Very blunt and crude, but maybe does the job!

[hughjonesd]

This seems to work fine for some basic tests:

dfr <- data.frame(a=rnorm(10), b = rnorm(10), grp = sample(1:3, 10, rep=T))

plot(dfr$a, dfr$b, col = viridis::magma(5), pch = 19)

pal2 <- saturation(viridis::magma, .8)
plot(dfr$a, dfr$b, col = pal2(5), pch = 19)

ggplot(dfr, aes(a,b,color=factor(grp))) + geom_point() + 
      saturation(scale_color_brewer(), 0.15)

Needs opacity fixing and some tests writing.

[jonclayden]

Great! I think this is a cleaner approach – thanks. There are some subtleties to resolve: I notice from ?continuous_scale and ?discrete_scale that the palette function behaves differently in each case, so that should be handled properly. I also think creating a ggproto subclass, as you did before, is a better bet than directly replacing the palette function. But I'm happy to merge this and then take it from here, unless you have anything to add...?

[hughjonesd]

Fine by me. I guess one question to answer is, do you want to extend this to mix(function, “red”) et cetera?... the only obstacle seems to be deciding what colorspace a function lives in....

[jonclayden]

True. Although scales does interpolation in Lab space, I think that the colours it generates are ultimately standard R colours, and so they're in sRGB space by default. For other functions, I think it's reasonable to put the onus on the function author to handle the space of its return values.

As for mixtures and complements, I'll look at the feasibility of supporting functions there, but I might end up leaving it for now, as the internals are a bit different.