Fix regression in geom_sf()

[teunbrand]

This PR aims to fix #6189.

Briefly, ggraph::GeomEdgeSf did not have a stroke aesthetic in GeomEdgeSf$default_aes.
That causes some calculations to go awry in GeomSf$draw_panel().
This PR makes a fallback for when stroke is missing.

devtools::load_all("~/packages/ggplot2/")
library(ggraph)

gr <- sfnetworks::as_sfnetwork(sfnetworks::roxel)
ggraph(gr, 'sf') + geom_edge_sf()

^{Created on 2024-11-19 with reprex v2.1.1}

[yutannihilation]

The current CRAN version looks up the default value of GeomLine. Maybe this is better than using the fixed value 0.5? Or, as this is just a safeguard to prevent the error, is it too much?

ggplot2/R/geom-sf.R

Lines 192 to 204 in 8fa3974

	defaults <- list(
	GeomPoint$default_aes,
	GeomLine$default_aes,
	modify_list(GeomPolygon$default_aes, list(fill = "grey90", colour = "grey35", linewidth = 0.2))
	)
	defaults[[4]] <- modify_list(
	defaults[[3]],
	rename(GeomPoint$default_aes, c(size = "point_size", fill = "point_fill"))
	)
	default_names <- unique0(unlist(lapply(defaults, names)))
	defaults <- lapply(setNames(default_names, default_names), function(n) {
	unlist(lapply(defaults, function(def) def[[n]] %||% NA))
	})

ggplot2/R/geom-sf.R

Line 217 in 8fa3974

stroke <- (x$stroke %||% defaults$stroke[1]) * .stroke / 2

[teunbrand]

In my mind, this is pureley to prevent an error. I'm happy to take a default from elsewhere, but the default value in in GeomLine/GeomPath is dynamic:

ggplot2/R/geom-path.R

Line 140 in 3d67907

linewidth = from_theme(linewidth),

At the point the safeguard is introduced, we don't have access to the theme, so we can't compute this default at that point. Ideally, ggraph::GeomEdgeSf$default_aes includes a stroke aesthetic.

[yutannihilation]

we don't have access to the theme, so we can't compute this default at that point

Ah, makes sense.

Ideally, ggraph::GeomEdgeSf$default_aes includes a stroke aesthetic.

Agreed.

[teunbrand]

Thank you for the review!

[agila5]

cc @loreabad6. I don't understand all details, but maybe you are interested in the conversation above and may decide to adjust the ggraph code.