workspace() and LastMain regression

[davidagold]

@maleadt and I uncovered a regression in the workspace() behavior. Compare the behavior of calling the constructors of Foo and Bar after workspace():

               _
   _       _ _(_)_     |  A fresh approach to technical computing
  (_)     | (_) (_)    |  Documentation: http://docs.julialang.org
   _ _   _| |_  __ _   |  Type "?help" for help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 0.5.0-rc0+110 (2016-08-01 16:59 UTC)
 _/ |\__'_|_|_|\__'_|  |  Commit e2d2925* (1 day old master)
|__/                   |  x86_64-apple-darwin15.5.0

julia> type Foo
          foo::Vector{Float32}
       end

julia> Foo([1.])
Foo(Float32[1.0])

julia> type Bar
          foo::String
       end

julia> Bar("bar")
Bar("bar")

julia> workspace()

julia> type Foo
          foo::Vector{Float32}
       end
WARNING: Method definition (::Type{Main.Foo})(Array{Float32, 1}) in module Main at REPL[1]:2 overwritten in module Main at REPL[7]:2.

julia> Foo([1.])
LastMain.Foo(Float32[1.0])

julia> type Bar
          foo::String
       end
WARNING: Method definition (::Type{Main.Bar})(String) in module Main at REPL[3]:2 overwritten in module Main at REPL[10]:2.

julia> Bar("bar")
Bar("bar")

However, on 0.4.6:

               _
   _       _ _(_)_     |  A fresh approach to technical computing
  (_)     | (_) (_)    |  Documentation: http://docs.julialang.org
   _ _   _| |_  __ _   |  Type "?help" for help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 0.4.6 (2016-06-19 17:16 UTC)
 _/ |\__'_|_|_|\__'_|  |  Official http://julialang.org/ release
|__/                   |  x86_64-apple-darwin13.4.0

julia> type Foo
          foo::Vector{Float32}
       end

julia> Foo([1.])
Foo(Float32[1.0f0])

julia> type Bar
          foo::String
       end
WARNING: Base.String is deprecated, use AbstractString instead.
  likely near no file:0
WARNING: Base.String is deprecated, use AbstractString instead.
  likely near no file:0

julia> Bar("bar")
WARNING: Base.String is deprecated, use AbstractString instead.
  likely near no file:0
WARNING: Base.String is deprecated, use AbstractString instead.
  likely near no file:0
WARNING: Base.String is deprecated, use AbstractString instead.
  likely near no file:0
Bar("bar")

julia> workspace()

julia> type Foo
          foo::Vector{Float32}
       end

julia> Foo([1.])
Foo(Float32[1.0f0])

julia> type Bar
          foo::String
       end
WARNING: Base.String is deprecated, use AbstractString instead.
  likely near no file:0
WARNING: Base.String is deprecated, use AbstractString instead.
  likely near no file:0

julia> Bar("bar")
WARNING: Base.String is deprecated, use AbstractString instead.
  likely near no file:0
WARNING: Base.String is deprecated, use AbstractString instead.
  likely near no file:0
WARNING: Base.String is deprecated, use AbstractString instead.
  likely near no file:0
Bar("bar")

Also, on 0.5,

               _
   _       _ _(_)_     |  A fresh approach to technical computing
  (_)     | (_) (_)    |  Documentation: http://docs.julialang.org
   _ _   _| |_  __ _   |  Type "?help" for help.
  | | | | | | |/ _` |  |
  | | |_| | | | (_| |  |  Version 0.5.0-rc0+110 (2016-08-01 16:59 UTC)
 _/ |\__'_|_|_|\__'_|  |  Commit e2d2925* (1 day old master)
|__/                   |  x86_64-apple-darwin15.5.0

julia> type Foo end

julia> workspace()

julia> type Foo end
WARNING: Method definition (::Type{Main.Foo})() in module Main at REPL[1]:1 overwritten in module Main at REPL[3]:1.

julia> @which Foo()
LastMain.Foo() at REPL[3]:1

julia> typeof(Foo()) == Main.Foo
true

julia> typeof(Foo()) == LastMain.Foo
false

[maleadt]

Bisected the first issue (ie. Foo resolving to LastMain) to #17405.

[JeffBezanson]

This is probably due to this line in toplevel.c:

    if (old_main) // don't block continued loading of incremental caches
        jl_main_module->uuid = old_main->uuid;

Giving 2 modules the same uuid doesn't bode well for telling things apart. @vtjnash Can we live without this?

[vtjnash]

No, as the comment says, this line is critical for being able to use using after calling workspace. There's nothing wrong with modules having the same id or for object_ids to be non-unique. Why is printing failing?

[JeffBezanson]

The type cache is equating two types that it shouldn't due to the object_id collision. While object_id necessarily has collisions since it might hash a large amount of data, it would be nice to avoid collisions this blatant. For now I'll see if I can make the type cache robust to this.