Fix linetype regression

[teunbrand]

This PR aims to fix #6096.

Briefly, it replaces na.value = "blank", which casts mapped value to characters, to na.value = NA, which will cast NA to whatever type mapped values have.

[yutannihilation]

Looks good! I think this should be documented in NEWS.md as a (probably very minor) breaking change. IIUC, the following code was valid at v3.5.1. I think there are no or very little users who actually specify this, but just in case.

p + scale_linetype_manual(values = 1:5, na.value = "dotted")

[teunbrand]

I'm happy to add a news bullet.

I'm not sure I agree it is a breaking change. In 3.5.1, using a numeric palette with character na.value also raises errors (if there are actual NAs to map). The only reason it is exposed now is because now append na.value to the palette here, so we always need to be able to cast na.value to the palette's type.

library(ggplot2)

df <- economics_long
df$variable[100:105] <- NA

ggplot(df, aes(date, value01, linetype = variable)) +
  geom_line() + 
  scale_linetype_manual(values = 1:5, na.value = "blank")
#> Error in grid.Call.graphics(C_lines, x$x, x$y, index, x$arrow): invalid line type: must be length 2, 4, 6 or 8

^{Created on 2024-09-16 with reprex v2.1.1}

[yutannihilation]

if there are actual NAs to map

Aw, got it. Sorry, I didn't notice this difference and tried with a data without actual NA. Thanks for the explanation, that makes sense!

[teunbrand]

Thanks for the review!