Contour plot interpolation method always treats data as linear

[andrew-matteson]

The implementation of the Catmull-Rom interpolation treats the data as linear. This results in very poor contours for plots with log-scale axes.

The linked example is a 5x5 contour generated in two different ways, with data that has x and y log-spaced. Code pen example: https://codepen.io/andrew-matteson/pen/oNWKJJZ

In the first plot, the x and y data is transformed to a linear scale by taking the log base 10. Ticks are managed with tick val / tick text. The contour data looks as expected.

In the second plot, the x and y data are taken directly, and the x and y axes are of type: 'log'. This produces nice axes, but very poor contours.

Screen shots:

In the absence of axis dependent interpolation, a workaround to get the nice "log style" axes with the nice contours would be appreciated.