Infinite recursion in div for Real

[iamed2]

#33040 caused this

The fallback div definition used to apply to T <: Real but is now for T <: AbstractFloat, which leaves out types such as FixedPointDecimal. This causes an infinite promote recursion.

julia> using FixedPointDecimals

julia> a = FixedDecimal{Int,1}(1.3)
FixedDecimal{Int64,1}(1.3)

julia> div(a, a, RoundToZero)
ERROR: StackOverflowError:
Stacktrace:
 [1] div(::FixedDecimal{Int64,1}, ::FixedDecimal{Int64,1}, ::RoundingMode{:ToZero}) at ./div.jl:234 (repeats 79985 times)

[Keno]

Yes, the removal of the fallback was deliberate. It was widely inaccurate for various types we had. FixedPointDecimals should provide this method for its particular type. Of course we should fix this to give a better error.

[Keno]

Also, I don't think this is a regression, at least not as stated. Three argument div is new as of that commit.

[NHDaly]

Ah, @Keno, it turns out that there was actually a regression, since the default implementation of ÷ had changed in #33040 to call 3-argument div. So this was now failing, where it didn't fail before:

julia> let x = FixedDecimal{Int,2}(1)
           x ÷ 2
       end
ERROR: StackOverflowError:
Stacktrace:
 [1] div(::FixedDecimal{Int64,2}, ::FixedDecimal{Int64,2}, ::RoundingMode{:ToZero}) at ./div.jl:234 (repeats 79998 times)

But your new fallback definition in #33679 does indeed fix the regression, so thanks! :)

julia> let x = FixedDecimal{Int,2}(1)
           x ÷ 2
       end
FixedDecimal{Int64,2}(0.00)

✅

[Keno]

Yeah, I saw that after making that comment and testing my fixed so I fixed it right along the original issue.