Add lineend and linejoin params to geom_rect() and geom_tile()

[yutannihilation]

Fix #3037

Just as #2132 for geom_segment(), this PR changes

1. geom_rect()'s default of linejoin from "butt" to "square", and
2. makes it configurable via linejoin parameter by users.
3. draw_key_polygon() is changed as well.

In addition, as a workaround for Windows, this PR

4. specifies the lineend parameter.

Note that I expect this breaks many visual tests as geom_rect() and draw_key_polygon() are used in many places.

library(ggplot2)


l <- function(linejoin) {
  geom_tile(aes(x = !!linejoin, y = 0), fill = NA, size = 5, width = 0.7, height = 0.7, linejoin = linejoin, colour = "black")
}

ggplot() +
  l("mitre") +
  l("round") +
  l("bevel") +
  ylim(-1, 1)

^{Created on 2019-05-01 by the reprex package (v0.2.1)}

[ptoche]

I'm new to this, so forgive me if I'm talking nonsense, but do you need to update the legend key as well?

ggplot2/R/legend-draw.r

Line 48 in 1c09bae

draw_key_rect <- function(data, params, size) {

[yutannihilation]

Thanks, you are right!

[thomasp85]

@yutannihilation can you have a look at this again and see if it is ready for review? I think exposing a lineend argument is weird in a graphic element that is always closed. Can we handle it under the hood based on the linejoin argument?

[yutannihilation]

I think exposing a lineend argument is weird in a graphic element that is always closed.

Yeah, I fixed the implementation to use linejoin only (sorry for not updating the description of this PR). But, I'm stuck in confusion about these two points:

1. Why do some visual tests fail with the open form, which polygonGrob() expects? (c.f. geom_tile borders missing notch at top left corners #3037 (comment))
2. Should we specify linend = "square" fixedly as a workaround for Windows' bug (c.f. geom_tile borders missing notch at top left corners #3037 (comment))? As geom_rect() is one of the fundamental geoms, I expect this will break many existing visual tests, but I'm not sure if this is worth.

Let me think a bit if I can finish this...

[thomasp85]

1. is probably as you mention that closed form is needed for correct munching - I didn't anticipate this when making the initial suggestion

2. I think lineend should simply follow the linejoin parameter and not be user-accessible... linejoin = 'round' -> lineend = 'round', everything else lineend = 'square'... Further, this is a graphic device problem, and I think there a limits to the lengths we need to go to to fix it in ggplot2

[yutannihilation]

is probably as you mention that closed form is needed for correct munching - I didn't anticipate this when making the initial suggestion

Thanks, then I'll revert it to closed form.

For the second one, I meant, changing the default lineend (butt) will break visual tests, which means we might have many broken revdeps. As you say, this is a graphic device problem. Do you think it's worth? (I agree it's only linejoin that we should expose to the users)

[thomasp85]

Are the visual tests not only run on linux systems, where the lineend parameter is rightfully ignored?

In any way I think it is worth breaking the tests... they are mainly there to avoid unforeseen regressions

[yutannihilation]

where the lineend parameter is rightfully ignored?

Oh, I didn't know this. Thanks.

Then, I'll try to finish this PR!

[thomasp85]

hmm... can I get you to look into the test failures? Apparently the tests are run on AppVeyor after all. It seems you missed updating the one in test-scales-breaks-labels.R

[yutannihilation]

Yes, I will. Sorry that I can't make much time for this...

[thomasp85]

No worries - didn't know you were hung up... Is there anything I can help with?

[yutannihilation]

It seems you missed updating the one in test-scales-breaks-labels.R

Thanks, this comment really helped me! I didn't merge the current master so the new visual tests added in #2334 were not included here.

Currently I'm trying to figure out what the draw_legend() should be. I didn't notice geom_rect() doesn't use draw_key_rect()...

[yutannihilation]

@thomasp85
I'm really sorry that I couldn't make this in due. Should this be still included in v3.2.0? I think this is ready for review now.

[thomasp85]

This is fine - well add it🙂. Reviewing later today

[yutannihilation]

Sure. I'll do my best to be responsive this week to let this be merged early.

[thomasp85]

LGTM

[yutannihilation]

Thanks!

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/