Speed up st_crs for SpatVector objects

[atsyplenkov]

Hey, guys.

Consider updating the st_crs() function for the SpatVector objects. The trick is terra::crs() always returns a character vector, therefore we can skip some checks implemented in st_crs() and we can gain a 2x speed increase in CRS extraction. When there is no CRS in SpatVector object, then the transition happens 1.2 times faster. When it comes to st_as_sf(), then it works 10% faster.

See examples below:

library(sf)
#> Linking to GEOS 3.12.1, GDAL 3.8.4, PROJ 9.3.1; sf_use_s2() is TRUE
library(terra)
#> terra 1.7.71

f <- system.file("ex/lux.shp", package="terra")
v <- vect(f)


st_crs2 = function(x) {
  if (!requireNamespace("terra", quietly = TRUE))
    stop("package terra required, please install it first") # nocov
  ctr <- terra::crs(x)
  if (ctr == "") {
    NA_crs_
  } else {
    crs <- sf:::CPL_crs_from_input(ctr)
    crs$input <- crs$Name
    crs
  }
}

bench::mark(
  original = st_crs(v),
  PR = st_crs2(v),
  check = T,
  relative = T,
  max_iterations = 1000L
)
#> # A tibble: 2 × 6
#>   expression   min median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <dbl>  <dbl>     <dbl>     <dbl>    <dbl>
#> 1 original    1.93   1.83      1         296.     1.00
#> 2 PR          1      1         1.86        1      1

crs(v) <- NA_character_

bench::mark(
  original = st_crs(v),
  PR = st_crs2(v),
  check = T,
  relative = T,
  max_iterations = 1000L
)
#> # A tibble: 2 × 6
#>   expression   min median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr> <dbl>  <dbl>     <dbl>     <dbl>    <dbl>
#> 1 original    1.24   1.20      1          Inf      NaN
#> 2 PR          1      1         1.11       NaN      NaN

^{Created on 2024-05-09 with reprex v2.1.0}

[rsbivand]

All geographical objects should always declare their CRS. Missing CRS are objects with missing essential data, with no information on their coordinate measurement units or relative shape in relation to other data objects. So helping users abuse their data is not a good idea at all. See

sf/R/read.R

Lines 546 to 549 in 97a179a

	if (is.na(st_crs(geom))) {
	message('writing: substituting ENGCRS["Undefined Cartesian SRS with unknown unit"] for missing CRS')
	st_crs(geom) = st_crs("ENGCRS[\"Undefined Cartesian SRS with unknown unit\",EDATUM[\"Unknown engineering datum\"],CS[Cartesian,2],AXIS[\"X\",unspecified,ORDER[1],LENGTHUNIT[\"unknown\",0]],AXIS[\"Y\",unspecified,ORDER[2],LENGTHUNIT[\"unknown\",0]]]")
	}

for a specific case.

[atsyplenkov]

Sure thing! However, I don't understand how my PR differs from the current behavior and philosophy of st_crs.

sf/R/crs.R

Lines 73 to 76 in 97a179a

	st_crs.character = function(x, ...) {
	if (is.na(x))
	NA_crs_
	else {

I am suggesting improving st_crs only for SpatVector objects, which rely on the terra::crs() function, which always returns a character vector.