Make autoSkip aware of major ticks

[benmccann]

This does the following:

• Skip major ticks only if they can't all fit
• If showing all major ticks, then skip minor ticks such that they're evenly distributed between major ticks
• Remove much of the time scale tick auto generation by generating every tick and then auto-skipping instead. This is so that we don't have two different "auto" algorithms to maintain. Also, by putting it in auto-skip it is available to all scales instead of just one. Thanks @simonbrunel for this suggestion

Interactive examples:

• This PR: https://plnkr.co/edit/REeume?p=preview
• Master: https://plnkr.co/edit/O9FincnNAZVmwtDA1mL9?p=preview

Closes #4600 #4612

[kurkle]

Couple of comments. Didn't do a through review. Can we have this in a pen/fiddle to test out?

[benmccann]

Can we have this in a pen/fiddle to test out?

I'm trying out a new one - plnkr :-) I added a link in the description

[kurkle]

Seems to work better than master in most cases.
week unit does not work well though:
PR:

Master:

[benmccann]

Thanks for testing this @kurkle! I think the issue is actually in the way that the sample was modified to chart week data. The PR seems to handle week fine in my testing, but the sample wasn't written with that unit in mind. If you just add week to the dropdown then it does what you pictured. But that's not quite valid because it will generate major ticks in strange places since the sample uses a custom afterBuildTicks which doesn't support weeks. Also, week is a unit that isn't auto-selected, so the unit is chosen as day and the major unit as month, which also breaks the custom afterBuildTicks in this sample since it assumes the data is generated daily when the unit is day.

Here's a slightly modified sample that works with week: https://plnkr.co/edit/jg9aAZ?p=preview

[kurkle]

There is a regression in tick label overlapping:

Quite valid use case IMO, one year, unit=month, maxRotation=0. It does not overlap always.

[kurkle]

Too many tests were changed IMO.
If "all" ticks are generated and then skipped, then tests should be altered to check for non skipped ones and the result should be quite the same in most cases.

[benmccann]

Thanks for the testing @kurkle. I've fixed the regression. The auto-skipper calculated to keep every 4th day, which was correct. The issue was in the way I changed the generate function in the time scale. It was generating a tick on every day, which was incorrect. Since it was a distribution: 'series' chart and the sample skipped weekends, it shouldn't have generated ticks on weekends. I've updated it to only generated ticks on days that have data points.

[kurkle]

Just a quick note about another indexOf. Will review in more depth later.

[benmccann]

Too many tests were changed IMO. If "all" ticks are generated and then skipped, then tests should be altered to check for non skipped ones and the result should be quite the same in most cases.

The tests that were changed were mostly tests for buildTicks. In these changed tests, only the tick generation is being tested and not the auto-skipping.

[benmccann]

I'm closing this temporarily. It has a performance regression since it generates more ticks. I will open a new version of this PR soon without the performance regression