Better extract

[rafaqz]

This PR mostly adds new keywords to extract

@tiemvanderdeure feel like reviewing this thing?

I made it return a vector - this isn't breaking as its still an iterable and we never promised the exact type or iterable.

I took the opportunity to polish the (pretty basic) implementation a little. As well as the new keywords, extract now properly handles flattening multiple polygons/feature collections/nested vectors of geoms so you always get a single "table" return value, and handles empty or all-missing inputs. The only known bug is mixed points and geometries in one vector or will fail. This is noted in the doc, and its kinda rare and not worth the effort right now.

Also once I squash the commits together you wont just be on that one empty commit but the whole thing, its just to get your name on the list.

[codecov-commenter]

Codecov Report

Attention: 17 lines in your changes are missing coverage. Please review.

Comparison is base (ffa56ec) 80.61% compared to head (d9570c4) 80.60%.
Report is 4 commits behind head on main.

❗ Current head d9570c4 differs from pull request most recent head 3f27f18. Consider uploading reports for the commit 3f27f18 to get more accurate results

Files	Patch %	Lines
src/methods/extract.jl	76.05%	17 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #565      +/-   ##
==========================================
- Coverage   80.61%   80.60%   -0.02%     
==========================================
  Files          57       57              
  Lines        4127     4212      +85     
==========================================
+ Hits         3327     3395      +68     
- Misses        800      817      +17     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[rafaqz]

@visr have you seen this anywhere else? we are getting failures in GDAL.jl gdalsetprojection only on macosx:

 Got exception outside of a @test
  GDALError (CE_Failure, code 1):
  	Invalid TOWGS84 node

See here: https://github.com/rafaqz/Rasters.jl/actions/runs/6697436007/job/18560020921#step:5:949

Linux tests are passing fine.

[visr]

Odd that this only happens on MacOS. I haven't seen this error before.

[tiemvanderdeure]

No point in using collect on a Vector, right?

Suggested change

	# use `extract` to get values for all layers at each observation point.
	# We `collect` to get a `Vector` from the lazy iterator.
	collect(extract(st, pnts))
	collect(extract(st, pnts; skipmissing=true))
	# use `extract` to get values for all layers at each observation point.
	extract(st, pnts; skipmissing=true)

[tiemvanderdeure]

Handling missings seems to work wonderfully when giving missing, but behaves unpredictably with NamedTuples containing missing.

E.g.

using Rasters
myrast = Raster(rand(X(10), Y(10)))
extract(myrast, ((X = missing , Y = missing), (X= 1, Y = 10))) # work but return Vector{Any}
extract(myrast, ((X= 1, Y = 10), (X = missing , Y = missing))) # errors
extract(myrast, ((X = missing , Y = 10),)) # Gives a StackOverflowError!

Or is this also what you mean by mixed geometries? Any point data that comes from a table and has some missing coordinates will look like this.

Maybe a solution could be to use a function that also skips geoms that contain missing instead of Base.skipmissing? It should be reasonable to treat something like (X = missing, Y = missing) in the exact same way as missing

I'll have time to look at this more tomorrow.

[rafaqz]

We should force the return type to be at least Vector{NamedTuple{keys,Any}}

We are not actually using Base.skipmissing, but filter doing pretty much what you suggest whenskipmissing=true is passed in. Otherwise I think people can just filter them later as a dataframe however they like.

[rafaqz]

@tiemvanderdeure handling this isn't possible because your points with missing coordinates are not valid GeoInterface geometries:

extract(myrast, ((X = missing , Y = missing), (X= 1, Y = 10)))

And to throw a better error there we need to disallow arbitrary objects like you tuple of points, its limited to an AbstractArray now