Moved .exit before .enter, should fix #1460

[kum-deepak]

Possible fix of #1460.

[kum-deepak]

@Frozenlock please test it out, see if it resolves the issue.

[Frozenlock]

Yup!

[kum-deepak]

@gordonwoodhull for your feedback.

[gordonwoodhull]

I'm willing to merge this, but as I asked in #1460 (comment), I would still like to know why this works.

If selection.exit() has no side effects, and transition.remove() is delayed by the transitionDuration, how come this has any effect?

Deepening the mystery, Mike's General Update Pattern series of blocks have .remove() after enter and update... up until the last example, where transitions are introduced. So I think there may be something about the asynchronous nature of transitions which make this necessary.

[kum-deepak]

@gordonwoodhull it indeed is a mystery to me that how it works. If we consider the original code and see http://localhost:8888/web/examples/scatter-brushing.html, symbols do get removed. So, in general there is no issue in .exit and .remove.

@Frozenlock, I could not find a link to a fiddle to check the problem. Please share the link (preferably the filter one), I will dig deeper into it.

[Frozenlock]

@kum-deepak My apologies, but the gif is from a pretty complex setup in Clojurescript.
In fact I don't program in javascript. I can read, tweak a few things, but that's about it.

The simplest way to test this is probably to have a button with which you can remove an entry in the chart's group. @gordonwoodhull will probably know more about what is going on, as he correctly guessed I was using fake groups.

[gordonwoodhull]

Okay I'll try to test this.

The scatter-brushing escape doesn't exhibit the problem because no bins are actually removed - the values just go to zero and the chart reacts by setting the size to zero based on the existenceAccessor.

This is what you get with straight crossfilter, which is probably why we didn't have any Jasmine tests for it.

[gordonwoodhull]

Ah, okay, the answer is really trivial. I was totally overthinking it. 😊

It's just that using this coding pattern we replace the symbols variable. The result selection of .merge(), of course, has no exiting elements.

Think I understood this before and just forgot.

[gordonwoodhull]

Thanks @kum-deepak, released in 3.0.5

[gordonwoodhull]

(added a test too!)