feat(r): Optimize conversion from sfc to ArrowArray

[paleolimbot]

This PR implements an optimized path for sf objects to ArrowArray when we know in advance all elements of the sfc are the same type and dimension (very common). This is ~2.3-3x faster than the visitor-based approach and results in very few datasets that are large enough for anybody to notice (~100ms). This version can also drop dimensions (e.g., for when you send a 3D geometry to S2 where the Z values will be ignored anyway).

I spent a reasonable amount of time attempting a two-pass conversion (one pass to count elements to perfectly preallocate, one pass to fill the buffers). This is slower than overallocating because (1) we have a pretty good idea of how many elements we have in advance (most multilinestrings have at least one element and at least one coordinate) and (2) the cost of reallocating is on par with the cost of calling the R API for every element in the sfc twice. The version here is more readable, too (e.g., can use GeoArrowBuilderOffsetAppend() instead of builder->view.buffers[i + 1].data.as_int32...).

library(sf)
#> Linking to GEOS 3.12.0, GDAL 3.7.2, PROJ 9.3.0; sf_use_s2() is TRUE

# https://github.com/geoarrow/geoarrow-data/releases/download/v0.1.0/ns-water-water_line.fgb.zip
multilinestrings <- read_sf("~/Downloads/ns-water-water_line.fgb")$geometry

bench::mark(
  geoarrow:::as_geoarrow_array.default(multilinestrings),
  geoarrow:::as_geoarrow_array.sfc(multilinestrings),
  iterations = 10,
  check = FALSE
)
#> # A tibble: 2 × 6
#>   expression                            min  median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>                        <bch:t> <bch:t>     <dbl> <bch:byt>    <dbl>
#> 1 geoarrow:::as_geoarrow_array.def… 264.5ms 283.3ms      3.56    2.87MB        0
#> 2 geoarrow:::as_geoarrow_array.sfc…  92.5ms  99.4ms      6.47   74.44KB        0

linestrings <- st_cast(multilinestrings, "LINESTRING")
bench::mark(
  geoarrow:::as_geoarrow_array.default(linestrings),
  geoarrow:::as_geoarrow_array.sfc(linestrings),
  iterations = 10,
  check = FALSE
)
#> # A tibble: 2 × 6
#>   expression                             min median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>                           <bch> <bch:>     <dbl> <bch:byt>    <dbl>
#> 1 geoarrow:::as_geoarrow_array.defaul… 294ms  308ms      3.28   302.1MB        0
#> 2 geoarrow:::as_geoarrow_array.sfc(li… 134ms  137ms      5.91    59.3KB        0


points <- wk::wk_vertices(linestrings)
bench::mark(
  geoarrow:::as_geoarrow_array.default(points),
  geoarrow:::as_geoarrow_array.sfc(points),
  iterations = 10,
  check = FALSE
)
#> # A tibble: 2 × 6
#>   expression                           min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>                      <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 geoarrow:::as_geoarrow_array.d…    1.42s    1.45s     0.680    59.1KB        0
#> 2 geoarrow:::as_geoarrow_array.s… 236.42ms 965.97ms     1.18     46.5KB        0

In all cases this is faster than generating WKB, although that might just be limitations of R or Rcpp that could be optimized in sf:

library(sf)
#> Linking to GEOS 3.12.0, GDAL 3.7.2, PROJ 9.3.0; sf_use_s2() is TRUE

# https://github.com/geoarrow/geoarrow-data/releases/download/v0.1.0/ns-water-water_line.fgb.zip
multilinestrings <- read_sf("~/Downloads/ns-water-water_line.fgb")$geometry

bench::mark(
  geoarrow:::as_geoarrow_array.default(multilinestrings),
  geoarrow:::as_geoarrow_array.sfc(multilinestrings),
  geoarrow:::as_geoarrow_array.default(multilinestrings, schema = geoarrow::na_extension_wkb()),
  sf:::CPL_write_wkb(multilinestrings),
  iterations = 10,
  check = FALSE
)
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
#> # A tibble: 4 × 6
#>   expression                           min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>                      <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 geoarrow:::as_geoarrow_array.d… 262.63ms 283.18ms     3.57     2.87MB    0    
#> 2 geoarrow:::as_geoarrow_array.s…  97.56ms 167.43ms     5.26    74.44KB    0    
#> 3 geoarrow:::as_geoarrow_array.d… 293.47ms  302.4ms     3.30     6.56KB    0    
#> 4 sf:::CPL_write_wkb(multilinest…    1.45s    1.64s     0.577  378.59MB    0.924

^{Created on 2023-11-20 with reprex v2.0.2}

[anthonynorth]

Maybe use wk::wk_vector_meta() here?

Suggested change

	if (class(x)[1] %in% c("sfc_POINT",
	# or use the labels instead
	if (wk::wk_vector_meta(x)$geometry_type %in% 1:6) {

[paleolimbot]

Good call!

[anthonynorth]

Maybe use wk::wk_vector_meta() here?

Suggested change

	if (!is.null(attr(x, "m_range"))) {
	if (wk::wk_vector_meta(x)$has_m) {

[paleolimbot]

Good call!