S3 proj_trans() Generic

[anthonynorth]

Re-implements proj_trans(), leveraging wk::wk_handle() and proj_trans_create(). This allows for inputs to take any form that wk::wk_handle() supports, including:

• Coordinate matrices
• Data frames with coordinates
• {wk} geometry vectors: xy, wkt, wkb, rct,
• {sf} geometry vectors
• {geos} geometry vectors
• {geoarrow} geometry vectors
• {s2} geography vectors (although this makes little sense)
• A data frame containing a geometry/geography vector from wk, sf, geos, geoarrow, s2

For coordinate matrices / data frames, there's an approximate 15-20% compute overhead compared with previous implementation. Benchmark:

library(PROJ)

xy <- wk::xy(
  runif(1e6, -180, 180),
  runif(1e6, -90, 90),
  "OGC:CRS84"
)

coords <- as.matrix(xy)
sfc <- sf::st_as_sfc(xy)

bench::mark(
  .Call(PROJ:::C_proj_trans_xy, wk::xy_x(xy), wk::xy_y(xy), "OGC:CRS84", "EPSG:3857"),
  .Call(PROJ:::C_proj_trans_list, unclass(xy), "OGC:CRS84", "EPSG:3857"),
  proj_trans(xy, "EPSG:3857"),
  sf::sf_project("OGC:CRS84", "EPSG:3857", coords),
  proj_trans(sfc, "EPSG:3857"),
  sf::st_transform(sfc, "EPSG:3857"),
  iterations = 5,
  check = FALSE
)
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
#> # A tibble: 6 × 6
#>   expression                           min   median `itr/sec` mem_alloc `gc/sec`
#>   <bch:expr>                      <bch:tm> <bch:tm>     <dbl> <bch:byt>    <dbl>
#> 1 ".Call(PROJ:::C_proj_trans_xy,… 167.98ms 168.92ms     5.92     15.3MB    0
#> 2 ".Call(PROJ:::C_proj_trans_lis…  163.6ms 164.08ms     6.04     15.3MB    1.21
#> 3 "proj_trans(xy, \"EPSG:3857\")" 213.58ms 213.74ms     4.64     15.4MB    0.929
#> 4 "sf::sf_project(\"OGC:CRS84\",… 164.49ms 168.31ms     5.95     15.3MB    1.19
#> 5 "proj_trans(sfc, \"EPSG:3857\"… 598.56ms 652.47ms     1.31     30.8MB    1.31
#> 6 "sf::st_transform(sfc, \"EPSG:…    3.03s    3.38s     0.293   114.8MB    0.820

^{Created on 2024-03-22 with reprex v2.1.0}

Closes #49

Breaking changes

• Z and T coordinate vectors are always supplied in the x parameter
• Renamed parameters: target -> target_crs, source -> source_crs
• source_crs defaults to the input crs if available, falling back to OGC:CRS84
• If coordinate matrices are named, expects names "x", "y" and optionally "z" and/or "m" columns. Previously names were ignored
• Coordinate data frames expect names "x", "y" and optionally "z" and/or "m" columns
• Output coordinate matrix or data frame will always contain names "x", "y" and optionally "z" and/or "m" depending on use_z and use_m parameters

[anthonynorth]

Failing tests are related to paleolimbot/wk#217 which hasn't yet landed on CRAN.

[mdsumner]

I'll check some more tomorrow, I think this is fine to merge and I will probably just do that after I do my "usage" checks for usual tricks.

[anthonynorth]

I haven't yet removed the existing C_proj_trans_xy() and C_proj_trans_list() functions. Useful for a benchmark, but wasn't sure what you want to do with them. We can refactor these to avoid taking any perf hit for the matrix/dataframe usage if it's important.

I haven't attempted to replicate this behaviour in the {wk} transformer. I think it's doable though

PROJ/src/C_proj_trans.c

Lines 78 to 81 in 7f4c4e7

	if (r) {
	Rprintf("Error detected, some values Inf (error code: %i)\n\n", r);
	Rprintf("' %s\n\n '", proj_errno_string(r));
	}

[mdsumner]

lgtm!

[mdsumner]

any reason not to merge this in? I'm keen to do that and move on, I can spend some time here now

[anthonynorth]

Sorry for the delay, I've been unable to get back to this. I'll drop the superseded c functions and then we can merge.

We could put in a guard for the failing tests also since they're dependent on wk > 0.9.1 which isn't on cran.