geom_line() render thread runs at 100%, apparently indefinitely, on wide-ranged input

[twest820]

Stumbled on this particular breakdown while looking at some fast(er) math things. ggplot 3.5.1 on R 4.4.0 with ragg 1.3.0 and RStudio 2024.04.2+764.

library(dplyr)
library(ggplot2)

theme_set(theme_bw() + theme(axis.line = element_line(linewidth = 0.3), panel.border = element_blank()))

arctan = tibble(theta = seq(0, 360), x = cos(pi/180 * theta), y = sin(pi / 180 * theta), quadrant = factor(if_else(x >= 0, if_else(y >= 0, 1, 4), if_else(y >= 0, 2, 3))), 
                arctan = atan2(y, x),
                yx = y/x, 
                horner = yx * (0.99997726 + yx^2 * (-0.33262347 + yx^2 * (0.19354346 + yx^2 * (-0.11643287 + yx^2 * (0.05265332 - yx^2 * 0.01172120))))))

ggplot() +
  geom_line(aes(x = y/x, y = arctan, group = quadrant, color = quadrant, linetype = "atan2"), arctan) +
  geom_line(aes(x = y/x, y = horner, group = quadrant, color = quadrant, linetype = "Horner"), arctan) +
  coord_cartesian(xlim = c(-1, 1) * 10, ylim = c(-1, 1) * pi) +
  labs(x = "y/x", y = "atan2(y, x)", color = "quadrant", linetype = "implementation")
# quickly shows console prompt but never renders and prompt doesn't take new input

Interrupt has no effect so recovery's via killing the R session and restarting.

Workaround's to only plot the 90° wedges of the circle where the Horner polynomial's well posed. This restricts the plotted y values to less than ±1 rather than going up to 10¹⁷⁶ or so.

arctan %>% filter(yx >= -1, yx <= 1)

[teunbrand]

Thanks for the report, I can reproduce the issue on windows.
It seems to me like this is a {ragg} issue because the plot seems fine with e.g. the png(type = "cairo") as device.

library(dplyr)
library(ggplot2)

theme_set(theme_bw() + theme(axis.line = element_line(linewidth = 0.3), panel.border = element_blank()))

arctan = tibble(
  theta = seq(0, 360), x = cos(pi/180 * theta), y = sin(pi / 180 * theta), 
  quadrant = factor(if_else(x >= 0, if_else(y >= 0, 1, 4), if_else(y >= 0, 2, 3))), 
  arctan = atan2(y, x),
  yx = y/x, 
  horner = yx * (0.99997726 + yx^2 * (-0.33262347 + yx^2 * (0.19354346 + yx^2 * (-0.11643287 + yx^2 * (0.05265332 - yx^2 * 0.01172120)))))
)

ggplot() +
  geom_line(aes(x = y/x, y = arctan, group = quadrant, color = quadrant, linetype = "atan2"), arctan) +
  geom_line(aes(x = y/x, y = horner, group = quadrant, color = quadrant, linetype = "Horner"), arctan) +
  coord_cartesian(xlim = c(-1, 1) * 10, ylim = c(-1, 1) * pi) +
  labs(x = "y/x", y = "atan2(y, x)", color = "quadrant", linetype = "implementation")

^{Created on 2024-07-05 with reprex v2.1.0}

In principle ggplot2 doesn't do anything wrong here, as the graphical objects that ggplot2 code yields are not invalid. For this reason, I'll close this issue here, but I'll signal @thomasp85 that there might be an issue with {ragg}.