Support guide_axis() on CoordTrans

[yutannihilation]

Address #3959 partly, fix #4551.

The plan was to support guide_axis() on all coords, but, as it might require some more consideration, I narrowed the scoped of this PR only to CoordCartesian and CoordTrans.

• Move labels(), setup_panel_guides() and train_panel_guides() to Coord
• Tweak these methods to support panel_params without ViewScales for backward compatibility
• Modify each setup_panel_params() to include ViewScale
• Add tests

---

library(ggplot2)

do_plot <- function() {
  df <- data.frame(x = 1, y = 1)
  p <- ggplot(df, aes(x, y)) +
    geom_point() +
    guides(
      x = guide_axis(title = "x (primary)"),
      y = guide_axis(title = "y (primary)"),
      x.sec = guide_axis(title = "x (secondary)"),
      y.sec = guide_axis(title = "y (secondary)")
    )
  
  p1 <- p +
    labs(tag = "p1")
  
  p2 <- p +
    scale_x_continuous(position = "top") +
    scale_y_continuous(position = "right") +
    labs(tag = "p2")
  
  p3 <- p1 +
    coord_trans() +
    labs(tag = "p1'")
  
  p4 <- p2 +
    coord_trans() +
    labs(tag = "p2'")
  
  patchwork::wrap_plots(p1, p2, p3, p4, nrow = 2)
}

do_plot()

devtools::load_all("~/GitHub/ggplot2/")
#> ℹ Loading ggplot2
do_plot()

^{Created on 2022-07-08 by the reprex package (v2.0.1)}

[yutannihilation]

Hmm, It turned out I failed to fix how CoordTrans treats Inf on #2972.. Will fix in this PR.

[thomasp85]

is this ready for review?

[yutannihilation]

No, not yet. As this will tweak a lot of internals, I think we should not include this in v3.3.1 to avoid possible breakages on extension packages.

[yutannihilation]

(Currently I'm stuck here...)

[thomasp85]

ok, no problem

[thomasp85]

@yutannihilation do you want to take this up again for next release?

[yutannihilation]

Sure. I forgot everything, but I'll try again.

[yutannihilation]

Current progress:

• CoordCartesian works fine (no change since 2 years ago).
• CoordTrans works now. There might be still room for improvement.
• CoordPolar doesn't work at the moment. I'm struggling with how to convert r and theta to x and y, and vice versa.
• I don't look into CoordSf yet, but since CoordSf is so special, I think I won't be able to include it in this pull request.

[thomasp85]

Is it possible to get this done for the next release (soon) or should we punt it?

[yutannihilation]

I don't think I can finish CoordPolar and CoordSf, sorry. Is it an option to narrow the scope of this pull request only to CoordCartesian and CoordTrans?

[thomasp85]

Yes, we can do that

[yutannihilation]

Thanks, I reverted the change on CoordPolar. Should be ready for review soon...

[yutannihilation]

Okay, I need to add some tests, but this should be basically ready for review now. I think I figured out how things should work.

[yutannihilation]

I was hoping to test the label without adding a visual case because then we'll need many for testing similar cases for issues like #4650, but I couldn't find any nice way...

[thomasp85]

LGTM, but @paleolimbot may be good to loop in if he has time

[paleolimbot]

My memory of these internals is poor but I like the approach (move stuff from coord-cartesian to the base coord class where maybe it can be helpful in more future or current coord subclasses), and I appreciate the comments documenting the idiosyncrasies. Thank you!

[yutannihilation]

Thank you for reviewing! I'm happy that I completed this at last.