strings/annotated test relies on that StyledStrings have been loaded

[KristofferC]

This is a reduction of a test in test/strings/annotated.jl that avoids loading StyledStrings (which now gets loaded when loading Test.jl):

str1 = Base.AnnotatedString("test", [(1:4, :label => 5)])
str2 = Base.AnnotatedString("case", [(2:3, :label => "oomph")])
@assert join([str1, str1], ' ') ==
    Base.AnnotatedString("test test",
                 [(1:4, :label => 5),
                  (6:9, :label => 5)])

Running this fails with:

ERROR: LoadError: AssertionError: join([str1, str1], ' ') == Base.AnnotatedString("test test", [(1:4, :label => 5), (6:9, :label => 5)])

However, if we add

using StyledStrings

to the top, the assert does not fire.

The reason for this is type piracy in StyledStrings: JuliaLang/StyledStrings.jl#61.

It is weird that things like join on base types give different results depending on if an stdlib has been (transitively) loaded.
The rendering of StyledStrings should probably move to Base to fix this.

[tecosaur]

@KristofferC to be clearer about what the plan/hope is here, are you thinking we ship 1.11 as-is, then move the code around in 1.12?

[KristofferC]

I'm coming around thinking that maybe pulling the plug completely on the StyledString stuff in stdlibs is the best way forward. Huge issues when it comes to the design (split between stdlibs and Base) and latency regressions that have been left unchecked for a year. Enough is starting to be enough.

[tecosaur]

(oops, misclick).

The split is a pain, and makes trying to solve the invalidations a bit of a challenge. I'd very much like to solve this, but I have yet to come up with a clean idea for doing so with the split as it currently is.

With that said, aside from pulling the plug I would very much like to explore whether there's any scope to adjust the base/stdlib split in such a way that we maintain the current functionality but resolve the invalidation issues (with more or less in Base), and/or see if some of the idea's I've been chatting with Cody on-and-off have the potential to make a meaningful dent in invalidations/latency.