mapreducedim attempts to compute f(zero(T)), but it may throw

[simonster]

This line assumes that f(zero(T)) is safe to call. This is not always the case:

julia> sum(x->sqrt(x-1), ones(5), 1)
ERROR: DomainError:
sqrt will only return a complex result if called with a complex argument.
try sqrt (complex(x))
 in sqrt at math.jl:133
 in anonymous at none:1
 in reducedim_init at reducedim.jl:96
 in sum at reducedim.jl:255

[timholy]

I suppose we could put that section in a try, and fall back to

julia/base/reducedim.jl

Lines 100 to 101 in 39d8aef

	z = zero(sum(f, A))
	Tr = typeof(z)

in the catch?

[StefanKarpinski]

Probably better to check applicability since inference can reason about that.

[timholy]

We can check applicability, but sqrt(-1) isn't something inference can reason about. I guess we need both?

[simonster]

If !isempty(A), maybe we could just use f(A[1]) here? Then we wouldn't have to compute the reduction twice when 0 isn't in the domain of f.

[JeffBezanson]

mapreduce here has the best current approach I think:

julia/base/reduce.jl

Line 133 in 97b1870

n = Int(length(A))

I can't believe we're computing the whole reduction twice in some cases. I don't know how I missed that. What's more, if zero is not applicable, the code still tries to do zero(sum(f,A)). Sure, it might be defined for the result of sum and not for T, but if we're bothering to check, why make that assumption?

Generally method_exists is a code smell. It's also irrelevant to this issue. While reduce has a little leeway in its definition, anything involving map is quite rigid: the argument function should only be called on elements of the passed collection.

[inkydragon]

Looks like it's fixed on 1.6.3?

julia> sum(x->sqrt(x-1), ones(5); init=1)
1.0

julia> versioninfo()
Julia Version 1.6.3
Commit ae8452a9e0 (2021-09-23 17:34 UTC)
Platform Info:
  OS: Windows (x86_64-w64-mingw32)
  CPU: Intel(R) Core(TM) i5-9400F CPU @ 2.90GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-11.0.1 (ORCJIT, skylake)

[brenhinkeller]

Yeah, seems to be fixed nowadays; I think this is safe to close but feel free to reopen if not!

julia> sum(x->sqrt(x-1), ones(5); init=1)
1.0

julia> sum(x->sqrt(x-1), ones(5))
0.0

julia> versioninfo()
Julia Version 1.8.2
Commit 36034abf260 (2022-09-29 15:21 UTC)
Platform Info:
  OS: macOS (arm64-apple-darwin21.3.0)
  CPU: 10 × Apple M1 Max
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-13.0.1 (ORCJIT, apple-m1)
  Threads: 1 on 8 virtual cores