Use type wrapper directly rather than typename in FieldError

[MasonProtter]

Motivation

FieldError currently prints nameof(exc.type) which just gives a simple symbol for a type's name when a user tries to access a non-existant field.

1. In https://discourse.julialang.org/t/better-error-message-for-modified-structs-in-julia-1-12/129265 it was argued that this is confusing when a type is redefined since the user isn't told that they're using an old instance, e.g.

julia> struct Point end

julia> p = Point();

julia> struct Point{T}
           x::T
           y::T
       end

julia> p.x
ERROR: FieldError: type Point has no field `x`; Point has no fields at all.
Stacktrace:
 [1] getproperty(x::@world(Point, 38528:38531), f::Symbol)
   @ Base ./Base_compiler.jl:54
 [2] top-level scope
   @ REPL[9]:1

The argument is that this should say something like FieldError: type @world(Point, 38528:38531) has no field x. Though, I will note that the world-info is available in the stacktrace, so it is in some sense redundant.

2. Relatedly, namespacing info is also lost because of the same mechanism.

julia> module Foo
       struct Point end
       p = Point()
       end;

julia> Foo.p.y
ERROR: FieldError: type Point has no field `y`; Point has no fields at all.
Stacktrace:
 [1] getproperty(x::Main.Foo.Point, f::Symbol)
   @ Base ./Base_compiler.jl:54
 [2] top-level scope
   @ REPL[11]:1

Similarly, it says Point rather than Main.Foo.Point in the error message. This info is also available in the stacktrace.

This PR

This PR simply replaced nameof with .name.wrapper to get the unparameterized version of the type so that it can be printed with the @world decorations and namespace as needed:

julia> struct Point end

julia> p = Point();

julia> struct Point{T}
           x::T
           y::T
       end

julia> p.x
ERROR: FieldError: type @world(Point, 38558:38559) has no field `x`; @world(Point, 38558:38559) has no fields at all.
Stacktrace:
 [1] getproperty(x::@world(Point, 38558:38559), f::Symbol)
   @ Base ./Base_compiler.jl:54
 [2] top-level scope
   @ REPL[24]:1

and

julia> module Foo
       struct Point end
       p = Point()
       end;

julia> Foo.p.y
ERROR: FieldError: type Main.Foo.Point has no field `y`; Main.Foo.Point has no fields at all.
Stacktrace:
 [1] getproperty(x::Main.Foo.Point, f::Symbol)
   @ Base ./Base_compiler.jl:54
 [2] top-level scope
   @ REPL[20]:1

Discussion / doubts

I'm a bit conflicted on whether or not this is a good idea. On one hand, it clutters the error message with info you could get by simply reading the stacktrace. On the other hand, many users are allergic to reading stacktraces and report confusion with things like this.

I'm interested if people have thoughts on the mechanism here, or if there's a better way to surface the worldage problem / namespacing stuff via hints.

[MasonProtter]

One further reason in favour of this which was mentioned on Discourse is that the stacktrace info is not always present if there was inlining:

julia> mutable struct Point
           x::Float64
           y::Float64
       end

julia> const a = Point(1.0, 2.0);

julia> mutable struct Point
           x::Float64
           y::Float64
           z::Float64
       end

julia> @inline f() = a.z;

Before this PR:

julia> f()
ERROR: FieldError: type Point has no field `z`, available fields: `x`, `y`
Stacktrace:
 [1] getproperty
   @ ./Base_compiler.jl:54 [inlined]
 [2] f()
   @ Main ./REPL[4]:1
 [3] top-level scope
   @ REPL[5]:1

After this PR:

julia> f()
ERROR: FieldError: type @world(Point, 38523:38526) has no field `z`, available fields: `x`, `y`
Stacktrace:
 [1] getproperty
   @ ./Base_compiler.jl:54 [inlined]
 [2] f()
   @ Main ./REPL[4]:1
 [3] top-level scope
   @ REPL[8]:1

[MasonProtter]

Triage is in favour