Border for thick lines in R plotting - r

Question

The picture below explains what I'd like to achieve in R. The blue line is added with a simple call to lines. This line shows some values of parameter over time. The line is thick (lwd=3) so it can be seen at a low zoom, and also so that the individual data points fit inside of it. This means that it overlaps with itself when there is chatter over a small x interval. Can the line be made to have a border to better resolve areas with overlap?

Answer 1

Draw a thicker line first and then draw thinner line on top of it
set.seed(42)
x = 1:10
y = sample(1:10)
plot(x, y, type = "l", lwd = 5, col = "red")
lines(x, y, lwd = 3, col = "green")

Answer 2

An alternative.
Under the assumption that seeing borders in order and on top of the "fill" color is important, then borrowing from d.b's excellent, fast, and very-simple answer that intersections give no indication of which direction came first (left of the 3-pack, below).
Altered data, to show overlaps:
set.seed(42)
x <- sample(10)
y <- sample(10)
One solution is to draw each segment individually, controlling colors each time.
segments2 <- function(x, y, lwd = c(2, 1), ..., border = NA, fill = NA) {
stopifnot(length(x) == length(y))
len <- length(x)
ign <- Map(function(x0, y0, x1 = x0, y1 = y0) {
if (!is.na(border)) lines(c(x0, x1), c(y0, y1), lwd = lwd[1], col = border, ...)
if (!is.na(fill)) lines(c(x0, x1), c(y0, y1), lwd = lwd[2], col = fill, ...)
}, x[-len], y[-len], x[-1], y[-1])
invisible()
}
This produces the middle plot below, but notice that each vertex has border-intrusion. A third option, much more complex, can mitigate that with a lot of trickery.
segments3 <- function(x, y, lwd = c(2, 1), ..., border = NA, fill = NA, lend = 0) {
stopifnot(length(x) == length(y))
len <- length(x)
dx <- x[-len] + diff(x) / 2
mx <- rbind(
c(NA, x[1], dx[1]),
cbind(dx[-(len-1)], x[-c(1,len)], dx[-c(1)]))
mx <- rbind(
mx[-(len-1),],
c(x[len], dx[len-1], NA),
mx[len-1,])
dy <- y[-len] + diff(y) / 2
my <- rbind(
c(NA, y[1], dy[1]),
cbind(dy[-(len-1)], y[-c(1,len)], dy[-c(1)]))
my <- rbind(
my[-(len-1),],
c(y[len], dy[len-1], NA),
my[len-1,])
for (rn in seq_len(nrow(mx))) {
lend0 <- if (rn %in% c(1L, len-1)) lend else 1
lines(mx[rn,], my[rn,], lwd = lwd[1], col = border, ..., lend = lend0)
lines(mx[rn,], my[rn,], lwd = lwd[2], col = fill, ..., lend = lend0)
}
}
It is likely possible to simplify this, but its performance is not heinous, and it does produce a slightly clearer plot.
par(mfrow=c(1, 3))
# simpler method
plot(x, y, type = "l", lwd = 5, col = "red", main = "Simpler")
lines(x, y, lwd = 3, col = "green")
# slower method, full-size
plot(x, y, type = "n", main = "Slower")
segments2(x, y, lwd = c(5, 3), border="red", fill="green")
# slowest method, fairly complex
plot(x, y, type = "n", main = "Slowest")
segments3(x, y, lwd = c(5, 3), border="red", fill="green")
Some notes:
lwd is a length-2 vector used to control the widths of the two lines; perhaps it would be more intuitive to have lwd be the main line and border.lwd be the border's thickness?
... arguments are passed through to lines, except
lend, which is tightly-controlled in order to keep mid-segment line endings controller but allow the user to override the two ends
if either border or fill are NA, then they just won't be drawn, which would then make one question why using this function
depending on your perspective, dimensions, etc, it is possible that a segment mid-point hints at a little junction, which can be seen as a hint in the left-most segment around (1.8, 3.7) ... this can go away just be resizing, and is I believe just an artifact of raster graphics in general