Colourbar does not have all colors of the scale.

[has2k1]

For some data it is possible for the colourbar to cover a fraction of the colourspace smaller than that covered by the data. It can be confusing when the guide for a continuous scale does span the full mapped space of an aesthetic.

Actually, this bug is probably triggered every time there is a colourbar, but in most cases it is small and imperceptible. With colourmaps like viridis, the cases where it is obvious should only go up.

In the reprex below, I had to do some contriving (nbin=7) to reveal it.

library(ggplot2)

ggplot(faithful, aes(x = eruptions, y = waiting)) +
  geom_point()  +
  stat_density_2d(aes(fill=..level..), geom='polygon') +
  xlim(0.5, 6) +
  ylim(40, 110) +
  scale_fill_viridis_c() +
  guides(fill=guide_colorbar(nbin=7))

The root of the issue is pretty does not guarantee that the breaks it computes are close enough to the limits that the difference is not perceptible when mapped to a colour.

A simple but partial solution is to increase the default guide$nbin parameter. This would give pretty enough wiggle room to work with and reduce incidences of the bug.

A better solution is recognising that the "raster" colourbar does not need to be generated using 'pretty' breaks, a seq(.limits[1], .limits[2], length=guide$nbin) can do. For this solution the current value of nbin need not change except for the craziest of colourmaps.

[hadley]

Would you mind pointing me towards the code involved?

[has2k1]

ggplot2/R/guide-colorbar.r

Line 193 in 582acfe

.bar <- discard(pretty(.limits, n = guide$nbin), scale$get_limits())

[hadley]

Yeah, there doesn't seem to be any need for pretty() there. Do you want to do a PR?