teardownMap slows perfomance drastically

[wclr]

In map.js there is function teardownMap added to remove _cid on orignal object.

This code calculates performance time with 1000 objects converted to can.Map, and with teardownMap it 10 times slower than without it.

            var map = new can.Map(),g
                    objects = [];

            for (var i = 0; i < 100; i++){
                objects.push({prop: 'prop', nest: {prop: 'prop', nest: {prop: 'prop'}}})
            }

            var start = new Date()

            for (var j = 0; j < 100; j++){
                map.attr('obj', objects)
            }
            var time = (new Date() - start)/1000

            // time is ten times more with tearing down than without it
            console.log('time', time)

I could add test but not sure what to check, as performance is relative.

[justinbmeyer]

How are you removing it exactly?

Sent from my iPhone

On Dec 7, 2013, at 5:05 AM, Alex Osh notifications@github.com wrote:

In map.js there is function teardownMap added to remove _cid on orignal object.

This code calculates performance time item with 1000 objects converted to can.Map, and with teardownMap it 10 times slower than without it.

var map = new can.Map(),
    objects = [];

for (var i = 0; i < 1000; i++){
    objects.push({prop: 'prop'})
}

var start = new Date()

map.attr('obj', objects)

var time = (new Date() - start)/1000

// time is ten times more with tearing down than without it
console.log('time', time)    

I could add test but not sure what to check, as performance is relative.

—
Reply to this email directly or view it on GitHub.

[wclr]

You may just put return there to test, cycle itself (even empty) causes performance drop

        teardownMap = function(){
            return
            for(var cid in madeMap){
                if(madeMap[cid].added) {
                    delete madeMap[cid].obj._cid;
                }
            }
            madeMap = null;
        },

[daffl]

Could this be fixed via #604?

[wclr]

@daffl
No, I tried. I've updated my code in OP, if you try you should see the time difference (previous version was incorrect for the case and shown no difference). On my PC I have ~5 sec for version with teardownMap and ~0.1 sec for version without it.

Performance drop is significant when you have quite large array to be converted to can.List many times in a row.

I noticed that because in my application I started to experience performance issues after update from master, debugged it and found out that it was due to e767fc6 commit.

[justinbmeyer]

I've confirmed this is indeed more than 10x slower. It seems that just setting madeMap = null; is the slow part. I'm going to have to think over this one.

[justinbmeyer]

Ok, I figured this out. The problem is not teardownMap. The issue is that when teardownMap is removed, only 300 objects are created instead of 300000.

This is because by removing teardownMap it doesn't re-create that big List of maps, it just uses the first one created when when the big plain array is passed to .attr('obj', objects) for the first time.

Closing... however, I'm going to work on getting a benchmark workflow into canjs.

[wclr]

So what does this mean? After e767fc6 performance became worse obviously. If what is the a way to improve it?

[justinbmeyer]

Performance did not become worse. There was a bug that made it appear faster than it was. You could get the same speed as changing your above code to:

var map = new can.Map(),
      objects = [];

            for (var i = 0; i < 100; i++){
                objects.push({prop: 'prop', nest: {prop: 'prop', nest: {prop: 'prop'}}})
            }
            var objectList = new can.List(objects)
            var start = new Date()

            for (var j = 0; j < 100; j++){
                map.attr('obj', objectList)
            }
            var time = (new Date() - start)/1000

            // time is ten times more with tearing down than without it
            console.log('time', time)

This was added to solve #521