Allowing text from points in geom_textpath

[AllanCameron]

At the moment, we need to specify a path for each label in geom_textpath, but it would make the geom easier to use if, instead of emitting a warning when each group only consists of a single observation, we interpret this as a request by the user to have a label at a single point. If we construct a little flat path for the label to sit on that is the same size as the text itself, it should behave much the same as geom_text, except curve in polar co-ordinates. This is much the same as what I was proposing in #5 except we don't need a separate geom to do it.

Since the calculation would only happen in the situation where otherwise the geom would fail, it doesn't add any overhead to the existing use cases.

Essentially , the behaviour would be like this:

library(ggplot2)
library(geomtextpath)

df <- data.frame(x = 1:10, y = c(2, 5, 8, 3, 1, 7, 9, 4, 6, 10) + 5, 
                 label = rownames(mtcars)[1:10])

p <- ggplot(df, aes(x, y, colour = label)) + 
       geom_point() +
       lims(y = c(0, 20)) +
       theme_bw() +
       theme(legend.position = "none")

p_text <- p + geom_text(aes(label = label), vjust = 1.5)

p_textpath <- p + geom_textpath(aes(label = label), vjust = 1.5)

p_text

p_textpath

p_text + coord_polar()

p_textpath + coord_polar()

^{Created on 2021-11-30 by the reprex package (v2.0.0)}

Does this fit with our conception of how the geom should behave? It means that under the correct circumstances geom_textpath could act as either exactly like geom_text, or exactly like geom_path, but usually some handy mixture of the two. I like this, but wanted to find out if there were any objections or something I'm missing before committing.

[teunbrand]

Yes I think it very much makes sense to do this. We'd have to make it clear in the docs, readme and possibly introduction vignette that this is the behaviour to be expected. But this raises the question whether point-textpath (sounds like a paradox), should support an angle aesthetic or always be plotted horizontally (in Cartesian coordinates). At the moment I'm in favour of no angle aesthetic, as it's ambiguous what that would imply for path-textpaths, but I can be persuaded otherwise.

I'm also curious how the small path would be calculated, because if you draw it too long, the hjust will be off and if you draw it too short it might not curve properly in polar coordinates. I'm presuming here that the path extension happens at the ggproto level before grob construction, to take advantage of the polar transformation (if applicable).

[AllanCameron]

Yes, this code is inside the draw_panel of geom_textpath

At the moment I am calculating the little path from the string width (taken from shape_string). It needs to be at least as long as the text string, otherwise the curvature doesn't work. I have also changed all linetypes to 0, since we assume that in this case the user doesn't need to know about the path.

You're right that this makes hjust useless. We can address in a few different ways:

1. Use the hjust to recalculate the little path.
2. Make the little path twice the string length. This would allow hjust to work in the [0, 1] range.
3. Simply tell people that hjust won't work in this scenario

Personally, I quite like number 2.

It wouldn't be difficult to implement angle aesthetic for the "point textpaths", and I actually quite like the idea. However, I agree that it doesn't make sense for "path textpaths". Since we are already detecting which of the two is intended by the user, I don't see a problem with implementing it only for point-textpaths. We should be able to warn users that angle will be ignored for "path textpaths", but only where angle is specifically passed as a parameter.

[teunbrand]

I have also changed all linetypes to 0

For the point-texpaths that makes perfect sense.

At the moment I am calculating the little path from the string width (taken from shape_string)

I'm not saying this won't work, but from a theoretical perspective, this returns metrics in pixels, which at draw_panel time still has an unknown relation with the NPC units that the data begets after coord transformation.

Use the hjust to recalculate the little path.

That would probably be the nicest for users, I've sketched a small function for this (y-axis == hjust).

x <- c(1,1,1)
y <- c(0, 0.5, 1)

small_path <- function(x, y, angle, width, hjust) {
  angle <- angle / 180 * pi
  
  xmin <- x + cos(angle + pi) * width * hjust
  xmax <- x + cos(angle) * width * (1 - hjust)
  ymin <- y + sin(angle + pi) * width * hjust
  ymax <- y + sin(angle) * width * (1 - hjust)
  data.frame(xmin = xmin, xmax = xmax, ymin = ymin, ymax = ymax)
}

path <- small_path(x, y, 0, 2, hjust = y)

plot(x, y, xlim = c(-1, 3), ylim = c(-0.5, 1.5))
segments(path$xmin, path$ymin, path$xmax, path$ymax)

# Change angle
path <- small_path(x, y, 45, 2, hjust = c(0, 0.5, 1))

plot(x, y, xlim = c(-1, 3), ylim = c(-0.5, 1.5))
segments(path$xmin, path$ymin, path$xmax, path$ymax)

^{Created on 2021-11-30 by the reprex package (v2.0.1)}

[AllanCameron]

Actually, I don't think it's possible to hack the hjust from within draw_panel because of the whole makeContent mechanism. The length of the text string relative to the path changes according to the device size, so if we grow the device, the little path will get bigger while the text remains the same size. This would have the effect of multiplying the hjust by the same proportion as the device size is multiplied.

[AllanCameron]

I'll push the commit anyway, since it seems to be a desirable feature, and we can work out how (or if) we handle hjust with "point textpaths" once we have had more of a play with it. If we feel the hjust is vital we can consider handling it inside makeContent

[AllanCameron]

I have implemented the angles for point-like textpaths inside draw_panel. Again, these don't behave quite the same as geom_text because the angle is changed as they are scaled. The angle is only accurate with coord_equal (although it should therefore always be accurate in polar coords). I think if we decide this is a bug rather than a feature (I'm not sure about that), it would have to be moved into the makeContent part.

Anyway, the current behaviour is this:

library(geomtextpath)
#> Loading required package: ggplot2

pie <- data.frame(values = c(5, 4, 2.5),
                  mids   = c(2.5, 7, 10.25),
                  fruit  = c("banana", "orange", "apple"),
                  colour = c("yellow", "orange", "green"),
                  x = 1)

p <- ggplot(pie, aes(x = x, y = mids, label = fruit)) + 
      geom_col(aes(x = x, y = values, fill = colour)) +  
      scale_fill_identity() +
      theme_void()

p_00 <- p + geom_textpath(size = 8, flip_inverted = TRUE, angle = 0)
p_90 <- p + geom_textpath(size = 8, flip_inverted = TRUE, angle = 90)
p_45 <- p + geom_textpath(size = 8, flip_inverted = TRUE, angle = 45)

p_00

p_00 + coord_polar(theta = "y")

p_90

p_90 + coord_polar(theta = "y")

p_45

p_45 + coord_polar(theta = "y")

^{Created on 2021-11-30 by the reprex package (v2.0.0)}

[AllanCameron]

Right. I think I've cracked this. I wasn't happy with the little paths and angles being outside of makeContent, so I've moved them inside. Now the angle parameter behaves as expected (in line with geom_text), as does the hjust. It does mean that the textpathGrob now takes 2 new parameters - angles and a switch for telling it to produce little polar paths instead of little straight paths. It also nearly melted my brain to produce little angled paths in polar co-ordinates.

All that aside, I'm very pleased with the results:

library(geomtextpath)
#> Loading required package: ggplot2

pie <- data.frame(values = c(5, 4, 2.5),
                  mids   = c(2.5, 7, 10.25),
                  fruit  = c("banana", "orange", "apple"),
                  colour = c("yellow", "orange", "green"),
                  x = 1)

p <- ggplot(pie, aes(x = x, y = mids, label = fruit)) + 
      geom_col(aes(x = x, y = values, fill = colour)) +
      geom_point() +
      scale_fill_identity() +
      theme_void()

p + geom_textpath(size = 8, flip_inverted = TRUE, angle = 0, hjust = 0)

p + geom_textpath(size = 8, flip_inverted = TRUE, angle = 0, hjust = 1)

p + geom_textpath(size = 8, flip_inverted = TRUE, angle = 0, hjust = 1) + 
    coord_polar(theta = "y")

p + geom_textpath(size = 8, flip_inverted = TRUE, angle = 45, hjust = 0)

p + geom_textpath(size = 8, flip_inverted = TRUE, angle = 45, hjust = 1)

p + geom_textpath(size = 8, flip_inverted = TRUE, angle = 45, hjust = 1) + 
    coord_polar(theta = "y")

p + geom_textpath(size = 8, flip_inverted = TRUE, angle = 90, hjust = 0)

p + geom_textpath(size = 8, flip_inverted = TRUE, angle = 90, hjust = 1)

p + geom_textpath(size = 8, flip_inverted = TRUE, angle = 90, hjust = 1) + 
    coord_polar(theta = "y")

^{Created on 2021-11-30 by the reprex package (v2.0.0)}