Chained promises not passing on rejection

Question

I am have a problem understanding why rejections are not passed on through a promise chain and I am hoping someone will be able to help me understand why. To me, attaching functionality to a chain of promises implies an intent that I am depending on an original promise to be fulfilled. It's hard to explain, so let me show a code example of my problem first. (Note: this example is using Node and the deferred node module. I tested this with Dojo 1.8.3 and had the same results)

var d = require("deferred");

var d1 = d();

var promise1 = d1.promise.then(
    function(wins) { console.log('promise1 resolved'); return wins;},
    function(err) { console.log('promise1 rejected'); return err;});
var promise2 = promise1.then(
    function(wins) { console.log('promise2 resolved'); return wins;},
    function(err) { console.log('promise2 rejected'); return err;});
var promise3 = promise2.then(
    function(wins) { console.log('promise3 resolved'); return wins;},
    function(err) { console.log('promise3 rejected'); return err;});
d1.reject(new Error());

The results of running this operation is this output:

promise1 rejected
promise2 resolved
promise3 resolved

Okay, to me, this result doesn't make sense. By attaching to this promise chain, each then is implying the intent that it will be dependant upon the successful resolution of d1 and a result being passed down the chain. If the promise in promise1 doesn't receive the wins value, but instead gets an err value in its error handler, how is it possible for the next promise in the chain to have its success function called? There is no way it can pass on a meaningful value to the next promise because it didn't get a value itself.

A different way I can describe what I'm thinking is: There are three people, John, Ginger, and Bob. John owns a widget shop. Ginger comes into his shop and requests a bag of widgets of assorted colours. He doesn't have them in stock, so he sends in a request to his distributor to get them shipped to him. In the mean time, he gives Ginger a rain check stating he owes her the bag of widgets. Bob finds out Ginger is getting the widgets and requests that he get the blue widget when she's done with them. She agrees and gives him a note stating she will. Now, John's distributor can't find any widgets in their supply and the manufacturer doesn't make them any more, so they inform John, who in turn informs Ginger she can't get the widgets. How is Bob able to get a blue widget from Ginger when didn't get any herself?

A third more realistic perspective I have on this issue is this. Say I have two values I want updated to a database. One is dependant on the id of the other, but I can't get the id until I have already inserted it into a database and obtained the result. On top of that, the first insert is dependant on a query from the database. The database calls return promises that I use to chain the two calls into a sequence.

var promise = db.query({parent_id: value});
promise.then(function(query_result) {
    var first_value = {
        parent_id: query_result[0].parent_id
    }
    var promise = db.put(first_value);
    promise.then(function(first_value_result) {
        var second_value = {
            reference_to_first_value_id: first_value_result.id
        }
        var promise = db.put(second_value);
        promise.then(function(second_value_result) {
            values_successfully_entered();
        }, function(err) { return err });
    }, function(err) { return err });
}, function(err) { return err });

Now, in this situation, if the db.query failed, it would call the err function of the first then. But then it would call the success function of the next promise. While that promise is expecting the results of the first value, it would instead get the error message from its error handler function.

So, my question is, why would I have an error handing function if I have to test for errors in my success function?

Sorry for the length of this. I just didn't know how to explain it another way.

UPDATE and correction

(Note: I removed a response I had once made to some comments. So if anyone commented on my response, their comments might seem out of context now that I removed it. Sorry for this, I am trying to keep this as short as possible.)

Thank you everybody who replied. I would like to first apologize to everybody for writing out my question so poorly, especially my pseudo code. I was a little too aggressive in trying to keep it short.

Thanks to Bergi's response, I think I found the error in my logic. I think I might have overlooked another issue that was causing the problem I was having. This is possibly causing the promise chain work differently than I thought it should. I am still testing different elements of my code, so I can't even form a proper question to see what I'm doing wrong yet. I did want to update you all though and thank you for your help.

Solution 1

To me, this result doesn't make sense. By attaching to this promise chain, each then is implying the intent that it will be dependant upon the successful resolution of d1 and a result being passed down the chain

No. What you are describing is not a chain, but just attaching all the callbacks to d1. Yet, if you want to chain something with then, the result for promise2 is dependent on the resolution of promise1 and how the then callbacks handled it.

The docs state:

Returns a new promise for the result of the callback(s).

The .then method is usually looked upon in terms of the Promises/A specification (or the even stricter Promsises/A+ one). That means the callbacks shell return promises which will be assimilated to become the resolution of promise2, and if there is no success/error handler the respective result will in case be passed directly to promise2 - so you can simply omit the handler to propagate the error.

Yet, if the error is handled, the resulting promise2 is seen as fixed and will be fulfilled with that value. If you don't want that, you would have to re-throw the error, just like in a try-catch clause. Alternatively you can return a (to-be-)rejected promise from the handler. Not sure what Dojo way to reject is, but:

var d1 = d();

var promise1 = d1.promise.then(
    function(wins) { console.log('promise1 resolved'); return wins;},
    function(err) { console.log('promise1 rejected'); throw err;});
var promise2 = promise1.then(
    function(wins) { console.log('promise2 resolved'); return wins;},
    function(err) { console.log('promise2 rejected'); throw err;});
var promise3 = promise2.then(
    function(wins) { console.log('promise3 resolved'); return wins;},
    function(err) { console.log('promise3 rejected'); throw err;});
d1.reject(new Error());

How is Bob able to get a blue widget from Ginger when didn't get any herself?

He should not be able. If there are no error handlers, he will just perceive the message (((from the distributor) from John) from Ginger) that there are no widgets left. Yet, if Ginger sets up an error handler for that case, she still might fulfill her promise to give Bob a widget by giving him a green one from her own shack if there are no blue ones left at John or his distributor.

To translate your error callbacks into the metapher, return err from the handler would just be like saying "if there are no widgets left, just give him the note that there are no ones left - it's as good as the desired widget".

In the database situation, if the db.query failed, it would call the err function of the first then

…which would mean that the error is handled there. If you don't do that, just omit the error callback. Btw, your success callbacks don't return the promises they are creating, so they seem to be quite useless. Correct would be:

var promise = db.query({parent_id: value});
promise.then(function(query_result) {
    var first_value = {
        parent_id: query_result[0].parent_id
    }
    var promise = db.put(first_value);
    return promise.then(function(first_value_result) {
        var second_value = {
            reference_to_first_value_id: first_value_result.id
        }
        var promise = db.put(second_value);
        return promise.then(function(second_value_result) {
            return values_successfully_entered();
        });
    });
});

or, since you don't need the closures to access result values from previous callbacks, even:

db.query({parent_id: value}).then(function(query_result) {
    return db.put({
        parent_id: query_result[0].parent_id
    });
}).then(function(first_value_result) {
    return db.put({
        reference_to_first_value_id: first_value_result.id
    });
}.then(values_successfully_entered);

Solution 2

@Jordan firstly as commenters noted, when using deferred lib, your first example definitely produces result you expect:

promise1 rejected
promise2 rejected
promise3 rejected

Secondly, even if it would produce output you suggest, it wouldn't affect execution flow of your second snippet, which is a bit different, more like:

promise.then(function(first_value) {
    console.log('promise1 resolved');
    var promise = db.put(first_value);
    promise.then(function (second_value) {
         console.log('promise2 resolved');
         var promise = db.put(second_value);
         promise.then(
             function (wins) { console.log('promise3 resolved'); },
             function (err) { console.log('promise3 rejected'); return err; });
    }, function (err) { console.log('promise2 rejected'); return err;});
}, function (err) { console.log('promise1 rejected'); return err});

and that, in case of first promise being rejected will just output:

promise1 rejected

However (getting to the most interesting part) even though deferred library definitely returns 3 x rejected, most of other promise libraries will return 1 x rejected, 2 x resolved (that leads to assumption you got those results by using some other promise library instead).

What's additionally confusing, those other libraries are more correct with their behavior. Let me explain.

In a sync world counterpart of "promise rejection" is throw. So semantically, async deferred.reject(new Error()) in sync equals to throw new Error(). In your example you're not throwing errors in your sync callbacks, you just returning them, therefore you switch to success flow, with an error being a success value. To make sure rejection is passed further, you need to re-throw your errors:

function (err) { console.log('promise1 rejected'); throw err; });

So now question is, why do deferred library took returned error as rejection?

Reason for that, is that rejection in deferred works a bit different. In deferred lib the rule is: promise is rejected when it's resolved with an instance of error, so even if you do deferred.resolve(new Error()) it will act as deferred.reject(new Error()), and if you try to do deferred.reject(notAnError) it will throw an exception saying, that promise can be rejected only with instance of error. That makes clear why error returned from then callback rejects the promise.

There is some valid reasoning behind deferred logic, but still it's not on par with how throw works in JavaScript, and due to that this behavior is scheduled for change with version v0.7 of deferred.

Short summary:

To avoid confusion and unexpected results just follow the good practice rules:

1. Always reject your promises with an error instances (follow rules of sync world, where throwing value that's not an error is considered a bad practice).
2. Reject from sync callbacks by throwing errors (returning them doesn't guarantee rejection).

Obeying to above, you'll get both consistent and expected results in both deferred and other popular promise libraries.

Comments

[Toilal]

When using angularJS with $q, the throw keyword is to replace with $q.reject(err).

[Michael Lorton]

To clean up @Toilal's remark, the preferred replacement to throw, is return $q.reject(err). throw will, I believe, still work; it is just much slower.