Add s2_convex_hull(), add na.rm argument to s2_convex_hull_agg()

[paleolimbot]

This PR adds a feature-wise convex-hull generator based on the implementation by @spiry34 in #151. It also adds na.rm, since this is how the other _agg() functions are implemented. This PR completes #150.

@spiry34 you're the most familiar with this code...is there anything I've missed or should have considered? You mentioned that the convex hull for lines is different and I haven't checked this in GEOS. I should probably check this in BigQuery to make sure that this implementation is consistent with that (since that's how I've tested the other functions).

library(s2)

s2_convex_hull(
  c(
    "GEOMETRYCOLLECTION(POINT(3.6 43.2), POINT (0 0), POINT(3.61 43.21))", 
    NA
  )
)
#> <s2_geography[2]>
#> [1] <POLYGON ((0 0, 3.61 43.21, 3.6 43.2, 0 0))>
#> [2] <NA>

s2_convex_hull_agg(c("POINT (0 0)", NA), na.rm = FALSE)
#> <s2_geography[1]>
#> [1] <NA>
s2_convex_hull_agg(
  c("GEOMETRYCOLLECTION(POINT(3.6 43.2), POINT (0 0), POINT(3.61 43.21))", NA),
  na.rm = TRUE
)
#> <s2_geography[1]>
#> [1] <POLYGON ((0 0, 3.61 43.21, 3.6 43.2, 0 0))>

^{Created on 2022-01-16 by the reprex package (v2.0.1)}

[codecov-commenter]

Codecov Report

Merging #163 (7569004) into main (90fa128) will decrease coverage by 0.02%.
The diff coverage is 94.87%.

@@            Coverage Diff             @@
##             main     #163      +/-   ##
==========================================
- Coverage   94.78%   94.75%   -0.03%     
==========================================
  Files          42       42              
  Lines        3200     3220      +20     
==========================================
+ Hits         3033     3051      +18     
- Misses        167      169       +2     

Impacted Files	Coverage Δ
src/geography.h	32.00% <0.00%> (-2.79%)	⬇️
R/s2-transformers.R	100.00% <100.00%> (ø)
src/geography-collection.h	73.60% <100.00%> (+0.42%)	⬆️
src/s2-transformers.cpp	96.70% <100.00%> (+0.14%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 90fa128...7569004. Read the comment docs.

[paleolimbot]

It's a good point about the convex hull of a segment/point. I don't know enough about the application here to know how important this is...it certainly could be included. I can see how it's difficult/verbose to replicate (my attempt is below):

library(s2)

fix <- function(result, original, tolerance = 1e-14) {
  simple <- s2_rebuild(
    result,
    options = s2_options(
      snap_radius = tolerance
    )
  )
  
  simple_boundary <- s2_rebuild(
    s2_boundary(result),
    options = s2_options(
      snap_radius = tolerance,
      edge_type = "undirected"
    )
  )
  
  simple_empty <- s2_is_empty(simple)
  simple_boundary_empty <- s2_is_empty(simple_boundary)
  simple[simple_boundary_empty] <- original[simple_boundary_empty]
  simple[simple_empty & !simple_boundary_empty] <- 
    original[simple_empty & !simple_boundary_empty]
  
  simple
}

geos::geos_convex_hull("MULTIPOINT (3.6 43.2, 0 0, 3.61 43.21)")
#> <geos_geometry[1]>
#> [1] <POLYGON ((0 0, 3.6 43.2, 3.61 43.21, 0 0))>
s2_convex_hull("MULTIPOINT (3.6 43.2, 0 0, 3.61 43.21)")
#> <s2_geography[1]>
#> [1] <POLYGON ((0 0, 3.61 43.21, 3.6 43.2, 0 0))>
fix(
  s2_convex_hull("MULTIPOINT (3.6 43.2, 0 0, 3.61 43.21)"), 
  "MULTIPOINT (3.6 43.2, 0 0, 3.61 43.21)"
)
#> <s2_geography[1]>
#> [1] <POLYGON ((0 0, 3.61 43.21, 3.6 43.2, 0 0))>


geos::geos_convex_hull("POINT (0 0)")
#> <geos_geometry[1]>
#> [1] <POINT (0 0)>
s2_convex_hull("POINT (0 0)")
#> <s2_geography[1]>
#> [1] <POLYGON ((0 0, -5.72949748e-14 3.03663366e-16, -3.03663366e-16 -5.72949748e-14, 0 0))>
fix(s2_convex_hull("POINT (0 0)"), "POINT (0 0)")
#> <s2_geography[1]>
#> [1] <POINT (0 0)>


geos::geos_convex_hull("LINESTRING (0 0, 3.6 43.2)")
#> <geos_geometry[1]>
#> [1] <LINESTRING (0 0, 3.6 43.2)>
s2_convex_hull("LINESTRING (0 0, 3.6 43.2)")
#> <s2_geography[1]>
#> [1] <POLYGON ((1.51774324 21.6094428, 3.6 43.2, 0 0, 1.51774324 21.6094428))>
fix(s2_convex_hull("LINESTRING (0 0, 3.6 43.2)"), "LINESTRING (0 0, 3.6 43.2)")
#> <s2_geography[1]>
#> [1] <LINESTRING (0 0, 3.6 43.2)>

^{Created on 2022-01-16 by the reprex package (v2.0.1)}

[paleolimbot]

BigQuery does this:

SELECT ST_CONVEXHULL(ST_GEOGFROMTEXT("POINT (0 1)"))

POINT(-5.73037018596824e-14 1)

SELECT ST_CONVEXHULL(ST_GEOGFROMTEXT("LINESTRING (0 0, 3.6 43.2)"))

LINESTRING(3.6 43.2, 1.51774323905594 21.6094428417663, 0 0)

[spiry34]

I really appreciate your work, and returning a polygon is definitely the most stable option for many usages.
I will still try to add a dedicated function able to return either a point/line/polygon depending on s2loop contents but I didn't succeed yet (I was just working one more bit on this). This will allow to fit standards. Anyway this will be just an add-on and will not modify your current implementation.
Glad to see soon this new version on the CRAN :-)
Many thanks for your contribution, all the best,
Sylvain

[paleolimbot]

I'm about to merge this one, but if you do get a working implementation feel free to PR it in!