add smart fillto to enable logscale y-axis stacked barplot

[Moelf]

fix #4549

related to #3004 and refactored a bit of code

barplot([1,2,3,1,2,3], [1,2,3,1,2,3], stack=[1,1,1,2,2,2], color=[1,1,1,2,2,2], gap=0; axis=(; yscale=log))

Before

After

[Moelf]

Do I need to provide reference plots? is there anything else I should do in the meantime?

[asinghvi17]

A mention in the docstring about the source of the heuristic would be nice, but we (maintainers) have to update the reference images

[Moelf]

@asinghvi17 #3004 (comment) we discussed that here, I can't find the exact code now, ROOT does something depends on if minimum is >1 or <1, but approximately /2 in both cases.

[Moelf]

maybe reference this? https://github.com/root-project/root/blob/d1cfaefcbdefbfb966ed4e8ab694d0c766944b8e/js/modules/hist2d/TH1Painter.mjs#L144-L146

[Moelf]

Bump

[SimonDanisch]

Looks good to me! @jkrumbiegel, do you have any objections?

[jkrumbiegel]

I don't know, it's kind of too arbitrary for me with these "smart" fillto values, I also didn't like the other log one. But I also don't use these kinds of stacked barplots so I don't really mind either.

[Moelf]

it's kind of too arbitrary for me with these "smart" fillto values

I agree it's arbitrary but I think we also agree it shouldn't be empty, because then it's unusable.

A nonarbitrary choice is to fill down to eps(), but that in practice means users are now forced to set ylims!() for every single yscale=log plot, which I thought would be too annoying. Thus smart fillto is a midway solution.

[ffreyer]

A nonarbitrary choice is to fill down to eps(), but that in practice means users are now forced to set ylims!() for every single yscale=log plot, which I thought would be too annoying. Thus smart fillto is a midway solution.

Since Axis relies on boundingbox() now and boundingbox() processes transformations you can have `-Inf`` in transformed data with sane limits by adding a method. If that's closer to the optimal solution I could update the pr for that.

[Moelf]

does that mean user must set limits themselves? or you're saying you can implement this /2 heuristics there too?

[ffreyer]

The latter. You can implement the heurisitic for boundingbox so that Axis uses it for limits while using something like eps() for the actual data.

[Moelf]

in that case feel free to do that, and I very much appreciate it