Improving Printf error messages

[Seelengrab]

This replaces all "invalid format string: $f style messages with more informative error messages about what exactly is wrong and where it goes wrong:

There are also some parts that I couldn't yet figure out - I'll comment these specifics down below for a conversation.

[Seelengrab]

Oh boy, looks like adding that additional error message that should cover what the docstring of @printf specifies broke something. Something about sysimg.jl and LibGit2?

LoadError("sysimg.jl", 19, 
    LoadError("/buildworker/worker/doctest_linux64/build/usr/share/julia/stdlib/v1.9/LibGit2/src/LibGit2.jl", 3,
    LoadError("/buildworker/worker/doctest_linux64/build/usr/share/julia/stdlib/v1.9/LibGit2/src/signature.jl", 58,
    LoadError("/buildworker/worker/doctest_linux64/build/usr/share/julia/stdlib/v1.9/LibGit2/src/signature.jl", 62,
        ArgumentError("First argument to `@printf` has to be either an `<: IO` to print to or a format `String`.")))))

[Seelengrab]

Ok, I forgot that this is a macro and of course the first argument is either a literal String or a Symbol. I've removed the check, so it should work now. We'll get a MethodError later on with Printf.format, which is also not a good error, so I may insert a $io isa IO || throw(ArgumentError(...)) in the returned expression later.

[Seelengrab]

Ok, no idea why the buildkite/julia-master job is still running (maybe macOS stuff?), but it seems like the change in error messages didn't break anything, which is good.

Looking through the existing tests, I'm not sure all of these are actually covered. Should I write some more to make sure they're explicitly covered?

[Seelengrab]

Alright, the error message is much nicer now:

and it's also smart about not going too far back:

The only thing left now are the BigFloat things, which I'm still not too sure about how to best communicate.

[halleysfifthinc]

Where is the LoadError part of the message coming from? InvalidFormatStringError is a new/independent <:Exception type with no relation to LoadError that I can see.

The extra new line between the error message and stacktrace might not be needed to sufficiently visually separate the text. With the indenting of the string and location arrow, I think there the error message will still be visually distinct from the stacktrace.

[Seelengrab]

LoadError comes from it happening during loading of the code, specifically during parsing (as that's when the code is run due to it happening in the macro).

Yeah, I thought I'd try it first to see if I like it. I'm not sold either way 🤷

[Seelengrab]

Alright, I've changed the \t to 4 spaces and clarified the error messages for the libmpfr case, as well as documented what that __BIG_FLOAT_MAX__ constant is for. Should be good to merge now.

[Seelengrab]

Error in testset Profile:
Error During Test at C:\buildbot\worker-tabularasa\tester_win32\build\share\julia\stdlib\v1.9\Profile\test\allocs.jl:109
Test threw exception
Expression: length((Allocs.fetch()).allocs) > 10
OutOfMemoryError()

Huh - unrelated!

[Seelengrab]

Does this need anything else (rebase, rerun of CI) or can this be merged, if there are no other comments?