I like the idea. One more thing to consider is how to handle make_sec_title(). I think we can just assign the same as make_title().

I saw that but didn't know what it did, so I didn't dare to touch anything. Is that for the second axis, if there is one?

On this topic: For any new function you add, or any related existing function where you figure out what it does, please add a few comments to the code in scale-.r so in the future we won't have these questions about what certain parts of the API actually do.

Is that for the second axis, if there is one?

Yes, I guess so, but I too don't know well about the Scale-related things. Honestly, I have no idea what is the case when there's no secondary.axis but the plot needs a secondary title...

The mechanism was added by #1868, so we might find some clue there.

Well, maybe @thomasp85 remembers what he did there. :-)

I think I got it. I need to add a similar trick here:

With that change, now it works also for secondary axis.

library(ggplot2)

df <- expand.grid(X1 = 1:10, X2 = 1:10)
df$value <- df$X1 * df$X2
p1 <- ggplot(df, aes(X1, X2)) + geom_tile(aes(fill = value))


p1 + scale_x_continuous(~ base::tolower(.x), 
                        sec.axis = sec_axis(trans = identity, 
                                            name = ~ gsub("x", "xx", .x))) +
  scale_fill_continuous(~ base::toupper(.x))

^{Created on 2021-01-11 by the reprex package (v0.3.0)}

How should I test this new functionality? My default idea is to create visual tests, but there must be a more specific way.

@clauswilke you think too highly of my memory 😄

But I'm sure I had my reasons...

@eliocamp are you in the process of preparing a PR?

and do your changes also work with p + labs(x = ~toupper(.x)), i.e. when setting label names outside of the scale?

and do your changes also work with p + labs(x = ~toupper(.x)), i.e. when setting label names outside of the scale?

It doesn't, and for it to fully work I'd need to change a lot of how labs work, I think.

Right now labs() changes the plot$labels variable, which is a character vector, so it cannot contain a function. A workaround is to apply the function to the label during update_labels() here:

However, if labs() is called before the relevant aesthetic is present, it doesn't work:

library(ggplot2)
df <- expand.grid(X1 = 1:10, X2 = 1:10)
df$lowercase <- df$X1 * df$X2

ggplot(df, aes(X1, X2)) + 
  geom_tile(aes(fill = lowercase)) +
  labs(fill =  ~base::toupper(.x))  

ggplot(df, aes(X1, X2)) + 
  labs(fill =  ~base::toupper(.x))  +
  geom_tile(aes(fill = lowercase)) 

^{Created on 2021-03-24 by the reprex package (v1.0.0)}

To solve this labs() would need to either add a new scale other modify the existing scale, which will probably break a lot of existing behaviour.

I would hold off on touching labs() until we get to #4269. Both @yutannihilation and I had reasonable prototypes for addressing this issue. But that's for a major feature release, not a patch release.

Any advice on how to add tests for this feature? Should it be a visual test or there's a better way of checking the text of axis and legends?

Any implementation of this could include a fix for #4631

Sounds like we could ignore labs() for now, and tackle that as a part of a bigger fix.

@eliocamp I think you should be able to find some part of the data structure that exposes the actual name and test that. I don't remember where it is, but if I was writing the test, I would start by doing a View(p) and str(p) and looking for the label.