return team.filter(function(user) {
		return user.type === type
	})
}
var project = findProject(123)
var team = project.run(getProjectTeam)
var designers = team.run(function(team) {
	return getTeamUsersByType(team, "designer")
})
var developers = team.run(function(team) {
	return getTeamUsersByType(team, "developer")
})

Now let's suppose that the team changed for the project and we need to fetch the project object from the server again. We would logically also want to update designers and developers so that the UI displays the correct users in their respective lists.

Fortunately, project is a stream, and team, designers and developers are streams derived from project. So to update the state of all these streams, we only need to do this:

findProject(123).run(project)

Doing so updates all the streams, and therefore there's no need to place the filtering code in the view, where the filtering code would recompute the same thing on every render.

Returning streams from m.request streamlines use cases where efficient reactivity is desired, without losing the composable semantics of promises.