Edges disappear after second identical plot() statement; when using transparent edge colors (alpha < 1).

[clpippel]

Describe the bug
Edges disappear after second identical plot() statement when using transparent edge colors (alpha < 1).

Note par()$xpd is flipped from FALSE to TRUE.

To reproduce

library(igraph)
g <- graph_from_literal(A +-+ B)
dev.new(); par()$xpd                # xpd = FALSE

# First  try after dev.new(), edge is  visible.
plot(g,  edge.color=rgb(0,0,0,0.3));
par()$xpd                           # xpd = TRUE

# Second try, edge is not visible.
# Edge becomes visible after save as .pdf").
plot(g,  edge.color=rgb(0,0,0,0.3)) 

# Third try. 
par(xpd=FALSE)                      # reset xpd 
plot(g,  edge.color=rgb(0,0,0,0.3)) # Edge is visible again.

Version information

• R/igraph version: [1] compiler_4.1.3 magrittr_2.0.1 rlang_1.0.5 pkgconfig_2.0.3

• R version: R version 4.1.3 (2022-03-10)

• Operating system: Windows 10 x64 (build 22621)
  

[clpippel]

Related problems with alpha channel are observed in Stack Overflow:

• igraph in R: Directed graphs (with arrows) causes vertex color transparency
• Unexpected arrow color behavior with semi-transparent edges (igraph, R)
  For example:

   plot( make_tree(5)
       , edge.arrow.mode=3
       , edge.color=rgb(0,0,0,0.2)
   )   # some arrow colors are rendered transparent.

[ntamas]

Is this happening with the latest development version from the main branch? I have recently fixed an issue that could be related to this in 133f74d

[ntamas]

Re the issue you mentioned in your second comment, there seems to be some issue there, but the edges are okay - the transparencies of the edge arrowheads seem to be incorrect to me with the latest version from main, at least on macOS. See the attached image.

[clpippel]

The transparencies of the arrow heads are also wrong on Windows, igraph version ‘1.3.4’. This is a separate issue.

Regarding the par()$xpd issue: how to download and try the latest development version for Windows?

[ntamas]

Ah sorry, I forgot that you are on Windows. I'll submit the source to win-builder and post a link to the results here.

[ntamas]

Here's the latest dev version for R 4.2:

https://win-builder.r-project.org/DFE1mD2399tN/

It will be online for the next 72 hours. You can install it by downloading it and then typing install.packages(file.choose(), repos=NULL) in the R command line.

[clpippel]

I have downloaded and installed the igraph version as requested.
[1] ‘1.3.5.9098’

To summarize:
(1) As expected the edges do not disappear. And par$xpd stays FALSE.
(2) The Directed graphs (with arrows) causes vertex color transparency problem is solved.
The following program shows no vertices with transparent colors:

library(igraph)
stack_graph <- make_ring(2, directed=TRUE)

plot(stack_graph, 
     layout = layout_in_circle(stack_graph), 
     vertex.shape = "sphere",
     edge.color = c("#66666660"),   # Transparency issues are caused by non valid RBG color code.
     edge.arrow.size = 2            # Changing this flag with anything > 0 will show transparency issues of the graph vertices.
     ) 

(3) The transparencies of the edge arrowheads are still incorrect.

     plot( make_tree(4)
       , edge.arrow.mode=3
       , edge.color=rgb(0,0,0,0.2)
   )   # some arrow colors are rendered transparent.

The problem does not occur with make_tree(n) with n < 4.

[ntamas]

@krlmlr I'm a bit stumped with this issue; can you take a look at it whenever it's convenient for you.

To reproduce:

library(igraph)
plot(make_tree(4), edge.arrow.mode=3, edge.color=rgb(0,0,0,0.2))

edge.arrow.mode essentially instructs igraph to draw arrows on both endpoints of an edge. The problem is that when the edge color uses alpha transparency, then the "real" arrowhead (i.e. the one at the head of the edge) is drawn correctly, but the "other" arrowhead (at the tail of the edge) is darker (i.e. uses a higher alpha than what it should be). If I change edge.arrow.mode to 2 (which draws an arrowhead only at the tail but not at the head) or 1 (which draws an arrowhead at the head but not at the tail), then everything is okay. There must be a logic error in the internal igraph.Arrows function, but I can't figure out what the problem is.

[krlmlr]

We would need more visual tests to have confidence in fixing or refactoring the plotting code.

[schochastics]

The difference in colors comes from the fact that the arrow heads are drawn multiple time when edge.arrow.mode=3 (in the code below it is called code). The problem is the variable r.arr.
When the forward edges are drawn, it is duplicated lx times (see 973)

rigraph/R/plot.R

Lines 966 to 973 in fc769f7

	## forward arrowhead
	if (code %in% c(2, 3)) {
	theta <- atan2((by2 - y1) * uin[2], (bx2 - x1) * uin[1])
	Rep <- rep(length(deg.arr), lx)
	p.x2 <- rep(bx2, Rep)
	p.y2 <- rep(by2, Rep)
	ttheta <- rep(theta, Rep) + rep(deg.arr, lx)
	r.arr <- rep(r.arr, lx)

The variable is not reset to its origin value and then again duplicated lx times for the backward edges (see L1003)

rigraph/R/plot.R

Lines 987 to 1003 in fc769f7

	## backward arrow head
	if (code %in% c(1, 3)) {
	x1 <- bx1
	y1 <- by1
	tmp <- x1
	x1 <- x2
	x2 <- tmp
	tmp <- y1
	y1 <- y2
	y2 <- tmp
	theta <- atan2((y2 - y1) * uin[2], (x2 - x1) * uin[1])
	lx <- length(x1)
	Rep <- rep(length(deg.arr), lx)
	p.x2 <- rep(x2, Rep)
	p.y2 <- rep(y2, Rep)
	ttheta <- rep(theta, Rep) + rep(deg.arr, lx)
	r.arr <- rep(r.arr, lx)

Hence, the arrow heads get drawn on top of each other lx times and that becomes evident when the transparency is not 1.

This is fixed in #1709