grid.grab(): strange behavior of "wrap" argument when working with lattice output

grid.grab() captures all of the current viewports and stores them as a grob. In theory, and often in practice, you can use grid.grab() to store an image now for drawing later on. But I have noticed some odd behavior of grid.grab() when working with lattice output. It is related to the wrap argument:

When wrap = TRUE, the saved grob should produce a drawing faithful to the original viewports. But sometimes it doesn't.
Even when using wrap = TRUE, grid.grab() warns me to use wrap = TRUE.

Here is a minimal example to illustrate these behaviors:

library(grid)
library(lattice)
xyplot(1:5 ~ 1:5)
trellis.focus()
panel.abline(h = 3)
panel.abline(h = 5)
myGrob <- grid.grab(wrap = TRUE)

At this point, I receive a warning—

In grabDL(warn, wrap, ...) :
  one of more grobs overwritten (grab WILL not be faithful; try 'wrap = TRUE')

—which is strange, given that I specified wrap = TRUE.

I proceed by running

grid.newpage()
grid.draw(myGrob)

According to Paul Murrell (R Graphics, 2nd ed., page 239), using grid.grab(wrap = TRUE) is "guaranteed to replicate the original output." But it doesn't. Specifically, the top horizontal line (from panel.abline(h = 5)) isn't reproduced.

I've run this example in R 3.3.2 and 3.5.1. It comes out the same in both cases. Are these bugs, or am I overlooking something?

Solution

Paul Murrell answered the question on the R-help mailing list. In short, the problem arises because, in the example above, the code creates two grobs with the same name, which in turn causes problems with grid.grab(). In his answer, Paul suggests that one workaround is to name the second grob: replace

panel.abline(h = 5)

with

panel.abline(h = 5, identifier = "abline2")

and there are no problems. (I've verified that this solution works.) Paul also notes that a coming version of grid.grab() will be better able to handle cases like the one that I described above.