Feature request: allow function as argument value in `scale_*(limits=)`

[econandrew]

Request is to allow limits arg in e.g. scale_x_continuous to accept a function which takes the default limits and returns new limits, in parallel with the same functionality in breaks and labels.

At present it seems like the two options are (1) specify literal limits, which is inflexible (2) calculate the preferred limits outside of the ggplot statement, but that is messy and can result in redoing calculations already happening inside ggplot (e.g. for position="stack").

It looks like this may be a one line change in scale-.r's get_limits, but I don't know my way around well enough to be sure.

Motivating example follows.

library(ggplot2)


p <- ggplot(mtcars) + geom_point(aes(cyl,mpg))

p # But suppose we don't like the default y-axis breaks...

# ...we have a general mechanism in breaks= to modify them with a function...
breaker <- function(by) {
  function(limits) {
    low <- floor(limits[1]/by)*by
    high <- ceiling(limits[2]/by)*by
    seq(low, high, by)
  }
}

# ...but it doesn't override the limits so we get an (IMHO) ugly y axis where
# the data exceeds the range of the tick marks...
p + scale_y_continuous(breaks = breaker(10))

# In principle we can function-ise the limits in the same way
limiter <- function(by) {
  function(limits) {
    low <- floor(limits[1]/by)*by
    high <- ceiling(limits[2]/by)*by
    c(low, high)
  }
}

# But for now the best we can do is manually override the limits
p + scale_y_continuous(breaks = breaker(10), limits = limiter(10)(range(mtcars$mpg)))

# Feature request: if limits arg is a function, pass it the existing limits automatically
p + scale_y_continuous(breaks = breaker(10), limits = limiter(10))
#> Warning in is.na(self$limits): is.na() applied to non-(list or vector) of
#> type 'closure'
#> Error in rep(yes, length.out = length(ans)): attempt to replicate an object of type 'closure'

[has2k1]

Is using the expand parameter not sufficient?

[econandrew]

I don't think so. Following on the example above, you might want a y-scale that ends exactly at multiples of 10, e.g.

p + scale_y_continuous(breaks = breaker(10), limits = limiter(10)(range(mtcars$mpg)), expand = c(0, 0))

I don't see how you can achieve that in a general way using expand alone, since it can only add/multiply by a fixed amount. You could add on whatever increment was necessary to (in the example case) bring the scale up to exactly 40, but again that would require replicating the limits calculation outside of the ggplot2 to figure out what the increment would be - or just hard-coding it, e.g.

p + scale_y_continuous(breaks = breaker(10), expand = expand_scale(add = c(0, 6.1)))

The other advantage of setting the limits rather than expand is that it will feed into the default calculation of the breaks, so in my example above you actually don't need to specify the breaks, changing the limits changes the breaks too:

p + scale_y_continuous(limits = limiter(10)(range(mtcars$mpg)))

[hadley]

I like the idea in principle, but it feels it bit too different to breaks/labels since if you supply an argument to limits, the default calculation still has to take place. That said, I'd still review a PR.