Fix float grouping

NEWS.md

@@ -14,6 +14,8 @@
 
 ## Bug fixes
 
+* fix bug in how `groupby` handles grouping of float columns
+  ([#2791](https://github.com/JuliaData/DataFrames.jl/pull/2791))
 * fix bug in how `issorted` handles custom orderings and improve performance
   of sorting when complex custom orderings are passed
   ([#2746](https://github.com/JuliaData/DataFrames.jl/pull/2746))

src/groupeddataframe/utils.jl

@@ -143,10 +143,28 @@ function refpool_and_array(x::AbstractArray)
         else
             return nothing, nothing
         end
-    elseif x isa AbstractArray{<:Union{Real, Missing}} &&
-        all(v -> ismissing(v) | isinteger(v), x) &&
-        !isempty(skipmissing(x))
-        minval, maxval = extrema(skipmissing(x))
+    elseif x isa AbstractArray{<:Union{Real, Missing}} && length(x) > 0
+        if !(x isa AbstractArray{<:Union{Integer, Missing}})
+            if !all(v -> (ismissing(v) | isinteger(v)) & (!isequal(v, -0.0)), x)
+                return nothing, nothing
+            end
+        end
+        if Missing <: eltype(x)
+            smx = skipmissing(x)
+            y = iterate(smx)
+            y === nothing && return nothing, nothing
+            (v, s) = y
+            minval = maxval = v
+            while true
+                y = iterate(smx, s)
+                y === nothing && break
+                (v, s) = y
+                maxval = max(v, maxval)
+                minval = min(v, minval)
+            end
+        else
+            minval, maxval = extrema(x)
+        end
         ngroups = big(maxval) - big(minval) + 1
         # Threshold chosen with the same rationale as the row_group_slots refpool method:
         # refpool approach is faster but we should not allocate too much memory either

test/grouping.jl

@@ -3710,6 +3710,14 @@ end
     end
 end
 
+@testset "grouping floats" begin
+    @test length(groupby_checked(DataFrame(a=[0.0, -0.0]), :a)) == 2
+    @test getindex.(keys(groupby_checked(DataFrame(a=[3.0, 2.0, 0.0]), :a)), 1) ==
+          [0, 2, 3]
+    @test getindex.(keys(groupby_checked(DataFrame(a=[3.0, 2.0, -0.0]), :a)), 1) ==
+          [3, 2, 0]
+end
+
 @testset "aggregation with matrix of Pair" begin
     df = DataFrame(a=["a", "b","a", "b"], x=1:4, y=11:14)
     gdf = groupby_checked(df, :a)