Trouble using Scale.continuous_scale_partial with the color aesthetic

[durcan]

When plotting bivariate histograms it is often desirable to use a log color scale so most of the dynamic range of the color scale is not used by a few of the bins. As an example, lets look at the "diamond" data-set and look at the carats vs price distribution:

using Lazy, Gadfly, RDatasets

d = dataset("ggplot2", "diamonds")

plot(d,
x=:Price,
y=:Carat,
Geom.hexbin(xbincount=70, ybincount=70)) |>
x->draw(SVG(15cm, 15cm), x)

produces:

Much of the structure of the long tail is hidden as the dynamic range is dominated by a few of the bins in the lower left of the plot. There is no Scale.color_log10, so using the definition as Scale.y_log10 as a rubric, I attempted to roll one myself:

color_log10 = Scale.continuous_scale_partial([:color], Scale.log10_transform) # define the scale

plot(d,
x=:Price,
y=:Carat,
color_log10, # use the new scale
Geom.hexbin(xbincount=60, ybincount=60)) |>
x->draw(PNG(15cm, 15cm), x)

This gives a somewhat cryptic error:

HexBinGeometry requires a continuous color scale.
while loading In[3], in expression starting on line 3

Especially given that color_log10 does seem to be a ContinuousScale constructor. Is this not the correct approach to modify color scales? I know that there are methods to produce entirely new scales (like Scale.ContinuousColorScale) but it is hard to make a sensible color scale when them (let alone one that is not linearly scaled). If there is a correct approach to this sort of problem it would be great to make it more accessible / documented.