Handling of `err` and `ans` global variables in REPL

[BioTurboNick]

Copying this from @mcabbott in merged #40642 to a new issue so it doesn't get lost.

===================
Is it possible that err, and in fact ans, should have a value on startup? This would prevent this:

 _/ |\__'_|_|_|\__'_|  |  Official https://julialang.org/ release
|__/                   |

julia> ans(x) = "function";

julia> 1+1
ERROR: invalid redefinition of constant ans

julia> 2;
ERROR: invalid redefinition of constant ans

in favour of this:

 _/ |\__'_|_|_|\__'_|  |  Official https://julialang.org/ release
|__/                   |

julia> 1+1
2

julia> ans(x) = "function";
ERROR: cannot define function ans; it already has a value

=======================

[JeffBezanson]

Yes, I think it would be reasonable to initialize them to nothing.

[StefanKarpinski]

Doesn't that turn the introduction of err into a breaking change? (Albeit only in the REPL.)

[BioTurboNick]

Could an alternative be to set a global inside the REPL module and export it only if not already in Main?

[JeffBezanson]

We have changed behavior in the repl before, e.g. the scope change. Fortunately may not need to in this case.

Could an alternative be to set a global inside the REPL module and export it only if not already in Main?

Yes I think that works --- if you define it first, you just get your variable and the REPL modifies its own variable which you can't see. But if you request ans or err without defining them, the using logic kicks in to import the variables. Good idea!

[vtjnash]

I made this version of the change, but I don't think I like it master...vtjnash:jn/MainInclude-ans-err

In particular, it means that ans gets hidden from names(Main)

So possibly we want to just declare global ans = nothing; global err = nothing early in REPL startup instead

[JeffBezanson]

Or the most benign version would be to check whether ans is a constant before trying to set it in the REPL, and just skipping it in that case to avoid the error/warning.