accumulate(op, v0, x) fails on arrays of non-numeric types

[chfin]

The following produces an error:

julia> accumulate((acc, x) -> acc+x[1], 0, [(1,2), (3,4), (5,6)])
ERROR: MethodError: no method matching rcum_promote_type(::##1#2, ::Type{Int64}, ::Type{Tuple{Int64,Int64}})
Closest candidates are:
  rcum_promote_type(::Any, ::Type{T<:Number}) where T<:Number at multidimensional.jl:585
  rcum_promote_type(::Any, ::Type{T}) where T at multidimensional.jl:586
  rcum_promote_type(::Any, ::Type{T}, !Matched::Type{S<:Number}) where {T, S<:Number} at multidimensional.jl:584
  ...
Stacktrace:
 [1] accumulate at ./multidimensional.jl:760 [inlined] (repeats 2 times)
 [2] macro expansion at ./REPL.jl:97 [inlined]
 [3] (::Base.REPL.##1#2{Base.REPL.REPLBackend})() at ./event.jl:73

The same happens for

accumulate((acc, x) -> x, "", ['a', 'b', 'c'])

It seems that accumulate(op, v0, x) tries to guess the type of the output array based on both the type of the initial value and the type of the input array, while it should only consider the first.

Changing this line:

julia/base/multidimensional.jl

Line 1105 in 4440b0f

T = rcum_promote_type(op, typeof(v0), eltype(x))

to:

    T = rcum_promote_type(op, typeof(v0))

seems to fix the issue, but I am not sure whether there is a reason to prefer the original version.

[jw3126]

accumulate(+,0, [1.1])

would fail with the proposed fix.

[simonbyrne]

A full solution to this would probably involve extending #25051 to accumulate

[motjuste]

I am facing this issue while trying cumulative counting along the lines of the following:

accumulate((acc, c) -> acc + (c == 'a' ? 1 : 0),  0, collect("cdaabcaa"))  # ERROR

My workaround for such a case was:

accumulate(+, map(c -> c == 'a' ? 1 : 0, collect("cdaabcaa")))

I would love to know if there is a better approach to do this!

[simonbyrne]

not yet, unfortunately.