Performance regression in vcat of heterogeneous AbstractVectors

[jishnub]

On v1.9.2, using FillArrays:

julia> @btime vcat($(Zeros(100)), $(ones(1)));
  122.875 ns (4 allocations: 976 bytes)

julia> @btime vcat($(zeros(100)), $(ones(1)));
  83.161 ns (1 allocation: 896 bytes)

On v1.10.0-beta1:

julia> @btime vcat($(Zeros(100)), $(ones(1)));
  1.467 μs (21 allocations: 1.28 KiB)

julia> @btime vcat($(zeros(100)), $(ones(1)));
  97.178 ns (1 allocation: 896 bytes)

julia> versioninfo()
Julia Version 1.10.0-beta1
Commit 6616549950e (2023-07-25 17:43 UTC)
Platform Info:
  OS: Linux (x86_64-linux-gnu)
  CPU: 8 × 11th Gen Intel(R) Core(TM) i5-1135G7 @ 2.40GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-15.0.7 (ORCJIT, tigerlake)
  Threads: 1 on 8 virtual cores
Environment:
  LD_LIBRARY_PATH = :/usr/lib/x86_64-linux-gnu/gtk-3.0/modules
  JULIA_EDITOR = subl

The performance is similar on the current nightly v"1.11.0-DEV.243"

[jishnub]

This appers fixed on master:

julia> @btime vcat($(zeros(100)), $(ones(1)));
  123.143 ns (1 allocation: 896 bytes)

julia> @btime vcat($(Zeros(100)), $(ones(1)));
  184.479 ns (1 allocation: 896 bytes)

julia> VERSION
v"1.11.0-DEV.387"

Compared to v1.9.3:

julia> @btime vcat($(zeros(100)), $(ones(1)));
  116.497 ns (1 allocation: 896 bytes)

julia> @btime vcat($(Zeros(100)), $(ones(1)));
  170.280 ns (4 allocations: 976 bytes)

The difference of a few ns is probably noise.

It would be good to isolate the commit and backport it (I'll see if I can find it)

[jishnub]

Bisect suggests that the regression was introduced by 5a922fa, and the fix comes from c8ad552, which bumped SparseArrays.

The commits are:
JuliaSparse/SparseArrays.jl@99c99b4...54f4b39

I wonder if this may be backported to v1.10?

Reopening this as a reminder to backport