Inaccurate printing of TypeError message

[englhardt]

While the examples below are clear cases where the user did not RTFM, they are still confusing and we could improve the error message:

isa

julia> isa(1, Int64)
true
julia> isa(Int64, 1)
ERROR: TypeError: in isa, expected Type, got Int64
Stacktrace:
 [1] top-level scope at REPL[5]:1

Int64 is a Type so we should prompt that the user passed the value 1 instead of a Type:
Expected: ERROR: TypeError: in isa, expected Type, got 1.

Union

julia> Union{Real, nothing}
ERROR: TypeError: in Union, expected Type, got Nothing
Stacktrace:
 [1] top-level scope at REPL[8]:1

Similar to isa(Nothing, nothing) that yields a similar error.
Expected: ERROR: TypeError: in isa, expected Type, got nothing.

As far as I can see, this part of the string in the TypeError message is generated by the macro @nospecialize in Base.

julia> versioninfo()
Julia Version 1.5.0-DEV.494
Commit 59d7522eae (2020-03-23 03:46 UTC)
Platform Info:
  OS: Linux (x86_64-pc-linux-gnu)
  CPU: Intel(R) Core(TM) i5-7300U CPU @ 2.60GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-9.0.1 (ORCJIT, skylake)

[jakobnissen]

The behaviour is correct, no? In the first case, the function expects a Type for the second argument, and you passed it 1, which is indeed a Int64. In the second case, it expects a type, and gets nothing, an instance of Nothing.

Maybe it could be phrased better as:
in Union, expected an instance of Type, got an instance of Nothing
?

[StefanKarpinski]

I think the issue is that the error messages, while accurate, could be clearer.