Overhead in vrt()

[kadyb]

Maybe there is room for optimizing the vrt() function?

library("sf")
library("terra") # 1.7.39
gdal() # 3.6.2

n = 4000
r = rast(ncols = n, nrows = n, vals = rnorm(n ^ 2))
filename = tempfile(fileext = "_.tif")
ff = makeTiles(r, 100, filename) # 1600 tiles
vrt = tempfile(fileext = ".vrt")

system.time({ terra::vrt(ff, filename = vrt) })
#> user  system elapsed 
#> 8.66    7.55   16.21

system.time({ sf::gdal_utils(util = "buildvrt", source = ff, destination = vrt) })
#> user  system elapsed
#> 2.54    2.84    5.39

[dfriend21]

I've also noticed that reading in a ".vrt" that's already been created and that references many rasters takes a while - at least it takes longer than I expect, although I'm not familiar with what's going on under the hood, so I'm not sure what a reasonable expectation is.

library(terra)
#> terra 1.7.39

r <- rast(system.file("ex/elev.tif", package = "terra"))
paths <- makeTiles(r, 3, tempfile(fileext = ".tif"))
length(paths)
#> [1] 960
system.time(r_vrt <- vrt(paths))
#>    user  system elapsed 
#>    3.44    2.66    6.36
system.time(rast(sources(r_vrt)))
#>    user  system elapsed 
#>    2.11    1.65    4.06

^{Created on 2023-08-24 with reprex v2.0.2}

[rhijmans]

I think the difference is because terra::vrt returns a SpatRaster:

system.time(x <- terra::vrt(ff, filename = vrt) )
#   user  system elapsed 
#  14.94   21.25   36.25 

Which is about the same duration as

system.time({ sf::gdal_utils(util = "buildvrt", source = ff, destination = vrt) })
#   user  system elapsed 
#   4.94    7.47   12.44 
system.time(y <- terra::rast(vrt) )
#   user  system elapsed 
#   9.47   15.17   24.64 

There could be an option to only create the .vrt file and not return a SpatRaster.

[rhijmans]

You can now request that the filename is returned instead of a SpatRaster:

f <- terra::vrt(ff, filename = vrt, return_filename=T)

[kadyb]

Thanks! Now I get:

system.time({ terra::vrt(ff, filename = vrt) })
#> user  system elapsed 
#> 8.17    7.70   15.87

system.time({ terra::vrt(ff, filename = vrt, return_filename = TRUE) })
#> user  system elapsed 
#> 3.30    2.45    5.75