Unexpected behavior in `hvcat` on Julia 1.12

[francis-biocurie]

Old behavior

On julia v1.11.5:

[zeros(1, 0) zeros(1,0); zeros(0,0) zeros(0, 0)]

gives, as expected:

1×0 Matrix{Float64}

New behavior

On julia v1.12-beta4, it gives this unexpected results:

2×0 Matrix{Float64}

and, weirdly, this:

row1 = [zeros(1, 0) zeros(1,0)]
row2 = [zeros(0,0) zeros(0, 0)]
[row1; row2]

gives a different result:

1×0 Matrix{Float64}

which is incoherent, in my opinion.

[mbauman]

Thanks for the report! Writing it as the transpose also yields a 1×0:

julia> [zeros(1,0); zeros(0,0);; zeros(1,0); zeros(0,0)]
1×0 Matrix{Float64}

[adienes]

I bet it's #39729

[BioTurboNick]

Most likely it's an error in the logic of the underlying hvncat code I wrote with respect to 0-dimension elements, and since that PR switched hvcat to rely on hvncat, the error is apparent. I can dig into it.