Add stacktrace colors to showing MethodErrors and method lists

[Seelengrab]

This basically takes the module coloring logic from printing stacktraces and applies it to method error printing as well. Also contains some formatting to make the "closest candidates" stand out more by having an empty line precede & follow that block.

Before (master)

After (This PR)

Code for producing the above

example.jl

f() = 1
f(x,y,z; f=true) = 3
f(x,y,z,k; f=false) = 4

include("subdir/subdir2/module.jl")

h() = f(1)
l() = h()

ambig(x, y) = 1
ambig(x::Integer, y) = 2
ambig(x, y::Integer) = 3
ambig(x::Int, y::Int) = 4
ambig(x::Number, y) = 5

subdir/subdir2/module.jl

module Example
    Main.f(x,y) = 2
    g() = :g
end

The one thing that kind of bothers me though is that the module coloring between the MethodError message and the followed stacktrace is not consistent - they're done seperately & in isolation and there's no easy way to "pass on" the existing coloring that's built in MethodError. The colors between method printing, stacktraces and MethodError are now also consistent. There may be a race condition on the underlying IdDict, should two tasks set a color at the same time. Since only one printing call would be affected and it's a non-functional part, I'd say this is fine.

Opinions, improvement ideas etc. are welcome! I've been bothered by how hard MethodErrors are to read for a long time now, and I think unifying the printing with stacktraces is a good first step to improve this.

Closes #40251 Closes #40913 Closes #30110 Closes #36129

[Seelengrab]

I do expect tests to fail, mostly due to doctests. I just wanted to have CI churn away at finding them instead of my laptop 😅 I'll fix them as they occur and after some feedback about the general layout/design.

[rikhuijzer]

In response to your request for comments on Zulip

I like how you spotted and fixed the inconsistency between using @ and at in the stacktrace versus the error message.

[Seelengrab]

I was made aware of #40251 and now I'm thinking about unifying the two, since they're (in part) about the same exact thing: making methods & their uses easier to read when displayed.

[Seelengrab]

Alright, I incorporated #40251 (with permission) into this PR, since that one kinda stalled and this one also touches method printing (though at a different place). I'll update the images at the top post.

I think this should be about it, maybe just some very minor things and this should be good to go!

[Seelengrab]

Ah of course, doctests - those will have to wait for tomorrow. Trying to fix those results in ugly expanded stacktraces to be printed, which I'll have to investigate why that happens.. I didn't touch any logic regarding that, I don't think.

[Seelengrab]

While fixing the failing tests, I noticed that the ambiguity errors have their own code path here - so I'm changing those too.

[Seelengrab]

Alright, the only thing that's missing now should be the doctests (for real this time). I've also added a NEWS.md entry and rebased onto current master.

I'm still not 100% sure how to fix the doctests best, without someone in the future running into having to specifically exclude the non-existence of [...] everywhere 🤔

[Seelengrab]

Alright, I think the doctests are fixed now, so this should be good to go once CI agrees. I took the liberty to squash the commits already, to keep history cleaner.

[Seelengrab]

Since this has been agreed on by triage, can this be merged soon-ish? It'd be unfortunate if this stalls and I have to keep fixing new doctests/other tests..

[oscardssmith]

LGTM

[Seelengrab]

I missed an inconsistency with colors :| Here's what it looks like now:

Should be perfectly consistent now.

EDIT modulo whitespace 😅

[Seelengrab]

Apologies for the force pushes after you added the mergeme label, fixing the colors broke some tests that should be fixed now (hopefully)..

[Seelengrab]

The Apple Darwin failure seems unrelated:

nested task error: rmprocs: pids [5] not terminated after 30 seconds.

as is the PowerPC build failure:

JIT session error: Symbols not found: [ __gnu_f2h_ieee, __gnu_h2f_ieee ]

No idea what's going on with the Profile failure in the "Allow Fail" category. Seems unrelated?

[tecosaur]

I have to say, I'm not a fan of the choice to have variable names be greyed out. To quote past me in #40251

I think argument names deserve more emphasis, in my PR I use the magenta colour for them. for example there's the join method with signature join(strings, delim, last) and I think it's worth applying emphasising argument names for cases such as those.

[Seelengrab]

The goal here was to make MethodError, methods(f) and stacktraces consistent in color - since it didn't seem like there was consensus about adding more colors there (and some said they dislike more color due to it adding more noise), I left the choice for later.

Perhaps this could be done in a future "customize all the colors" REPL theming PR, like suggested in #41435?