Faster iterating over (almost) empty Dict or Set

[petvana]

It is possible to utilize the existing code 0e12723 defining idxfloor to improve performance of iterating over (almost) empty Dict/Set.

However, this PR cannot solve asymptotic complexity because the current implementation cannot shrink Dict so far (there is a todo for that).

julia/base/dict.jl

Lines 238 to 243 in b10aa56

	if newsz <= oldsz
	# todo: shrink
	# be careful: rehash!() assumes everything fits. it was only designed
	# for growing.
	return d
	end

.

Master

julia> using BenchmarkTools

julia> d = Set{Int}();
julia> @btime minimum(d, init = 0);
  20.737 ns (0 allocations: 0 bytes)

julia> d = Set([1]);
julia> @btime minimum(d, init = 0);
  21.904 ns (0 allocations: 0 bytes)

julia> d = Set(1:10_000); empty!(d);
julia> @btime minimum(d, init = 0);
  6.581 μs (0 allocations: 0 bytes)

julia> d = Set(1:10_000); empty!(d); push!(d,1);
julia> @btime minimum(d, init = 0);
  6.665 μs (0 allocations: 0 bytes)

PR

julia> using BenchmarkTools

julia> d = Set{Int}();
julia> @btime minimum(d, init = 0);
  14.968 ns (0 allocations: 0 bytes)

julia> d = Set([1]);
julia> @btime minimum(d, init = 0);
  16.615 ns (0 allocations: 0 bytes)

julia> d = Set(1:10_000); empty!(d);
julia> @btime minimum(d, init = 0);
  14.933 ns (0 allocations: 0 bytes)

julia> d = Set(1:10_000); empty!(d); push!(d,1);
julia> @btime minimum(d, init = 0);
  821.388 ns (0 allocations: 0 bytes)