[Error printing] Distinguish having no applicable function and manually thrown MethodErrors

[rapus95]

In order to progress on the long way to perfect, beautiful and userfriendly error messages and to reduce irritations like those:

julia> div(5,2,RoundFromZero)
ERROR: MethodError: no method matching div(::Int64, ::Int64, ::RoundingMode{:FromZero})
Closest candidates are:
  div(::Real, ::Real, ::RoundingMode) at div.jl:247
  div(::Missing, ::Number, ::RoundingMode) at missing.jl:121
  div(::T, ::T) where T<:Union{Int16, Int32, Int64, Int8} at int.jl:230
  ...
Stacktrace:
 [1] div(::Int64, ::Int64, ::RoundingMode{:FromZero}) at .\div.jl:247
 [2] top-level scope at REPL[51]:1

julia> @which div(5,2,RoundFromZero)
div(x::Real, y::Real, r::RoundingMode) in Base at div.jl:247

(This particular error has already a bug report there #34519)
where it clearly looks like there are applicable methods and in fact are, I'd suggest to somehow leverage that information for the user that there indeed are functions which have compatible signature but can't handle that actual signature.

Every time I stumble over such a case I first check it like 3 times whether I overlooked something in the signature but come to the conclusion that it should find its applicable function. Once I dive into the function via @edit I see that a MethodError was thrown manually.

At the very least it looks somewhat inconsistent to me. An idea to tackle that problem would be to introduce a manual flag (naming of that flag is open to discussion) which can be passed to the MethodError constructor and is true by default. Only in the places where the MethodError is thrown because there are no applicable methods at all, that flag should hold the value false. Then, on printing the error, we could easily distinguish those cases.

Even though the steps to solve that error are the same in both cases, namingly, defining the missing function, it helps reducing confusion and increases consistency IMO.

[JeffBezanson]

The manual flag should not be necessary, since as you say we can look at the method table to see that there is a non-ambiguous matching method.