"image" pattern bug with type = "tile"

[trevorld]

• Sometimes the "image" pattern doesn't work on certain platforms and R provides a "magick_image_readpath()" error.
• I've noticed before that the "tile" pattern_type seems to reliably choke on CRAN/Rhub Windows and OSX machines but seems to work fine on Linux and Solaris machines. I suspect this is some subtle upstream magick bug that only affects certain platforms (and/or only certain magick versions?). It isn't something I've been able to reproduce locally on my Linux laptop (where it works fine) but I have been able to reproduce indirectly via R-builder/Rhub checks on relevant platforms.
• This bug has also been noticed downstream: "image" pattern bug with 'type = "tile"' ggpattern#37

[trevorld]

Reproduced in RStudio on a Windows machine:

logo_filename   <- system.file("img", "Rlogo.png" , package="png")
gridpattern::grid.pattern_image(filename = logo_filename, type = "tile")

Invalid Parameter - 247x247
Error in magick_image_readpath(path, density, depth, strip, defines) : 
  rsession.exe: UnableToOpenBlob `C:\Users\Frank\AppData\Local\Temp\RtmpMj8nXg\file2bac745d50c2.png': No such file or directory @ error/blob.c/OpenBlob/2924

[trevorld]

Issue is we call convert via system() but on Windows convert is usually something other than imagemagick so it is choking.

It should theoretically be possible to tile images using magick::image_blank() but for some reason it is inconsistently buggy:

library("grid")
library("magick")
library("png")
logo_filename   <- system.file("img", "Rlogo.png" , package="png")
pseudo = paste0("tile:", logo_filename)
grid.raster(magick::image_blank(600, 600, pseudo_image = pseudo))

[trevorld]

image_blank() inconsistency reported upstream: ropensci/magick#344

[trevorld]

{magick} maintainer suggests manually tiling image using image_append() or image_montage() (would need to do some manual calculations and then crop/chop at the end).