Box plot with log axis results in log-scaled box widths

[jonmmease]

Originally reported in plotly/plotly.py#1415

When plotting a box plot on a log axis, the box widths seem to follow the log pattern.

CodePen: plotly/plotly.py#1415

var y0=[],y1=[]
for ( i = 0; i < 50; i ++) 
{
	y0[i] = Math.random();
	y1[i] = Math.random();
}

var trace1 = {
	x: y0.map(v => 1),
   y: y0,
  type: 'box'
};

var trace2 = {
  x: y0.map(v => 10),
  y: y1,
  type: 'box'
};

var trace3 = {
  x: y0.map(v => 100),
  y: y1,
  type: 'box'
};

var data = [trace1, trace2, trace3];
var layout = {xaxis: {type: 'log'}}

Plotly.newPlot('myDiv', data, layout, {showSendToCloud: true});

Since the width of the boxes is not a value in data coordinates, it seems to me that we shouldn't scale it according to the log axis transform. What do you think?

[KonradTUM]

It would be really useful to input a box width in pixel. So it would be possible to output plots with different axis scales but same look. Don't know how to achieve this at the moment.

[etpinard]

@KonradTUM

It would be really useful to input a box width in pixel.

Maybe try the new box width attribute added in #3234

[etpinard]

Thanks @jonmmease I never tried plotting a box trace on a log position axis. Good to know!

[s417-lama]

I fixed this issue in s417-lama#1 .
Would you review the PR? @etpinard