ES6 promise settled callback?

问题:

I want to run the same action whether my Promise resolved successfully or not. I don't want to bind the same function to both args of .then. Isn't there a .always like jQuery has? If not, how do I achieve this?

回答1:

Isn't there a .always like jQuery has?

No, there's not (yet). Though there is an active proposal, so maybe ES2018.

If not, how do I achieve this?

You can implement the finally method yourself like this:

Promise.prototype.finally = function(cb) {
    const res = () => this
    const fin = () => Promise.resolve(cb()).then(res)
    return this.then(fin, fin);
};

or more extensively, with passing resolution information to the callback:

Promise.prototype.finally = function(cb) {
    const res = () => this
    return this.then(value =>
        Promise.resolve(cb({state:"fulfilled", value})).then(res)
    , reason =>
        Promise.resolve(cb({state:"rejected", reason})).then(res)
    );
};

Both ensure that the original resolution is sustained (when there is no exception in the callback) and that promises are awaited.

回答2:

With async/await, you can a combination of await with try/finally, like so:

async function(somePromise) {
  try {
    await somePromise();
  } finally {
    // always run this-- even if `somePromise` threw something
  }
}

Here's a real example I have running in production with Node, using Babel's async-to-generator plugin.

// Wrap promisified function in a transaction block
export function transaction(func) {
  return db.sequelize.transaction().then(async t => {
    Sequelize.cls.set('transaction', t);
    try {
      await func();

    } finally {
      await t.rollback();
    }
  });
}

I use this code inside a mocha test alongside the Sequelize ORM to start a DB transaction, and regardless of the outcome of the DB calls within the test, always roll back at the end.

This is roughly analogous to Bluebird's .finally() method, but IMO, far nicer syntax!

(Note: In case you're wondering why I'm not awaiting on the first Promise- it's an implementation detail of Sequelize. It uses CLS to 'bind' a SQL transaction to a Promise chain. Anything that incurs inside the same chain is scoped to the transaction. Anything outside isn't. Therefore, awaiting on the Promise would have 'closed' the transaction block and broken the chain. I threw this example in to show you how 'vanilla' Promise handling can be mixed alongside async functions, and play well together.)

回答3:

If you don't/can't update the prototype, the way to hack a finally is:

executeMyPromise()
.then(function(res){ return {res: res}; })
.catch(function(err){ return {err: err}; })
.then(function(data) {
    // do finally stuff
    if (data.err) {
        throw data.err;
    }
    return data.res;
}).catch(function(err) {
    // handle error
});

回答4:

Here is my implementation of .finally().

Promise.prototype.finally = function(cb) {
   return this.then(v=>Promise.resolve(cb(v)),
                    v=>Promise.reject(cb(v)));
};

I tested it:

(new Promise((resolve,reject)=>{resolve(5);})).finally(x=>console.log(x));  //5

(new Promise((resolve,reject)=>{reject(6);})).finally(x=>console.log(x));  //6

(new Promise((resolve,reject)=>{reject(7);}))
.then(x=>x,y=>y)
.catch(x=>{throw "error";}) 
.finally(x=>{console.log(x); throw "error"; return x;})  // 7
.then(x=>console.log(x),y=>console.log('e'));  //e
// Uncaught (in promise) undefined

回答5:

Summary:

We now also have access to Promise.prototype.finally(). This is a function which can be put on the promise chain as a last element to perform some cleanup. it works the following way when compared to Promise.then and Promise.catch:

• Promise.then only gets called when the promise is resolved (if you only put it the first argument callback function)
• Promise.catch only gets called when the promise is rejected
• Promise.finally always gets called when a promise is fullfilled, so both when the promise gets rejected or resolved.

Example:

let Prom = new Promise((res, rej) => {
  let random = Math.random();
  
  if (random > 0.5) {
    res(1);
  } else {
    rej('Error occured')
  }

  
});


Prom.then((val) => {
  console.log(val);
  return val * 10;
}).catch((err) => {
  console.log(err);
}).finally(() => {
  console.log('finally executed');
})

In the above exapmle we can observe that finally is always executed regardless if the promise resolves or rejects. Not that finally should ideally always be at the end of the promise chain to do some cleanup which should be executed regardless of Promise outcome.

The advantage of using finally is that it avoids the need for code duplication because it is executed for both a resolved and a rejected promise. Otherwise we would have to use hacks like:

.then(onfullfilled, onfullfilled)

or

.then(onfullfilled)
.catch(onfullfilled)

Note that now we have to define the onfullfilled function as a named function outside the promisehandler itself (or pass in 2 anonymous function copies which is even less elegant). Promise.finally solves this problem for us.

回答6:

No need to introduce new concepts

const promise = new Promise((resolve, reject) => {
  /*some code here*/
});

promise.then(() => {
  /* execute success code */
}, () => {
  /* execute failure code here */
}).then(() => {}, () => {}).then(() => {
  /* finally code here */
});

回答7:

To extend Bergi answer.

Returning Promise.reject() in catch handler will prevent finnalizing 'then' to be called.

So if you going to handle promise error 2+ times you should use boilerplate like this:

return myPromise()
.then(() => ... )
.catch((error) => {
  ...
  myFinnaly();
  return Promise.reject(error);
})
.then(() => myFinnaly());