note on flatten

Author: mcabbott

src/flatten.jl

@@ -10,6 +10,10 @@ Should be equal to `collect(Iterators.flatten(iter))`, but faster.
 Acting on a vector of vectors, this is equal to `vcat(iter...)` or `reduce(vcat, iter)`,
 and to `vec(stack(iter))` when this is allowed, i.e. when `allequal(length.(iter))`.
 
+!!! Warning
+   `flatten` is an experiment which got committed here by accident, more or less.
+   I'm not sure it's part of the long-term future of LazyStack.
+
 # Examples
 ```
 julia> v = flatten((1:2, 8:9))
@@ -96,6 +100,8 @@ end
 
 # _typed_flatten(::Type{T}, ::Union{HasShape, HasLength}, A) where {T} = _flatten(collect(A))
 
+# Non-array iterators, whose elements are arrays or tuples, should probably be collected & sent to the above path.
+
 function _typed_flatten(::Type{T}, ::IteratorSize, A) where {T}
     xit = iterate(A)
     nothing === xit && return T[]
@@ -297,4 +303,41 @@ vectors of varying eltype
   min 2.713 μs, mean 9.107 μs (2 allocations, 78.17 KiB)
   min 68.750 μs, mean 77.771 μs (9 allocations, 326.55 KiB)
 
+=#
+
+
+#=
+
+julia> f(x, y) = reduce(vcat, map(current -> searchsorted(x, current), y));  # collected
+
+julia> g(x, y) = mapreduce(current -> searchsorted(x, current), vcat, y);  # pairwise
+
+julia> h(x, y) = flatten(current -> searchsorted(x, current), y);  # append!
+
+julia> let
+           Random.seed!(1)
+           list = sort(rand(1:99, 10^6))
+           needed = unique(sort(rand(1:99, 10)))
+           @btime f($list, $needed)   # reduce(vcat, map(...
+           # @btime g($list, $needed) # mapreduce(...
+           @btime h($list, $needed)   # flatten(...
+           @btime flatten([searchsorted($list, current) for current in $needed])
+       end;
+  min 17.958 μs, mean 74.229 μs (4 allocations, 789.84 KiB)
+  min 19.917 μs, mean 57.096 μs (2 allocations, 807.81 KiB)  # here the guess is good
+  min 19.542 μs, mean 72.105 μs (3 allocations, 789.70 KiB)
+
+julia> let
+           Random.seed!(42)
+           list = sort(rand(1:99, 10^6))
+           needed = unique(sort(rand(1:99, 10)))
+           @btime f($list, $needed)   # reduce(vcat, map(...
+           # @btime g($list, $needed) # mapreduce(...
+           @btime h($list, $needed)   # flatten(...
+           @btime flatten([searchsorted($list, current) for current in $needed])
+       end;
+  min 17.791 μs, mean 67.903 μs (4 allocations, 790.41 KiB)
+  min 37.375 μs, mean 128.942 μs (3 allocations, 2.13 MiB)  # here the guess is not so good
+  min 19.417 μs, mean 77.225 μs (3 allocations, 790.27 KiB)
+
 =#