scale_y_log10 Fails with a Very Large Range

[billdenney]

The second figure should not fail; log axes should be possible.

library(ggplot2)
#> Warning: package 'ggplot2' was built under R version 3.4.3
ggplot(data.frame(y = c(9.88131291682493e-324, 1000), x = 1:2), aes(x = x, y = y)) + 
  geom_line()

ggplot(data.frame(y = c(9.88131291682493e-324, 1000), x = 1:2), aes(x = x, y = y)) + 
  geom_line() + scale_y_log10()
#> Error in seq.default(min, max, by = by): 'from' must be a finite number

[mcol]

Most of what's going on is within the ScaleContinuous section of scale-.r.

When transform() is called on the original example, the points are 9.881313e-324 1.000000e+03 at entry, and their log10 is computed, turning them into -323.0052 3.0000.

Axis limits are a bit larger than this, becoming (-339.30544, 19.30026), and the call to self$trans$inverse(limits) changes it to 0.000000e+00 1.996454e+19. With these values, the call to self$trans$breaks(limits) fails because self$trans$breaks() needs to compute a log, which produces -Inf for 0.

[thomasp85]

@hadley, since this is a matter of log10_trans() not handling input gracefully I suggest we move this issue to scales.

There is a larger discussion to be had around scale expansions extending beyond what is meaningful for a scale, but that is slightly orthogonal to this I think (and a huge change)

[hadley]

@thomasp85 sounds good to me.