`geom_sf()`: add support for `scale_x_continuous(n.breaks = )`

[pat-s]

Currently this argument does not have an effect when using geom_sf().

library("ggplot2")
nc <- sf::st_read(system.file("shape/nc.shp", package = "sf"), quiet = TRUE)
ggplot(nc) +
  geom_sf(aes(fill = AREA)) +
  scale_x_continuous(n.breaks = 2)

^{Created on 2021-09-18 by the reprex package (v2.0.1)}

[thomasp85]

@clauswilke do you have a rough idea of why coord_sf() behaves like this?

[clauswilke]

I'll take a look.

[thomasp85]

Thanks... My guess is it comes down to sf, but you never know...

[clauswilke]

I think it's these two lines:

ggplot2/R/coord-sf.R

Lines 226 to 227 in f5e01ba

	lat = scale_y$breaks %|W|% NULL,
	lon = scale_x$breaks %|W|% NULL,

We're reading out the value of the breaks element directly instead of calling get_breaks():

ggplot2/R/scale-.r

Line 636 in f5e01ba

get_breaks = function(self, limits = self$get_limits()) {

But I don't fully understand how the other coords use this code. I believe they call guide_train.axis() which then calls get_breaks():

ggplot2/R/guides-axis.r

Lines 64 to 67 in f5e01ba

	guide_train.axis <- function(guide, scale, aesthetic = NULL) {
	aesthetic <- aesthetic %||% scale$aesthetics[1]
	breaks <- scale$get_breaks()

but my problem is that I see how coord_cartesian() does this but not how coord_polar() does this. Yet things seem to work for coord_polar().

Maybe @paleolimbot can weigh in?

[paleolimbot]

I'm far removed from this to have a quick answer other than that I recall that the intention was to use the guide axis framework for the other Coord subclasses once the bugs had been ironed out for CoordCartesian. Perhaps this is a good time to do that for CoordSf? I'm nearly at a place where I could take that on but it will be a few weeks.

[clauswilke]

But do I understand correctly that the guide axis framework is not needed to get the axis breaks right? So we could just fix the two lines in coord_sf() and be done with it?

I would suspect that coord_sf() is a particularly challenging coord to move to the guide axis framework, as it does a lot of non-standard things such as turning projected coordinates into long/lat values for labels.

[paleolimbot]

Yes, likely much easier! $get_breaks() is probably what should be used.

[hadley]

I think the concern with using get_breaks() is that the default break computation used by sf might be much better. I don't have any evidence for this, so it might be worth doing a little spelunking in the sf code to figure out exactly what it's doing so we can make sure our defaults match.

[teunbrand]

Some additional limitations:

1. setting breaks = NULL does not remove gridlines or axis.

library(ggplot2)

nc <- sf::st_read(system.file("shape/nc.shp", package = "sf"), quiet = TRUE)

ggplot(nc) +
  geom_sf() +
  scale_x_continuous(breaks = NULL)

2. Providing a function to breaks does not work.

ggplot(nc) +
  geom_sf() +
  scale_x_continuous(breaks = scales::breaks_width(1))
#> Error in min(lon): invalid 'type' (closure) of argument

^{Created on 2023-09-26 with reprex v2.0.2}