Secondary axis doesn't show axis ticks or labels with coord_trans

[anthonytw]

Exactly the same issue as #2072, which I take it was fixed since it is closed.

The secondary axis ticks and labels do not show up when coord_trans is present, even if using the identity transformation. The axis label ("name" parameter) does still appear.

For example, roughly taking from the examples in sec_axis:

ggplot(mtcars, aes(cyl, mpg)) + geom_point()  +
    scale_y_continuous(sec.axis = sec_axis(~.+10, name='MPG + 10'))

works as expected, but

ggplot(mtcars, aes(cyl, mpg)) + geom_point()  +
    scale_y_continuous(sec.axis = sec_axis(~.+10, name='MPG + 10')) +
    coord_trans()

loses the axis ticks and labels (but keeps the title "MPG + 10").

[clauswilke]

Reprex:

library(ggplot2)
ggplot(mtcars, aes(cyl, mpg)) + geom_point()  +
  scale_y_continuous(sec.axis = sec_axis(~.+10, name='MPG + 10'))

ggplot(mtcars, aes(cyl, mpg)) + geom_point()  +
  scale_y_continuous(sec.axis = sec_axis(~.+10, name='MPG + 10')) +
  coord_trans()

^{Created on 2018-11-09 by the reprex package (v0.2.1)}

@dpseidel Is this distinct from #2978?

[dpseidel]

It seems so. At least it predates my changes from #2796 and is present with version 3.0.0. I'll need to look into this further to be certain. Will do when I'm back at my desktop.

[yutannihilation]

This problem comes from here in CoordTrans$setup_panel_params(); out is redefined so the params for the secondary axis (out$sec.range, out$sec.major, out$sec.minor, etc) are lost.

ggplot2/R/coord-transform.r

Lines 192 to 195 in 868fdb7

	out <- list(
	range = out$range, labels = out$labels,
	major = out$major_source, minor = out$minor_source
	)

If I replace the code with below, at least the labels are shown (the locations are wrong, though).

  out$major <- out$major_source
  out$minor <- out$minor_source
  out$arrange <- scale$axis_order()

[ptoche]

If out$sec.major ends up having length 0, then the axis is removed following the logic explained here: #2983 (comment), I guess.

[dpseidel]

I played around with this a little on my branch containing my most recent sec.axis patch (#3040) and can confirm the misalignment is still an issue. Interestingly, it's unique to all the other misalignment issues. It seems that when coord_trans() is specified sec.axis is passed the unexpanded range -- the transformed range of the data, not the transformed range of the primary scale (as it is when no coord is specified). This leads to misalignment relative to the primary axis.

# loading local dpseidel/revert2796 + suggested coord_trans patch
library(ggplot2)

c <- ggplot(mtcars, aes(cyl, mpg)) +
  geom_point() +
  scale_y_continuous(sec.axis = sec_axis(~ . + 10, name = "MPG + 10")) +
  coord_trans()

A screenshot of output from within debug(get("break_info", AxisSecondary)) to demonstrate the actual break specifications calculated:

Notably, and likely a separate bug worth flagging in its own issue, the layer_scales function runs the break_info function with the expected, expanded, range even when passed coord_trans() and returns properly aligned break values; a bug I have encountered when debugging other sec.axis issues and which can give users a false reading of the position of tick axes (as it does below):

layer_scales(c)$y$break_info()
#> $range
#> [1] 10.4 33.9
#> 
#> $labels
#> [1] "15" "20" "25" "30"
#> 
#> $major
#> [1] 0.1957447 0.4085106 0.6212766 0.8340426
#> 
#> $minor
#> [1] 0.0893617 0.1957447 0.3021277 0.4085106 0.5148936 0.6212766 0.7276596
#> [8] 0.8340426 0.9404255
#> 
#> $major_source
#> [1] 15 20 25 30
#> 
#> $minor_source
#> [1] 12.5 15.0 17.5 20.0 22.5 25.0 27.5 30.0 32.5
#> 
#> $sec.range
#> [1] 20.4 43.9
#> 
#> $sec.labels
#> [1] "25" "30" "35" "40"
#> 
#> $sec.major
#> [1] 0.1957447 0.4085106 0.6212766 0.8340426
#> 
#> $sec.minor
#> [1] 0.0893617 0.1957447 0.3021277 0.4085106 0.5148936 0.6212766 0.7276596
#> [8] 0.8340426 0.9404255
#> 
#> $sec.major_source
#> [1] 25 30 35 40
#> 
#> $sec.minor_source
#> [1] 22.5 25.0 27.5 30.0 32.5 35.0 37.5 40.0 42.5

^{Created on 2018-12-16 by the reprex package (v0.2.1)}

Since this is a scale expansion issue, misalignment is fixed if one specifies expand = c(0,0) but we will likely want to dig deeper into this (and merge an appropriate coord_trans patch) once the sec.axis refactoring is complete.

[thomasp85]

@dpseidel do you have any idea how to handle this most gracefully?

[dpseidel]

I'll need to look back into it. I'll plan my own private dev day this weekend to dig into this and other open axis/scale issues 👍

[paleolimbot]

This seems related to #3338 in that coord_trans() is handling scale expansion differently than other coords.