don't warn about redefining methods in Main

[stevengj]

This PR closes #18725 by silencing the "Method definition overwritten" warnings when both the new and old methods are in Main.

The warnings are useful for modules, where overwritten methods are usually an error, but are incredibly annoying and confusing during interactive work in the REPL (or IJulia etc.). (Now that #265 is fixed, redefining functions is much less dangerous, so most of the concerns raised in #18725 should now be moot.)

With this PR:

julia> f(x) = 3
f (generic function with 1 method)

julia> f(x) = 12
f (generic function with 2 methods)

julia> f(4)
12

julia> module Foo
           f(x) = 3
           f(x) = 4
       end
WARNING: Method definition f(Any) in module Foo at REPL[7]:2 overwritten at REPL[7]:3.
Foo

[JeffBezanson]

Yes, with #265 fixed I'd say this is ok. Not sure about this, but maybe we should limit this to interactive functions, i.e. functions not in any file?

[stevengj]

@JeffBezanson, people regularly include a file with function definitions repeatedly. (And in IJulia the input cell "In[xx]" is treated as the "filename", so that might be affected too.)

[KristofferC]

Previous implementation and discussion at #18746 (also includes test)

[tkelman]

the version with a test was better

[stevengj]

We don't normally test warnings. If we want to do that, we need a better test framework than setting up redirect_stdout manually every time

[tkelman]

The work had been done manually for this case. There are some convenience macros in the Pkg tests that Carlo wrote that would be worth moving to a more general place and making more use of.

[timholy]

Even better, we already have #18165. It's now very easy to do this correctly.

[stevengj]

#19903 shows what I had in mind for a nicer test framework for warnings than manual redirect_stderr calls, and adds a test for this PR.