Breaking change in vcat between 0.6.2 and 0.6.3

[a-parton]

Between 0.6.2 and 0.6.3 an error appears to have been introduced when concatenating an array with an empty array. I've looked through the commits and the only change that appears to affect vcat is an improvement to the support of abstract arrays, which makes me think that this might not have been a deliberate move.

In 0.6.3:

julia> versioninfo()
Julia Version 0.6.3
Commit d55cadc350 (2018-05-28 20:20 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Core(TM) i5-7200U CPU @ 2.50GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Haswell)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.9.1 (ORCJIT, broadwell)

julia> vcat([],[1.0 2.0 3.0])
ERROR: ArgumentError: number of columns of each array must match (got (1, 3))
Stacktrace:
 [1] typed_vcat(::Type{Any}, ::Array{Any,1}, ::Array{Float64,2}) at ./abstractarray.jl:1141
 [2] vcat(::Array{Any,1}, ::Array{Float64,2}) at ./sparse/sparsevector.jl:990

and in 0.6.2:

julia> versioninfo()
Julia Version 0.6.2
Commit d386e40c17 (2017-12-13 18:08 UTC)
Platform Info:
  OS: macOS (x86_64-apple-darwin14.5.0)
  CPU: Intel(R) Xeon(R) CPU E5-1680 v2 @ 3.00GHz
  WORD_SIZE: 64
  BLAS: libopenblas (USE64BITINT DYNAMIC_ARCH NO_AFFINITY Sandybridge)
  LAPACK: libopenblas64_
  LIBM: libopenlibm
  LLVM: libLLVM-3.9.1 (ORCJIT, ivybridge)

julia> vcat([],[1.0 2.0 3.0])
1×3 Array{Any,2}:
 1.0  2.0  3.0

[mbauman]

Thanks for the report! This is also the case on 0.7, and I think this was indeed an unintended break in behavior for the backport. A quick bisect confirms that 72c3e62 was indeed the culprit, intending to just fix vcat(1:8, fill(1, (4, 1))).

Empty arrays still work, but they must be 0x3:

julia> vcat(reshape([],0,3),[1.0 2.0 3.0])
1×3 Array{Any,2}:
 1.0  2.0  3.0

Should we figure out a way to restore the old behavior on 0.6.4? We're also a little inconsistent in applying this rule:

julia> vcat(zeros(0), rand(2,2,2))
2×2×2 Array{Float64,3}:
# …

julia> vcat(zeros(0,2), rand(2,2,2))
ERROR: DimensionMismatch("tried to assign 0×2 array to 0×2×2 destination")
Stacktrace:
 [1] throw_setindex_mismatch at ./indices.jl:138 [inlined]
 [2] setindex_shape_check(::Array{Float64,2}, ::Int64, ::Int64, ::Int64) at ./indices.jl:164
 [3] macro expansion at ./multidimensional.jl:638 [inlined]
 [4] _unsafe_setindex!(::IndexLinear, ::Array{Float64,3}, ::Array{Float64,2}, ::UnitRange{Int64}, ::UnitRange{Int64}, ::UnitRange{Int64}) at ./multidimensional.jl:631
 [5] _setindex! at ./multidimensional.jl:625 [inlined]
 [6] setindex! at ./abstractarray.jl:997 [inlined]
 [7] __cat(::Array{Float64,3}, ::Tuple{Int64,Int64,Int64}, ::Tuple{Bool}, ::Array{Float64,2}, ::Vararg{Any,N} where N) at ./abstractarray.jl:1363

[StefanKarpinski]

Should we figure out a way to restore the old behavior on 0.6.4?

Yes — 0.6.4 should endeavor to match the behavior of 0.6.0 except for bugs. If 0.6.3 deviated from that in some way, 0.6.4 should revert the deviation.

[ararslan]

I've pushed a reversion for that change in #27621.

[JeffBezanson]

This has been fixed in a 0.6 point release. In 1.0 we give the error, but it is arguably correct. In other cases I believe we'd need to give more errors, which would be a breaking change. That can be discussed separately.