Odd subset of vector types can not be written to disk

[alex-s-gardner]

These types all error: Vector{Int8}, Vector{UInt16}, Vector{Int16}, Vector{Float32}

These types do not: UInt8(), Int8(), UInt16(), Int16(), Float32(), Float64(), Vector{UInt8}, Vector{Float64}

This is likely an ArchGDAL issue but thought I'd start the discussion here.

import GeoInterface as GI
using DataFrames
using GeoDataFrames

outfile = joinpath(tempdir(), "test.geojson");
pt = GI.Point(1,1, crs=GFT.EPSG(4326))
colvals = [UInt8(1), Int8(1), UInt16(1), Int16(1), Float32(1), Float64(1), [UInt8.([1,2])], [Int8.([1,2])], [UInt16.([1,2])], [Int16.([1,2])], [Float32.([1,2])], [Float64.([1,2])]]

for cv in colvals
    df = DataFrame(geometry=pt, test=cv)
    try
        GeoDataFrames.write(outfile, df)
    catch e
        println("causes write error: $cv")
    end
end

causes write error: Vector{Int8}[[1, 2]]
causes write error: Vector{UInt16}[[0x0001, 0x0002]]
causes write error: Vector{Int16}[[1, 2]]
causes write error: Vector{Float32}[[1.0, 2.0]]

Typical error message looks like this:

ERROR: MethodError: no method matching setfield!(::ArchGDAL.Feature, ::Int64, ::Vector{Int8})
You may have intended to import Base.setfield!
The function `setfield!` exists, but no method is defined for this combination of argument types.

Closest candidates are:
  setfield!(::ArchGDAL.AbstractFeature, ::Integer, ::DateTime, ::Int64)
   @ ArchGDAL ~/.julia/packages/ArchGDAL/aXTAq/src/ogr/feature.jl:713
  setfield!(::ArchGDAL.AbstractFeature, ::Integer, ::DateTime)
   @ ArchGDAL ~/.julia/packages/ArchGDAL/aXTAq/src/ogr/feature.jl:713
  setfield!(::ArchGDAL.AbstractFeature, ::Integer, ::Vector{UInt8})
   @ ArchGDAL ~/.julia/packages/ArchGDAL/aXTAq/src/ogr/feature.jl:704
  ...

Stacktrace:
  [1] (::GeoDataFrames.var"#22#27"{Vector{…}, DataFrameRow{…}, Vector{…}})(feature::ArchGDAL.Feature)
    @ GeoDataFrames ~/.julia/packages/GeoDataFrames/eP9Sg/src/io.jl:224
  [2] addfeature(f::GeoDataFrames.var"#22#27"{Vector{…}, DataFrameRow{…}, Vector{…}}, layer::ArchGDAL.FeatureLayer)
    @ ArchGDAL ~/.julia/packages/ArchGDAL/aXTAq/src/ogr/featurelayer.jl:490
  [3] (::GeoDataFrames.var"#20#25"{…})(layer::ArchGDAL.FeatureLayer)
    @ GeoDataFrames ~/.julia/packages/GeoDataFrames/eP9Sg/src/io.jl:210
  [4] createlayer(::GeoDataFrames.var"#20#25"{…}; kwargs::@Kwargs{…})
    @ ArchGDAL ~/.julia/packages/ArchGDAL/aXTAq/src/context.jl:268
  [5] (::GeoDataFrames.var"#19#24"{…})(spatialref::ArchGDAL.SpatialRef)
    @ GeoDataFrames ~/.julia/packages/GeoDataFrames/eP9Sg/src/io.jl:183
  [6] newspatialref(::GeoDataFrames.var"#19#24"{…}; kwargs::@Kwargs{})
    @ ArchGDAL ~/.julia/packages/ArchGDAL/aXTAq/src/context.jl:268
  [7] newspatialref
    @ ~/.julia/packages/ArchGDAL/aXTAq/src/context.jl:265 [inlined]
  [8] #18
    @ ~/.julia/packages/GeoDataFrames/eP9Sg/src/io.jl:177 [inlined]
  [9] create(f::GeoDataFrames.var"#18#23"{…}, args::String; kwargs::@Kwargs{…})
    @ ArchGDAL ~/.julia/packages/ArchGDAL/aXTAq/src/context.jl:268
 [10] create
    @ ~/.julia/packages/ArchGDAL/aXTAq/src/context.jl:265 [inlined]
 [11] write(fn::String, table::DataFrame; layer_name::String, crs::Nothing, driver::Nothing, options::Dict{…}, geom_columns::Tuple{…}, chunksize::Int64, kwargs::@Kwargs{})
    @ GeoDataFrames ~/.julia/packages/GeoDataFrames/eP9Sg/src/io.jl:176
 [12] write(fn::String, table::DataFrame)
    @ GeoDataFrames ~/.julia/packages/GeoDataFrames/eP9Sg/src/io.jl:112
 [13] top-level scope
    @ ./Untitled-1:12
Some type information was truncated. Use `show(err)` to see complete types.

[evetion]

This is indeed a limitation by GDAL. Would you expect a better error, or an automatic conversion?

[alex-s-gardner]

Are you sure about Vector{Float32} that seems like a pretty basic type... looking through the GDAL docs I can't readily determine what types are supported. I see that GeoTiff these types are supported.

I would automatic conversion with a warning?

[evetion]

Yeah, gdal defines a list of doubles (float64) only. https://gdal.org/en/latest/doxygen/ogr__api_8h.html#a83a56d01128bf247016c139faed157be. ArchGDAL follows it to the letter, but should probably be more lenient and just convert. All your other types can also be converted without data loss. You just won't get them back as the same type.