Allow scale_{x/y}_sqrt() to have breaks at 0.

DESCRIPTION

@@ -41,7 +41,7 @@ Imports:
     MASS,
     mgcv,
     rlang (>= 1.0.0),
-    scales (>= 0.5.0),
+    scales (>= 1.2.0),
     stats,
     tibble,
     withr (>= 2.0.0)

NEWS.md

@@ -1,5 +1,9 @@
 # ggplot2 (development version)
 
+* Automatic break calculation now squishes the scale limits to the domain
+  of the transformation. This allows `scale_{x/y}_sqrt()` to find breaks at 0   
+  when appropriate (@teunbrand, #980).
+
 * Using multiple modified aesthetics correctly will no longer trigger warnings. 
   If used incorrectly, the warning will now report the duplicated aesthetic 
   instead of `NA` (@teunbrand, #4707).

R/scale-.r

@@ -638,6 +638,10 @@ ScaleContinuous <- ggproto("ScaleContinuous", Scale,
       return(numeric())
     }
 
+    # Ensure limits don't exceed domain (#980)
+    domain <- self$trans$transform(self$trans$domain)
+    limits <- oob_squish(limits, domain)
+
     # Limits in transformed space need to be converted back to data space
     limits <- self$trans$inverse(limits)
 

tests/testthat/test-scales-breaks-labels.r

@@ -367,3 +367,15 @@ test_that("functional limits work for continuous scales", {
     ggplot(mpg, aes(class)) + geom_bar(aes(fill = drv)) + scale_y_continuous(limits = limiter(50))
   )
 })
+
+test_that("limits are squished to transformation domain", {
+  # Breaks should not be calculated on ranges outside domain #980
+  sc1 <- scale_x_sqrt()
+  sc2 <- scale_x_sqrt()
+
+  sc1$train(c(0, 10))
+  sc2$train(c(-10, 10))
+
+  expect_equal(sc1$get_breaks(), sc2$get_breaks())
+  expect_equal(sc2$get_breaks()[1], 0)
+})