don't assume groups are non-empty (#991)

grouping on an empty data frame should not crash. combine can't create additional columns but returning an empty data frame is better than crashing.



Similarly, showing an GroupedDataFrame in the REPL should not assume that it's non-empty.

src/groupeddataframe/grouping.jl

@@ -217,7 +217,7 @@ function combine(ga::GroupApplied)
     gd, vals = ga.gd, ga.vals
     # Could be made shorter with a rep(x, lengths) function
     # See JuliaLang/julia#16443
-    idx = Vector{Int}(sum([size(val, 1) for val in vals]))
+    idx = Vector{Int}(sum(Int[size(val, 1) for val in vals]))
     j = 0
     for i in 1:length(vals)
         n = size(vals[i], 1)

src/groupeddataframe/show.jl

@@ -1,8 +1,10 @@
 function Base.show(io::IO, gd::GroupedDataFrame)
     N = length(gd)
     println(io, "$(typeof(gd))  $N groups with keys: $(gd.cols)")
-    println(io, "First Group:")
-    show(io, gd[1])
+    if N > 0
+        println(io, "First Group:")
+        show(io, gd[1])
+    end
     if N > 1
         print(io, "\n⋮\n")
         println(io, "Last Group:")

test/grouping.jl

@@ -36,4 +36,8 @@ module TestGrouping
     x = pool(collect(1:20))
     df = DataFrame(v1=x, v2=x)
     groupby(df, [:v1, :v2])
+
+    df2 = by(e->1, DataFrame(x=Int64[]), :x)
+    @test size(df2) == (0,1)
+    @test sum(df2[:x]) == 0
 end