Fix hover filter to display close period points

[archmoj]

Fix for #5292 the hover part.
Please note that the spikeline part was addressed in #5542.

@plotly/plotly_js

[nicolaskruchten]

Some high-level feedback on the behaviour here:

1. we would like to keep the invariant we currently have that "whatever points show up in the one x unified hoverlabel are the same points that get a hoverlabel in x mode", which I don't believe is uniformly true with this PR
2. when the scatter points are in period mode with periodalignment = end we would like the hover calculation to be made based on the data coordinate of the point rather than the pixel coordinate.
3. I could see a case for saying that if I have bars in period=m3 mode, and scatter points (in any mode) on the first of the month, then I should not be able to see a hoverlabel with a Jan 1 point and a bar that represents Q4 i.e. the last three months of the previous year: Jan 1 is not in Q4.

[nicolaskruchten]

We should also remember to test this against grouped bars.

[archmoj]

1. we would like to keep the invariant we currently have that "whatever points show up in the one x unified hoverlabel are the same points that get a hoverlabel in x mode", which I don't believe is uniformly true with this PR

x hovermode does not appear to be perfect IMHO. See this codepen. Should we log a bug for "bars and scatters don't show together in x hovermode when hovering over middle or right side of the bars"?

[archmoj]

1. we would like to keep the invariant we currently have that "whatever points show up in the one x unified hoverlabel are the same points that get a hoverlabel in x mode", which I don't believe is uniformly true with this PR

x hovermode does not appear to be perfect IMHO. See this codepen. Should we log a bug for "bars and scatters don't show together in x hovermode when hovering over middle or right side of the bars"?

In addition this codepen may illustrate another bug concerning x hovermode and period alignements.

[nicolaskruchten]

Re compare mode (aka hovermode='x') please see #5553 and #5554 for a description of the desired behaviours, which could be implemented either in this PR or a different one.

[archmoj]

1. we would like to keep the invariant we currently have that "whatever points show up in the one x unified hoverlabel are the same points that get a hoverlabel in x mode", which I don't believe is uniformly true with this PR

Good call. Addressed in 1e59d89.

[alexcjohnson]

I see one bit of problematic behavior: If I make the plot from your test and zoom in a bit, it's possible to get the spikelines and hover labels out of sync. Oddly enough the behavior is somewhat reversed between x and x unified but both are problematic:

[archmoj]

I see one bit of problematic behavior: If I make the plot from your test and zoom in a bit, it's possible to get the spikelines and hover labels out of sync. Oddly enough the behavior is somewhat reversed between x and x unified but both are problematic:

Thanks for the comment. It seems like the different bug fixed by the other PR #5542.
So I merged master into this branch in 4d831e2.

[alexcjohnson]

Excellent! Good catch that the problem I mentioned was just another flavor of the one you already solved 🎉
💃