Float32Arrays with `NaN`s as its first elements not plotting anymore on version > 1.6.30

[AdilRabbani]

Hi 👋

It seems like Float32Arrays with some NaN values at the start are no longer plotted after version 1.6.30. Just one value at the very start of the array is enough to break the plotted chart. It then shows up like this:

I've also created a fiddle here: https://jsfiddle.net/adilrabbani/0s6L7mhc/3/ (the second series has a NaN at first index)

[leeoniya]

i'm surprised that it ever worked. uPlot only checks the arrays for null and undefined using a single == null check per value. it would be quite expensive to and an extra check per value for NaN, though it we see that the data is a typed array we can potentially replace the null check with a NaN check 🤔

[AdilRabbani]

though it we see that the data is a typed array we can potentially replace the null check with a NaN check 🤔

This would be great if it could be added! Our data comes in as Float32Arrays and we would prefer to stick to it. Although we could just switch to a number array considering that we only have to do it for the values and not the time data...

[leeoniya]

i looked at this a bit and it's more involved, since this handling needs to be passed down to getMinMax functions and other places. i dont think it's worth it.

i remembered we already had this discussion, what has changed since then that my answer should be different now?

#974 (comment)

[AdilRabbani]

I decided to stick with Float32Arrays as the gaps can still be shown as NaNs in the tool-tip. Moreover, since we use that data later on to do some other calculations, I'm concerned there will be a performance hit using normal arrays (https://www.measurethat.net/Benchmarks/Show/33193/1/array-vs-float32array)

Normal arrays would also take double the amount of memory if I'm not mistaken? 🤔 (4 bytes vs 8 bytes per number)

[AdilRabbani]

If the performance hit is negligible for using Float32Arrays in uPlot and its mostly that the code gets more involved, I can volunteer to have a look at the code and make a PR? But yeah if there are other concerns as you mention it being not worth it then I'll just switch to normal arrays...