Canceling a reduction

[verma]

What would be the general strategy for canceling a Promise.reduce call? The promise itself cancels but the the current promise being waited upon for fulfillment is not notified of this cancellation.

E.g. how can I cancel this reduction?:

https://github.com/verma/plasio/blob/master/js/ui.js#L614

The fDataLoader and loadData methods are cancelable promises.

[petkaantonov]

The promise returned by Promise.reduce cannot be chained with the promises you return from the callback, so if you want to be able to cancel it, you can't use .reduce.

This should work however:

// do a progress function based on which file we're processing
var pfuncDataLoad = function(p, msg) {
    progress((currentLoadIndex + p*0.5) / maxLoadIndex, msg);
};

var pfuncDecompress = function(p, msg) {
    progress((currentLoadIndex + 0.5 + p*0.5) / maxLoadIndex, msg);
};

var soFar = [];
var cur = Promise.resolve().cancellable();

files.forEach(function(fname) {
    cur = cur.then(function() {
        return fDataLoader(fname, pfuncDataLoad).then(function(data){
            return loadData(data, pfuncDecompress);
        })
        .then(function(r) {
            var ret = {
                header: r[0],
                batcher: r[1]
            };

            // TODO: This needs to be fixed for mutliple URLs
            //
            ret.header.name = fname.name || name;
            currentLoadIndex ++;
            sofar.push(ret);
        })
    });
});

var loaderPromise = cur.then( ... )... rest of the code

[petkaantonov]

The promise returned by Promise.reduce cannot be chained with the promises you return from the callback, so if you want to be able to cancel it, you can't use .reduce.

This should work however:

// do a progress function based on which file we're processing
var pfuncDataLoad = function(p, msg) {
    progress((currentLoadIndex + p*0.5) / maxLoadIndex, msg);
};

var pfuncDecompress = function(p, msg) {
    progress((currentLoadIndex + 0.5 + p*0.5) / maxLoadIndex, msg);
};

var soFar = [];
var cur = Promise.resolve().cancellable();

files.forEach(function(fname) {
    cur = cur.then(function() {
        return fDataLoader(fname, pfuncDataLoad).then(function(data){
            return loadData(data, pfuncDecompress);
        })
        .then(function(r) {
            var ret = {
                header: r[0],
                batcher: r[1]
            };

            // TODO: This needs to be fixed for mutliple URLs
            //
            ret.header.name = fname.name || name;
            currentLoadIndex ++;
            sofar.push(ret);
        })
    });
});

var loaderPromise = cur.then( ... )... rest of the code

[verma]

Thanks for your response and help.

I tried your solution every which way, the weird thing is that it works sometimes while other times it doesn't. The Promise.cancel is invoked but it doesn't go anywhere. I mean I don't see it being caught or raising an error anywhere.

Could be something funky I am doing. I am also uncertain on how to help you reproduce this issue.

[verma]

Thanks for your response and help.

I tried your solution every which way, the weird thing is that it works sometimes while other times it doesn't. The Promise.cancel is invoked but it doesn't go anywhere. I mean I don't see it being caught or raising an error anywhere.

Could be something funky I am doing. I am also uncertain on how to help you reproduce this issue.

[verma]

So I took a closer look at the cancellation logic.

When we do Promise.resolve().cancellable(); we create a promise which is cancelable and already resolved. So at the top of the chain this.isCancelable() returns false since the promise is already resolved.

Promise.prototype.isCancellable = function Promise$isCancellable() {
    return !this.isResolved() &&
        this._cancellable();
};

At each level, to branch up to the parent, the code does this:

if (!this.isCancellable()) return this;
var parent;
if ((parent = this._cancellationParent) !== void 0) {
    parent.cancel(SYNC_TOKEN);
    return;
}

https://github.com/petkaantonov/bluebird/blob/master/js/browser/bluebird.js#L309

I feel that the top node (lets say C0) and the one below it (C1) have a little bit of a problem with this. C1 checks if its _cancellationParent is not null (which it isn't) and branches into the condition, C0.cancel(...) returns right away since C0.isCancellable() is false (since it is already resolved), which causes the control to come back to C1, which returns right away as well. This patterns repeats all the way to the bottom and promise fails to cancel (i.e. call _rejectUncheched function).

Still investigating, just wanted to update you. I am not entirely sure what's going on, I have very little knowledge of the whole thing.

Seems to me like the the C1 node here should have had its _cancellationParent set to null.

[verma]

So I took a closer look at the cancellation logic.

When we do Promise.resolve().cancellable(); we create a promise which is cancelable and already resolved. So at the top of the chain this.isCancelable() returns false since the promise is already resolved.

Promise.prototype.isCancellable = function Promise$isCancellable() {
    return !this.isResolved() &&
        this._cancellable();
};

At each level, to branch up to the parent, the code does this:

if (!this.isCancellable()) return this;
var parent;
if ((parent = this._cancellationParent) !== void 0) {
    parent.cancel(SYNC_TOKEN);
    return;
}

https://github.com/petkaantonov/bluebird/blob/master/js/browser/bluebird.js#L309

I feel that the top node (lets say C0) and the one below it (C1) have a little bit of a problem with this. C1 checks if its _cancellationParent is not null (which it isn't) and branches into the condition, C0.cancel(...) returns right away since C0.isCancellable() is false (since it is already resolved), which causes the control to come back to C1, which returns right away as well. This patterns repeats all the way to the bottom and promise fails to cancel (i.e. call _rejectUncheched function).

Still investigating, just wanted to update you. I am not entirely sure what's going on, I have very little knowledge of the whole thing.

Seems to me like the the C1 node here should have had its _cancellationParent set to null.

[petkaantonov]

You are right, can you check if b3a5a0d works for you as well?

[petkaantonov]

You are right, can you check if b3a5a0d works for you as well?

[petkaantonov]

Err I mean d818ff3

[petkaantonov]

Err I mean d818ff3

[verma]

Yes, seems to be working great! Thanks!

[verma]

Yes, seems to be working great! Thanks!

[petkaantonov]

It is now included in the version 1.2 release

[petkaantonov]

It is now included in the version 1.2 release