Fine-tune stacktrace printing

[mcabbott]

This is a small tweak to #36134, as discussed there. This gist has the same changes.

[mcabbott]

This seems a bit crude, but I got lost trying to find the right printing function to overload for this purpose (of printing only the type parameters dimly, not the outermost type). It should still use aliases like Matrix{Int}.

It's also possible this should be a method of print_within_stacktrace instead of its own function.

[mcabbott]

A question raised in this Discourse thread is whether the stack trace should also truncate very long types. The example there has lines over 7000 characters long. It would be trivial to make print_type_stacktrace truncate this, but are there be cases where the whole type is needed? Maybe it could do this only for interactive use?

(cc @tomerarnon)

[tomerarnon]

whether the stack trace should also truncate very long types.

While I'm personally of the opinion that this should definitely happen, it wasn't clear to me that the discussion had settled on a precise course of action (and while there did seem to be some support for the idea, it was less than I expected, judging based on small numbers of likes and participants).

We brought up a handful of behaviors, (i.e. opt-in/out, and/or toggleable through an ENV variable, and/or a new overloadable _showtype(T) for stacktrace purposes) and I'm not sure which would be best. If you want to run with any of those ideas, I'm more than happy to lend a hand, either here or in a separate pr.

[mcabbott]

Latest commit restores bold function names. I don't see precisely how these were disabled in #36134, so they may have been put back differently, torture tests welcomed. (I have updated the gist linked above, too.)

1.5.2, master, and this PR (with Fira Code, whose bold is pretty strong):

(Merge conflict is with #37750, this line: https://github.com/JuliaLang/julia/pull/37750/files#diff-c616894fef3ba781d1e1cf77a08e69f366ffd9e5b1cfe1220e7bdae85377e0b2R2067)

[KristofferC]

Could you perhaps rebase this on master to get a fresh CI run and get rid of the merge conflict?

[KristofferC]

I'd like to (again) bring up that the red color for the Base module is non-ideal. I tried this out and got:

and immediately my brain went, "Oh what's that error message, Base.Math? That's weird". If anything, it makes those modules stand out more than other colors. Perhaps just make them gray instead? For inlined methods, there are already no module printed so it isn't that much of a missing tooth perhaps?

[mcabbott]

Thanks for the screenshot, maybe they are too close. (Notice also :bright_red on master, at [14] in screenshot just above.) And maybe grey looks less odd with the bold to keep the rhythm. Like this, or :light_black to match the paths?

 @eval Base STACKTRACE_FIXEDCOLORS = IdDict(Base => :normal, Core => :normal)

Will rebase once I've figured out what #37750 did & how not to break it.

Here's the current look, VScode's default dark theme, JuliaMono, with Base now in :light_black:

[jkrumbiegel]

Cool, I like it much better with bold. Especially for light themes, as black is not that much more contrasting with white compared to dark gray.

[KristofferC]

Let's try this out then :)

[mcabbott]

Some constructors don't get printed in bold, but perhaps they should:

julia> PermutedDimsArray{Float64,2,(2,1),(1,2),Array{Float64,2}}(ones(2,2))
ERROR: ArgumentError: (2, 1) and (1, 2) must be inverses
Stacktrace:
 [1] PermutedDimsArray{Float64, 2, (2, 1), (1, 2), Matrix{Float64}}(data::Matrix{Float64})  # <-- not in bold
   @ Base.PermutedDimsArrays ./permuteddimsarray.jl:15
 [2] top-level scope
   @ REPL[54]:1

julia> Matrix{Int}([1 2; 3 4.5])
ERROR: InexactError: Int64(4.5)
Stacktrace:
  [1] Int64
    @ ./float.jl:607 [inlined]
...
 [10] Matrix{Int64}(x::Matrix{Float64})   # <-- not in bold
    @ Base ./array.jl:562
 [11] top-level scope
    @ REPL[57]:1