Small optimizations

[kurkle]

• Only extend element if configuration is provided
• Time scale lookup, check index validity before accessing table[-1]

A 20 times sequential uPlot (parsing: false) comparison:
Master (average ~125ms/chart):

MR (average ~105ms/chart):

[benmccann]

It seems to me like this can only happen when lo and hi both equal 0. Is that right? Why do we need a special check for that? It would seem the code comment (given value is outside table (before first item)) no longer applies in that case

I think we should get rid of the interpolation stuff entirely, which would be much faster and simpler. We can instead check the distribution to be a lot more efficient. Something like:

getPixelForValue(value) {
	const me = this;
	if (distribution === 'series') {
		return getPixelForDecimal(me._timestamps.indexOf(value) / me._timestamps.length);
	} else {
		return getPixelForDecimal((value - me.max) / (me.min - me.max));
	}
}

[kurkle]

This happens when lo = 0 and hi = 1. (0 + 1) >> 1 = 0
indexOf would be a lot slower than the lookup table.

[kurkle]

Another problem with indexOf is that you have to have all the timestamps in place to be able to get a pixel for it.
For example:

data: [{x: 1577659670, y: 1}, {x: 1577659680, y: 2}]

getPixlForValue(1577659675) // undefined 

[benmccann]

Yeah, you're right that indexOf would be a O(n) lookup. We could use a Map instead to make it O(1). Especially since a Map is already being created (since #6871 - though we'd need the value to be the index rather than true).

I don't think that we'd ever call getPixelForValue for a value that's not in the dataset provided by the user, so I'd be fine with explicitly not supporting that as I don't see a use case for it.