`ggsave()`: Optionally switch between `ragg` and `grDevices`

[hsbadr]

Currently, ragg doesn't catch segmentation faults due to invalid res, width, height, units, ... etc. (see r-lib/ragg#82). So, would it make sense to add sanity checks before calling the device or give an option to switch between ragg and grDevices by default (without specifying the device argument), even if ragg is available?

ggplot2/R/save.r

Lines 180 to 188 in 6d94f0d

	if (requireNamespace('ragg', quietly = TRUE)) {
	png_dev <- ragg::agg_png
	jpeg_dev <- ragg::agg_jpeg
	tiff_dev <- ragg::agg_tiff
	} else {
	png_dev <- grDevices::png
	jpeg_dev <- grDevices::jpeg
	tiff_dev <- grDevices::tiff
	}

[yutannihilation]

If you are complaining about the compatibility with ragg, it's already fixed. Did you try the dev version?

#4388

[hsbadr]

Did you try the dev version?

Yes. I'm not complaining; I'm just reporting that ragg causes segmentation fault with some devices (e.g., png) if any of res, width, height, or units is invalid. So, I'm requesting to add sanity checks before calling the device or give an option to disable ragg even if it's installed.

ggplot2/R/save.r

Lines 180 to 188 in 6d94f0d

	if (requireNamespace('ragg', quietly = TRUE)) {
	png_dev <- ragg::agg_png
	jpeg_dev <- ragg::agg_jpeg
	tiff_dev <- ragg::agg_tiff
	} else {
	png_dev <- grDevices::png
	jpeg_dev <- grDevices::jpeg
	tiff_dev <- grDevices::tiff
	}

[clauswilke]

Please provide a reproducible example that triggers a segmentation fault.

[hsbadr]

@clauswilke I'm getting segmentation fault with the development version of ggplot when calling the png device, if ragg is installed. I'm on R-devel.

Here's the simple ggplot example:

> library(ggplot2)
> p <- ggplot(mtcars, aes(mpg, wt)) + geom_point()
> ggsave("mtcars.png", plot = p)
Saving 7 x 7 in image

 *** caught segfault ***
address 0x38, cause 'memory not mapped'

Traceback:

 1: grid.Call(C_stringMetric, as.graphicsAnnot(x$label))
 2: descentDetails.text(list(label = "gjpqyQ", x = 0.5, y = 0.5,     just = "centre", hjust = NULL, vjust = NULL, rot = 0, check.overlap = FALSE,     name = "GRID.text.19", gp = list(fontsize = 8.8, fontfamily = "",         fontface = NULL), vp = NULL))
 3: (function (x) {    UseMethod("descentDetails", x)})(list(label = "gjpqyQ", x = 0.5, y = 0.5, just = "centre",     hjust = NULL, vjust = NULL, rot = 0, check.overlap = FALSE,     name = "GRID.text.19", gp = list(fontsize = 8.8, fontfamily = "",         fontface = NULL), vp = NULL))
 4: grid.Call(C_convert, x, as.integer(whatfrom), as.integer(whatto),     valid.units(unitTo))
 5: convertUnit(x, unitTo, "y", "dimension", "y", "dimension", valueOnly = valueOnly)
 6: convertHeight(grobDescent(textGrob(label = "gjpqyQ", gp = gpar(fontsize = size,     cex = cex, fontfamily = family, fontface = face))), "inches")
 7: font_descent(gp$fontfamily, gp$fontface, gp$fontsize, gp$cex)
 8: title_spec(label, x = x, y = y, hjust = hjust, vjust = vjust,     angle = angle, gp = gp, debug = debug, check.overlap = check.overlap)
 9: titleGrob(label, x, y, hjust = hj, vjust = vj, angle = angle,     gp = modify_list(element_gp, gp), margin = margin, margin_x = margin_x,     margin_y = margin_y, debug = element$debug, ...)
10: element_grob.element_text(list(family = "", face = "plain", colour = "grey30",     size = 8.8, hjust = 0.5, vjust = 1, angle = 0, lineheight = 0.9,     margin = c(2.2, 0, 0, 0), debug = FALSE, inherit.blank = TRUE),     x = c(0.0299806576402321, 0.223404255319149, 0.416827852998066,     0.610251450676983, 0.8036750483559, 0.997098646034816), margin_y = TRUE,     label = c("10", "15", "20", "25", "30", "35"), check.overlap = FALSE)
11: (function (element, ...) {    UseMethod("element_grob")})(list(family = "", face = "plain", colour = "grey30", size = 8.8,     hjust = 0.5, vjust = 1, angle = 0, lineheight = 0.9, margin = c(2.2,     0, 0, 0), debug = FALSE, inherit.blank = TRUE), x = c(0.0299806576402321, 0.223404255319149, 0.416827852998066, 0.610251450676983, 0.8036750483559, 0.997098646034816), margin_y = TRUE, label = c("10", "15", "20", "25", "30", "35"), check.overlap = FALSE)
12: exec(element_grob, label_element, `:=`(!!position_dim, break_positions),     `:=`(!!label_margin_name, TRUE), label = break_labels, check.overlap = check.overlap)
13: draw_axis_labels(break_positions = break_positions[indices],     break_labels = break_labels[indices], label_element = label_element,     is_vertical = is_vertical, check.overlap = check.overlap)
14: FUN(X[[i]], ...)
15: lapply(dodge_indices, function(indices) {    draw_axis_labels(break_positions = break_positions[indices],         break_labels = break_labels[indices], label_element = label_element,         is_vertical = is_vertical, check.overlap = check.overlap)})
16: draw_axis(break_positions = guide$key[[aesthetic]], break_labels = guide$key$.label,     axis_position = guide$position, theme = theme, check.overlap = guide$check.overlap,     angle = guide$angle, n.dodge = guide$n.dodge)
17: guide_gengrob.axis(guide, theme)
18: guide_gengrob(guide, theme)
19: panel_guides_grob(panel_params$guides, position = "bottom", theme = theme)
20: f(...)
21: coord$render_axis_h(range, theme)
22: f(...)
23: self$facet$draw_panels(panels, self$layout, self$panel_scales_x,     self$panel_scales_y, self$panel_params, self$coord, data,     theme, self$facet_params)
24: f(..., self = self)
25: layout$render(geom_grobs, data, theme, plot$labels)
26: ggplot_gtable.ggplot_built(data)
27: ggplot_gtable(data)
28: print.ggplot(x)
29: print(x)
30: grid.draw.ggplot(plot)
31: grid.draw(plot)
32: ggsave("mtcars.png", plot = p)

Possible actions:
1: abort (with core dump, if enabled)
2: normal R exit
3: exit R without saving workspace
4: exit R saving workspace

Uninstalling ragg fixes the problem because it switches to grDevices.

ggplot2/R/save.r

Lines 180 to 188 in 6d94f0d

	if (requireNamespace('ragg', quietly = TRUE)) {
	png_dev <- ragg::agg_png
	jpeg_dev <- ragg::agg_jpeg
	tiff_dev <- ragg::agg_tiff
	} else {
	png_dev <- grDevices::png
	jpeg_dev <- grDevices::jpeg
	tiff_dev <- grDevices::tiff
	}

Can you make the switch between ragg and grDevices optional? i.e., replacing if (requireNamespace('ragg', quietly = TRUE)).

[yutannihilation]

So, I'm requesting to add sanity checks before calling the device or give an option to disable ragg even if it's installed.

Ah, OK. Sorry, I failed to see the context.

But I cannot reproduce your issue on R 4.1.0. Maybe it's somehow related to your installation of ragg package?

library(ggplot2)
p <- ggplot(mtcars, aes(mpg, wt)) + geom_point()
ggsave("mtcars.png", plot = p)
#> Saving 7 x 5 in image

^{Created on 2021-06-05 by the reprex package (v2.0.0)}

[hsbadr]

Maybe it's somehow related to your installation of ragg package?

Yes, I think it's related to the system libraries or the R options. I'm using Cairo with x11 devices on an HPC server. Disabling my R profile fixes the problem too. But, still, it'd be useful to be able to opt in/out the ragg default, regardless if it's installed.

[hsbadr]

@clauswilke @yutannihilation Here's exactly why it fails with my R profile. I spoke too soon; I was testing on my local machine. The problem persists on the server (on R-devel). I can fix it by explicitly setting the device argument though.

[clauswilke]

I think we still need somebody else to reproduce this problem and/or have a concrete explanation of what is going wrong. The bug report started out with talking about invalid parameter inputs for the ragg devices, but I don't see how use of ggsave() with default parameters could lead to invalid parameter inputs.

I see two possibilities here:

1. It's something else that is going wrong
2. There is indeed some obscure situation under which ggsave() produces invalid parameter inputs. But it most likely depends on some other factor that we haven't identified yet. For example, does this only happen when using R devel?

[hsbadr]

@clauswilke Thanks you! With running into the same issue with simple cases, I think it's an ragg issue on some systems or R-devel (possibility # 1). Would it be possible to let the users decide if they want to use ragg by default? or maybe defaulting to grDevices while fully supporting the ragg devices?

That'd mean changing:

ggplot2/R/save.r

Lines 180 to 188 in 6d94f0d

	if (requireNamespace('ragg', quietly = TRUE)) {
	png_dev <- ragg::agg_png
	jpeg_dev <- ragg::agg_jpeg
	tiff_dev <- ragg::agg_tiff
	} else {
	png_dev <- grDevices::png
	jpeg_dev <- grDevices::jpeg
	tiff_dev <- grDevices::tiff
	}

to:

  png_dev <- grDevices::png
  jpeg_dev <- grDevices::jpeg
  tiff_dev <- grDevices::tiff

Then, ragg devices can be used with the device argument. Alternatively, a global option to prefer the default would do it.

[hsbadr]

does this only happen when using R devel?

Yes, but I think it could be a combination of system libs (e.g., libpng) too since ragg passes the R-CMD-check on R-devel.

[hsbadr]

have a concrete explanation of what is going wrong.

I'll let you know when I find the culprit.

[hsbadr]

@clauswilke FYI. The common thing between my environment and #4499 is building R with Intel compilers and MKL.

[yutannihilation]

Just curious. It seems ragg doesn't work at all on your environment. Why do you still want to keep it installed?

[hsbadr]

Just curious. It seems ragg doesn't work at all on your environment. Why do you still want to keep it installed?

Good question. Other packages import/suggest it so it gets installed.

You're right; the problem here is ragg on my system, but this issue requests that at least it isn't used by default with ggplot2::ggsave(). I can specify the device to fix the problem, but that isn't an option when other packages call it.