Some values returned by `bquote()` trigger errors (3.5.1.9000)

[aphalo]

@teunbrand
Thanks for your dedication to 'ggplot2'. I can see lots of nice and useful enhancements coming!

The new support in (== 3.5.1.9000) of lambdas as arguments to paprameter name of scales breaks support for some calls as returned by bquote(). Expressions still work as before, as works converting the value returned by bquote() with as.expression(). The culprit seems to be specific tilde characters triggering an attempt to interpret the call as a lambda or formula. First reprex below is for 3.5.1.9000 from R-Universe (183e7ad) and the second for 3.5.1 from CRAN. (I haven't tested all possibilities, but hopefully this reprex provides a hint of what could be the problem.)

Additionally 'gganimate' supports calls but not expressions. See 'gganimate' issue #590.

3.5.1.9000

library(ggplot2)
packageVersion("ggplot2")
#> [1] '3.5.1.9000'

ggplot(mtcars, aes(factor(cyl), mpg)) + 
  geom_boxplot() + 
  scale_y_continuous(name = expression("Efficiency "~~(mi~gl^{-1})))

ggplot(mtcars, aes(factor(cyl), mpg)) + 
  geom_boxplot() + 
  scale_y_continuous(name = bquote("Efficiency"*(mi~gl^{-1})))

ggplot(mtcars, aes(factor(cyl), mpg)) + 
  geom_boxplot() + 
  scale_y_continuous(name = bquote("Efficiency"~(mi~gl^{-1})))
#> Error in `allow_lambda()`:
#> ! Can't convert `x`, a two-sided formula, to a function.

eff <- "Efficiency"
ggplot(mtcars, aes(factor(cyl), mpg)) + 
  geom_boxplot() + 
  scale_y_continuous(name = bquote(.(eff)~(mi~gl^{-1})))
#> Error in `allow_lambda()`:
#> ! Can't convert `x`, a two-sided formula, to a function.

^{Created on 2025-03-27 with reprex v2.1.1}

3.5.1

library(ggplot2)
packageVersion("ggplot2")
#> [1] '3.5.1'

ggplot(mtcars, aes(factor(cyl), mpg)) + 
  geom_boxplot() + 
  scale_y_continuous(name = expression("Efficiency "~~(mi~gl^{-1})))

ggplot(mtcars, aes(factor(cyl), mpg)) + 
  geom_boxplot() + 
  scale_y_continuous(name = bquote("Efficiency"*(mi~gl^{-1})))

ggplot(mtcars, aes(factor(cyl), mpg)) + 
  geom_boxplot() + 
  scale_y_continuous(name = bquote("Efficiency"~(mi~gl^{-1})))

eff <- "Efficiency"
ggplot(mtcars, aes(factor(cyl), mpg)) + 
  geom_boxplot() + 
  scale_y_continuous(name = bquote(.(eff)~(mi~gl^{-1})))

^{Created on 2025-03-27 with reprex v2.1.1}

[teunbrand]

Thanks for the report with the examples! I can reproduce this on my end. I don't think ggplot2 advertises somewhere that bquote() are inputs for scale names, and thus I'm unsure what degree of commitment ggplot2 has towards preserving this behaviour. I think if it is easy to preserve, we might as well try.

[teunbrand]

I think the trouble is coming from allow_lambda() that appears to recognise bquote() results as a formula.

x <- bquote("Efficiency"~(mi~gl^{-1}))
rlang::is_formula(x)
#> [1] TRUE

^{Created on 2025-03-27 with reprex v2.1.1}

The fix might be very simple by just checking for a one-sided (LHS) formula instead of any-sided formula.

Is there ever a reason to use bquote() with an expression that doesn't have both sides of the ~ operator (i.e. does it ever look like a single-sided formula)?

[aphalo]

The current situation is that bquote() is the only approach supported by 'gganimate'. I think this is because the labels are recomputed/updated for each frame in the animation. This is a useful behaviour, I think. Having different requirements in 'ggplot2' and 'gganimate' seems problematic to me.
It should be possible (easy?) to test for the presence of a lhs. I see you have already implemented this!
Thanks!

[teunbrand]

I see you have already implemented this!

I'm going to keep the issue open to keep track of it, in case #6385 doesn't get merged :)