deprecate repmat() ?

[bjarthur]

given repeat, should repmat be deprecated? the latter cannot handle more than 2 dimensions, and does not have the flexibility provided by the former's inner argument. happy to draft a PR if there is consensus.

[fredrikekre]

repeat is (in the current form) ~1000 times slower than repmat and uses ~40 times more memory:

julia> using BenchmarkTools

julia> for N in (10, 50, 100)
           A = rand(N, N)
           outer = (N, N)
           @btime repmat($A, $N, $N)
           @btime repeat($A, outer = $outer)
       end
  11.313 μs (202 allocations: 84.45 KiB)
  10.518 ms (119017 allocations: 3.11 MiB)
  5.407 ms (5002 allocations: 47.84 MiB)
  7.397 s (79971519 allocations: 1.98 GiB)
  106.325 ms (20002 allocations: 763.55 MiB)
  112.596 s (1294889019 allocations: 31.96 GiB)

[JeffBezanson]

ref JuliaLang/julia#15553

[oscardssmith]

As a temporary solution, could repmat be deprecated, and a fast path added that uses the current repmat code when applicable?

[nalimilan]

repmat is faster, but its return type is also inferrable, which isn't currently the case for repeat. So I'm afraid we'll have to keep it until the compiler improves. Cf. JuliaLang/julia#14082

[johnmyleswhite]

Would the return type be inferrable if we only allowed tuples as arguments for inner and outer and didn't use keyword arguments?

[timholy]

Yes, with those constraints one could make it inferrable.

[nalimilan]

This discussion already happened at JuliaLang/julia#12953 and JuliaLang/julia#14082, and it was decided to keep keyword arguments.

[JeffBezanson]

Closed by JuliaLang/julia#26039