Select processing of large groups of selects takes too long

[Westbrook]

Description

Not sure if this is ultimately an iron-list, or an array-selector, or a further up, issue, but selecting large groups of content, whether by the technique outlined in #124 or in #361 takes an inordinate amount of time. Explicitly, I'm seeing ~10s processing times for lists of 5000 items. I understand that's a lot of items, BUT that still seems crazy long.

My research points to this coming from the way selectItem applied code from array-selector When you follow that into the array-selector, you can see that it's pushing into it's selected array the subsequent property effects of which seems to be the cause of the processing delay.

Expected outcome

Selecting a large amount of items takes an virtually indistinguishably longer time than selecting one or a few.

Actual outcome

Selecting 5000 items at a time causes a ~10s block on the main thread.

Live Demo

https://jsbin.com/tobosaleki/1/edit?html,console,output

Steps to reproduce

1. Visit the JS Bin demo
2. Click 'select all'
3. Wait, see the window.performance based timing display on completion
4. Click 'select none'
5. Wait, a really small amount of time, see the window.performance based timing display on completion
6. Be in awe at the difference

Browsers Affected

• All

[Westbrook]

@keanulee any thoughts on how I might approach this problem in a way that would be enticing to the team?

[keanulee]

Not super familiar with array-selector or how the selection logic actually works, but with iron-list#2.0-preview, there's a new selectIndex method that will let you more efficiently select multiple items (see https://github.com/PolymerElements/iron-list/tree/2.0-preview#changes-in-v2). You can use iron-list#2.0-preview with Polymer 1.0 too (since it's "hybrid").

[Westbrook]

That is exciting, I'll have to look into this more deeply. However, a quick pass on testing that out yields only slightly better results why working with the "hybrid" version:

https://jsbin.com/buhewosuli/edit?html,console,output

And, while the test in this if is esoteric (to say the least), the comments seem to point out that in V1 I still can't get around using the underlying array-selector method from which the issue seems to originate:

https://github.com/PolymerElements/iron-list/blob/2.0-preview/iron-list.html#L1528

Am I reading that right? What's more, it seems that a full v2.0 demo actually has worse performance here:

https://jsbin.com/tanuzodire/1/edit?html,console,output

Still learning my polygit with v2, so I might have messed something up in that JS Bin, let me know if you see anything particularly weird there. I definitely have the move to Polymer 2.0 on my to-do list, but it's still too much of a moving target (especially re: these sorts of issues) to sell time on in to my office.

[keanulee]

In that case maybe it's worthwhile to add a selectItems method, but I don't want to add it to 1.0/master since that means it'll need to be ported to 2.0 (where Collection doesn't exist). Feel free to make a 2.0 version, but for it to be merged it'll have to be hybrid (work with Polymer 1 and 2).

[Westbrook]

I'll have to catch up on some of my 2.0-isms anyways, so here's as good a time as any. Let me see what I can do.

[Westbrook]

@keanulee I've got a branch coming together for this that's gonna rely on some PRs to the various Polymer branches, but I wanted to share the progress so far:

V1: https://jsbin.com/fiqanewevo/1/edit?html,console,output
V2: https://jsbin.com/wolilumeni/4/edit?html,console,output

In that the PR for this would rely on the referenced changed to the Polymer library, would it be better if I wait till those are accepted or should I go ahead and put in the iron-list PR now?

[keanulee]

You could put it in a PR now, but I don't think anyone would look at them until the Polymer ones are merged.

[dman777]

Any follow ups to this? I am hitting the same issue.

[Westbrook]

Sorry @dman777 I never got around to doing this work. The upward facing pressure on the Polymer library did outline a path forward, however. In conjunction with this issue, the conversation in #457 there should be more than enough information to pick up where I left off.