Type piracy in package

[KristofferC]

All of this stuff:

StyledStrings.jl/src/io.jl

Lines 252 to 297 in 6901610

	write(io::IO, s::Union{<:AnnotatedString, SubString{<:AnnotatedString}}) =
	_ansi_writer(io, s, write)::Int
	print(io::IO, s::Union{<:AnnotatedString, SubString{<:AnnotatedString}}) =
	(write(io, s); nothing)
	escape_string(io::IO, s::Union{<:AnnotatedString, SubString{<:AnnotatedString}},
	esc = ""; keep = ()) =
	(_ansi_writer(io, s, (io, s) -> escape_string(io, s, esc; keep)); nothing)
	function write(io::IO, c::AnnotatedChar)
	if get(io, :color, false) == true
	termstyle(io, getface(c), getface())
	bytes = write(io, c.char)
	termstyle(io, getface(), getface(c))
	bytes
	else
	write(io, c.char)
	end
	end
	print(io::IO, c::AnnotatedChar) = (write(io, c); nothing)
	function show(io::IO, c::AnnotatedChar)
	if get(io, :color, false) == true
	out = IOBuffer()
	show(out, c.char)
	print(io, ''', AnnotatedString(String(take!(out)[2:end-1]), map(a -> (1:ncodeunits(c), a), c.annotations)), ''')
	else
	show(io, c.char)
	end
	end
	function write(io::IO, aio::Base.AnnotatedIOBuffer)
	if get(io, :color, false) == true
	# This does introduce an overhead that technically
	# could be avoided, but I'm not sure that it's currently
	# worth the effort to implement an efficient version of
	# writing from a AnnotatedIOBuffer with style.
	# In the meantime, by converting to an `AnnotatedString` we can just
	# reuse all the work done to make that work.
	write(io, read(aio, AnnotatedString))
	else
	write(io, aio.io)
	end
	end

looks to be a pretty bad type piracy and should probably be moved to Base.

[kimikage]

Somewhat related to #57, regarding the justification for this library to do the piracy.

[tecosaur]

Loosely at best, since here we're talking about moving some functionality from this library to Base, and #57 is more about what/how functionality should be backported to the compat version.

[kimikage]

That's right. However, isn't the "result" of that the same?
That is, when they are transferred to Base, julia1-compat pirates them anew.

[tecosaur]

I can't say I follow. In Julia >1.11 the julia1-compat version shouldn't be used.

[kimikage]

I am just referring to julia1-compat, i.e., in Julia <1.11.

The above piracy could be eliminated from StyledString as stdlib. As a result, the piracy left in julia1-compat will go from "mere backporting" to "pure piracy".

In other words, I am talking about the justification for piracy here, not how the code behaves.

[ronisbr]

Hi!

Just for the records, it seems that this issue is causing a huge precompilation overhead in PrettyTables.jl when using Julia 1.11. For more information, see JuliaLang/julia#56080.

[tecosaur]

Good to have, and FWIW you have some sympathy for the effort in tracking this down.

I don't think the core issue has actually been stated here, so it would probably be worth me doing so for the record.

During the series of triage discussions that occurred around StyledStrings, it was decided that an initial proposal to add the functionality to Base would be split into a Base part and a new stdlib part (StyledStrings). This was done to prioritise keeping Base slim, but had the side-effect of creating the invalidations giving us headaches now.

There's ongoing discussion about how this should best be resolved, ranging from just evicting everything currently in Base (and the convenience that comes with it), to defining a "stub" print method in Base to control invalidations that's (somehow) substituted for the "fully-fledged" print defined in this library, or even brining all the code required for printing an AnnotatedString with color into Base.

Cody also had an interesting idea I half-remember about making a more central, pluggable, print method to solve this issue and make it possible to customise printing potentially.

This is an issue that's cared about, without an obvious fix, but with a few ideas that need exploring.

[topolarity]

Cody also had an interesting idea I half-remember about making a more central, pluggable, print method to solve this issue and make it possible to customise printing potentially.

Yeah the idea is to split the behavior into the AnnotatedString vs. Face portions, so that the AnnotatedString portion (which iterates annotations) can live in Base and the Face portion (which updates the actual printer state for display, resolves colors, etc.) can be separate

The additional wrinkle right now is that an AnnotatedString is just a set of Symbols mapped to regions, so we don't know which copy of StyledStrings in memory to ask to generate/display Faces for them (see #91). We might need to associate an AnnotatedString with some kind of library-specific type that allows us to dispatch the resolution / display of its annotations to the correct library.

You wouldn't be allowed to "mix" annotations from different libraries in the same string - but that also dramatically simplifies the display logic since you only need to worry about "your own" display state on the library side. The key thing to get rid of is the hard-coded behavior where we assume there is one special library in memory that knows how to display annotations in an AnnotatedString (and we invalidate on it)

[KristofferC]

It seems to invalidate the package precompilation functionality:

❯ julia +nightly --project -e '@time @eval Base.Precompilation.precompilepkgs(); 
                               using StyledStrings; 
                               @time @eval Base.Precompilation.precompilepkgs()'
  0.082379 seconds (84.51 k allocations: 10.849 MiB, 51.15% compilation time)
  1.092273 seconds (1.46 M allocations: 83.460 MiB, 4.71% gc time, 96.65% compilation time: 100% of which was recompilation)

Causes latency problems in e.g Pkg and package loading when the REPL (which depends on StyledStrings) is loaded.