This proposal expresses a design that leans towards solving problems with middleware. This is motivated by the need to support migratability from Hubot, deliver extensibility in the framework, and to lay the ground work for a plugin API in the future.
Please add feedback Does this feel unintuitive? Complicated? Are there use cases that you beleive should be easy hard to implement using this API? Please comment below.
There are some wrinkles that still need to be ironed out. These are noted below where they come up.
You create an app by calling a constructor, which is a top-level export.
const { Slapp } = require('slapp');
// NOTE: the parameters to the constructor are not fully specified here
const app = new Slapp(token, options);
/* Add functionality here */
app.start();
NOTE: This change eliminates the attachToExpress()
method in favor of abstracting the specific reciever. start()
calls
through to the reciever to initialize it in whichever manner it wants to. The default reciever is built on express, so this
call would tell it to listen. Alternatively, the default reciever may export the express app, so that you can use it as part
of an existing HTTP server.
Apps typically react to incoming events, which can be events, actions, commands, options requests, etc. For each type of events, there's a method to attach a listener function.
// Listen for an event from the Events API
app.event(eventType, fn);
// Listen for an action from a button, menu, dialog submission, or message action
app.action(callbackId, fn);
// Listen for a command
app.command(commandName, fn);
// Listen for options requests
app.options(callbackId, fn);
There's a special method that's provided as a convenience to handle Events API events with the type message
. When using
this method to attach a listener function, it will not be called for messages that were sent from this app (can be disabled
using the ignoreSelf
option on the constructor). Also, you can include a string or RegExp pattern
before the listener
function to only call that listener function when the message text matches that pattern.
app.message([pattern ,] fn);
Most of the app's functionality will be inside listener functions (the fn
parameters above). These functions are called
with arguments that make it easier to build a rich app.
-
payload: object
(aliases:message
,event
) - The contents of the event. The exact structure will depend on which type of event this listener is attached to. For example, for an event from the events API, it will be an event type structure. NOTE: Fully specify which structures this could be (see@slack/events-api
,@slack/interactive-messages
for examples). -
context: object
- The event context. This object contains data about the message and the app, such as thebotId
. See advanced usage for more details. -
say?: (string|object) => void
- A function to respond to an incoming event. This argument is only available when the listener is triggered for event that contains achannel_id
(includingmessage
events). Call this function to send a message back to the same channel as the incoming event. It accepts both simple strings (for plain messages) and objects (for complex messages, including blocks or attachments). -
ack?: (string|object|undefined) => void
- A function to acknowledge that an incoming event was recieved by the app. Incoming events from actions, commands, and options requests must be acknowledged by calling this function. See acknowleding events for details. -
respond?: (string|object) => void
- A function to respond to an incoming event. This argument is only available when the listener is triggered for an event that contains aresponse_url
(actions and commands). Call this function to send a message back to the same channel as the incoming event, but using the semantics of theresponse_url
. NOTE: Fully specify what this means in each case for buttons, menus, dialog submissions, slash commands, and message actions. -
body: object
- An object that contains the whole body of the event, which is a superset of the data inpayload
. Some types of data are only available outside the event payload itself, such asapi_app_id
,authed_users
, etc. This argument should rarely be needed, but for completeness it is provided here.
The arguments are grouped into properties of one object, so that its easier to pick just the ones your listener needs (using object destructuring). Here is an example where the app sends a simple response, so there's no need for most of these arguments:
// Reverse all messages the app can hear
app.message(({ message, say }) => {
const reversedText = message.text.split('').reverse().join('');
say(reversedText);
});
Listeners can use the full power of all the methods in the Web API (given that your app is installed with the appropriate
scopes). Each app has a client
property that can be used to call these methods. See the WebClient
documentation for
a more complete description of how it can be used.
// Reacts to any message that contains "happy" with a 😀
app.message('happy', ({ message }) => {
// Calls the "reactions.add" Web API method
app.client.reactions.add({ name: ':grinning:', channel: message.channel_id })
.then(console.log)
.catch(console.error);
});
Some types of events need to be acknowledged in order to ensure a consistent user experience inside the Slack client (web, mobile, and desktop apps). This includes all actions, commands, and options requests. Listeners for these events need to call the ack()
function, which is passed in as an argument.
In general, the Slack platform expects an acknowledgement within 3 seconds, so listeners should call this function as soon as possible.
Depending on the type of incoming event a listener is meant for, ack()
should be called with a parameter:
-
Button clicks, menu selections, and slash commands: Either call
ack()
with no parameters, astring
to to update the message with a simple message, or anobject
to replace it with a complex message. Replacing the message to remove the interactive elements is a best practice for an action should only be performed once. -
Message actions: Call
ack()
with no parameters. TODO: is there message replacement for actions? -
Dialog submissions: Call
ack()
with no parameters when the inputs are all valid, or an object describing the validation errors if any inputs are not valid. -
Options requests: Call
ack()
with an object containing the options for the user to see.
If an app does not call ack()
within the time limit, Slapp will generate an error. See handling errors for more details.
If an error occurs in a listener function, its strongly recommended to handle it directly. There are a few cases where
those errors may occur after your listener function has returned (such as when calling say()
or respond()
, or forgetting to call ack()
). In these
cases, your app will be notified about the error in an error handler function. Your app should register an error handler
using the Slapp.prototype.error(fn)
method.
app.error((error) => {
// Check the details of the error to handle special cases (such as stopping the app or retrying the sending of a message)
console.error(error);
});
NOTE: Specify the types of errors (corresponding to code
s), and the properties of each type.
If you do not attach an error handler, the app will log these errors to the console by default.
The app.error()
method should be used as a last resort to catch errors. It is always better to deal with errors
in the listeners where they occur because you can use all the context available in that listener.
If the app expects that using say()
or respond()
can fail, it's always possible to use app.client.chat.postMessage()
instead, which returns a Promise
that can be caught to deal with the error.
Apps are designed to be extensible using a concept called middleware. Middleware allows you to define functionality that processes a whole class of events before (and after) the listener function is called. This makes it easier to deal with common concerns in one place (e.g. authentication, logging, etc) instead of spreading them out in every listener.
In fact, middleware can be chained so that any number of middleware functions get a chance to run before the listener, and they each run in the order they are added.
Middleware are just functions - nearly identical to listener functions. They can choose to respond right away, to extend
the context
argument and continue, or trigger an error. The only difference is that middleware use a special next()
argument, a function that's called to let the app know it can continue to the next middleware (or listener) in the chain.
There are two types of middleware: global and listener. Each are explained below.
Global middleware are used to deal app-wide concerns, where each incoming event should be processed. They are added to
the app using app.use(middleware)
. You can add multiple middleware, and each of them will run before any of the listener
middleware, or the listener functions run. Remember, the order matters!
As an example, let's say your app can only work if the user who sends every incoming message is identified using an internal authentication service (e.g. an SSO provider, LDAP, etc). Here is how you might define a global middleware to make that data available to each listener.
// Authentication middleware - Calls Acme identity provider to associate the incoming event with the user who sent it
function authWithAcme({ payload, context, say, next }) {
const slackUserId = payload.user;
// Assume we have a function that can take a Slack user ID as input to find user details from the provider
acme.lookupBySlackId(slackUserId)
.then((user) => {
// When the user lookup is successful, add the user details to the context
context.user = user;
// Pass control to the next middleware (if there are any) and the listener functions
next();
})
.catch((error) => {
// Uh oh, this user hasn't registered with Acme. Send them a registration link, and don't let the middleware/listeners continue
if (error.message === 'Not Found') {
// In the real world, you would need to check if the say function was defined, falling back to the respond function if not,
// and then falling back to only logging the error as a last resort.
say(`I'm sorry <@${slackUserId}, you aren't registered with Acme. Please use <https://acme.com/register> to use this app.`);
return;
}
// This middleware doesn't know how to handle any other errors. Pass control to the previous middleware (if there
// are any) or the global error handler.
next(error);
});
}
app.use(authWithAcme);
// The listener now has access to the user details
app.message('whoami', ({ say, context }) => { say(`User Details: ${JSON.stringify(context.user)}`) });
Listener middleware are used to deal with shared concerns amongst many listeners, but not necessarily for all of them. They are added as arguments that precede the listener function in the call that attaches the listener function. This means the methods described in Listening for events are actually all variadic (they take any number of parameters). You can add as many listener middleware as you like.
As an example, let's say your listener only needs to deal with messages from humans. Messages from apps will always have a
subtype of bot_message
. We can write a middleware that excludes bot messages:
// Listener middleware - filters out messages that have subtype 'bot_message'
function noBotMessages({ message, next }) {
if (!message.subtype || message.subtype !== 'bot_message') {
next();
}
}
// The listener only sees messages from human users
app.message(noBotMessages, ({ message }) => console.log(
`(MSG) User: ${message.user}
Message: ${message.text}`
));
Subtype matching is so common, that Slapp ships with a listener middleware helper that filters all messages that match
a given subtype. The following is an example of the opposite of the one above - the listener only sees messages that are
bot_message
s.
const { Slapp, subtype } = require('slapp');
// Not shown: app initialization and start
// The listener only sees messages from bot users (apps)
app.message(subtype('bot_message'), ({ message }) => console.log(
`(MSG) Bot: ${message.bot_id}
Message: ${message.text}`
));
The examples above all illustrate how middleware can be used to process an event before the listener (and other middleware in the chain) run. However, some middleware are designed to process the event after the listener finishes. In general, a middleware can run both before and after the remaining middleware chain.
In order to process the event after the listener, a function is passed to next()
. The function recieves two arguments:
-
error
will be falsy when the middleware chain finished handling the event normally. When some later middleware callsnext(error)
(including an error parameter), then this argument is set to an error value. -
done
is a callback that must be called when processing is complete. When there is no error, or the incoming error has been handled,done()
should be called with no parameters. If instead the middleware is propigating an error up the middleware chain,done(error)
should be called with the error for its only parameter.
The following example shows a global middleware that calculates the total processing time for the middleware chain:
function logProcessingTime({ next }) {
const startTimeMs = Date.now();
next((error, done) => {
// This middleware doesn't deal with any errors, so it propogates any truthy value to the previous middleware
if (error) {
done(error);
return;
}
const endTimeMs = Date.now();
console.log(`Total processing time: ${endTimeMs - startTimeMs}`);
// Continue normally
done();
});
}
app.use(logProcessingTime)