Add precompile signatures to Markdown to reduce latency.

[PallHaraldsson]

Fixes #55706 that is seemingly a 4472x regression, not just 16x (was my first guess, based on CondaPkg, also fixes or greatly mitigates JuliaPy/CondaPkg.jl#145), and large part of 3x regression for PythonCall.

[IanButterworth]

Does it need to be wrapped like here:

Yes, in case people are using Markdown without compiled modules.

[IanButterworth]

It's generally less brittle and cross-platform safe to add a precompile workload, as long as its self contained. Seems possible for this?

[PallHaraldsson]

I'll fix that trivial thing. Last time I checked Markdown didn't work, as is on 1.12-DEV so I kind of expect CI errors... But those will be unrelated, so should I make a PR to some (1.11) branch?

[KristofferC]

Just change it to a workload and it can be backported.

[PallHaraldsson]

You mean like Markdown.mdexpr(" ... ") workload? Or even with an empty string? Either way, I'm not sure where, just in same place?! Or rather the macro version: md" ... "? I was timing the former, since the timing for the latter was bogus for some reason... It's functions/methods that get precompiled, but not macros? It's unclear to me if the latter is slower, and macros should sometimes, then, be avoided...

[PallHaraldsson]

Don't try to run this until I (change and) undraft it. It seems to have opened a can of worms, see here: JuliaLang/Pkg.jl#4017

[PallHaraldsson]

I changed and undrafted, but I'm not sure if this is the right place for the workload. Also before rc3 may have been broken, because of my older change, not sure about this one. Unchanged RC2 is at least known ok, maybe the bug I had has always been there in RC3.

The speed regression is from 1.10, but actually also from rc1, even rc2 if I recall.

[PallHaraldsson]

I think this already works, but can be improved. The workload helps to start Markdown, but is seemingly not realistic enough, since this is improved but not as much as I would like, likely because I see Markdown there:

$ julia +1.11 --trace-compile=stderr -e "@time using CondaPkg"
precompile(Tuple{typeof(Base.setindex!), Base.EnvDict, Bool, String})
precompile(Tuple{typeof(Base.Threads.lock_profiling), Bool})
precompile(Tuple{typeof(Base.getindex), Base.Threads.Atomic{Int64}})
precompile(Tuple{typeof(Base.gc_num)})
precompile(Tuple{typeof(Base.cumulative_compile_timing), Bool})
precompile(Tuple{typeof(Base.cumulative_compile_time_ns)})
precompile(Tuple{typeof(micromamba_jll.find_artifact_dir)})
precompile(Tuple{typeof(Base.invokelatest), Any})
precompile(Tuple{typeof(JLLWrappers.get_julia_libpaths)})
precompile(Tuple{typeof(Base.getproperty), Markdown.MD, Symbol})
precompile(Tuple{typeof(Markdown.hashheader), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, Markdown.MD})
precompile(Tuple{Type{NamedTuple{(:eat,), T} where T<:Tuple}, Tuple{Bool}})
precompile(Tuple{typeof(Markdown.list), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, Markdown.MD})
precompile(Tuple{Type{NamedTuple{(:padding,), T} where T<:Tuple}, Tuple{Bool}})
precompile(Tuple{typeof(Base.signbit), Int64})
precompile(Tuple{typeof(Markdown.fencedcode), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, Markdown.MD})
precompile(Tuple{typeof(Markdown.blockquote), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, Markdown.MD})
precompile(Tuple{Type{NamedTuple{(:allow_whitespace, :allowempty, :eat), T} where T<:Tuple}, Tuple{Bool, Bool, Bool}})
precompile(Tuple{typeof(Markdown.admonition), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, Markdown.MD})
precompile(Tuple{Type{NamedTuple{(:rep,), T} where T<:Tuple}, Tuple{Bool}})
precompile(Tuple{typeof(Markdown.blocktex), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, Markdown.MD})
precompile(Tuple{typeof(Markdown.blockinterp), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, Markdown.MD})
precompile(Tuple{typeof(Markdown.indentcode), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, Markdown.MD})
precompile(Tuple{typeof(Markdown.footnote), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, Markdown.MD})
precompile(Tuple{typeof(Markdown.github_table), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, Markdown.MD})
precompile(Tuple{typeof(Markdown.horizontalrule), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, Markdown.MD})
precompile(Tuple{typeof(Markdown.setextheader), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, Markdown.MD})
precompile(Tuple{typeof(Base.:(==)), Char, Char})
precompile(Tuple{Type{NamedTuple{(:breaking,), T} where T<:Tuple}, Tuple{Bool}})
precompile(Tuple{typeof(Markdown.paragraph), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, Markdown.MD})
precompile(Tuple{typeof(Markdown.inline_code), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, Markdown.MD})
precompile(Tuple{typeof(Markdown.en_dash), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, Markdown.MD})
precompile(Tuple{typeof(Base.copy!), Array{Any, 1}, Array{Any, 1}})
precompile(Tuple{typeof(Base.first), Array{Any, 1}})
precompile(Tuple{typeof(Base.Broadcast.broadcasted), typeof(Base.:(-)), Tuple{UInt64, UInt64}, Tuple{UInt64, UInt64}})
precompile(Tuple{typeof(Base.Broadcast.materialize), Base.Broadcast.Broadcasted{Base.Broadcast.Style{Tuple}, Nothing, typeof(Base.:(-)), Tuple{Tuple{UInt64, UInt64}, Tuple{UInt64, UInt64}}}})
precompile(Tuple{Type{Base.GC_Diff}, Base.GC_Num, Base.GC_Num})
precompile(Tuple{typeof(Base.getproperty), Base.GC_Diff, Symbol})
precompile(Tuple{Type{NamedTuple{(:value, :time, :bytes, :gctime, :gcstats, :lock_conflicts, :compile_time, :recompile_time), T} where T<:Tuple}, Tuple{Nothing, Float64, Int64, Float64, Base.GC_Diff, Int64, Float64, Float64}})
precompile(Tuple{typeof(Base.getproperty), NamedTuple{(:value, :time, :bytes, :gctime, :gcstats, :lock_conflicts, :compile_time, :recompile_time), Tuple{Nothing, Float64, Int64, Float64, Base.GC_Diff, Int64, Float64, Float64}}, Symbol})
precompile(Tuple{typeof(Base.:(*)), Float64, Float64})
precompile(Tuple{typeof(Base.gc_alloc_count), Base.GC_Diff})
precompile(Tuple{typeof(Core.kwcall), NamedTuple{(:msg,), Tuple{Nothing}}, typeof(Base.time_print), Base.TTY, Float64, Int64, Int64, Int64, Int64, Float64, Float64, Bool})
precompile(Tuple{Base.var"#1109#1110"{Nothing, Float64, Int64, Int64, Int64, Float64, Float64, Bool, String}, Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}})
precompile(Tuple{typeof(Base.prettyprint_getunits), Int64, Int64, Int64})
precompile(Tuple{typeof(Base.Ryu.writefixed), Float64, Int64})
  3.523722 seconds (1.89 M allocations: 105.742 MiB, 4.43% gc time, 77.01% compilation time)

Pkg also has a 1543x regression:

@time using Pkg
  0.965739 seconds (549.50 k allocations: 34.620 MiB, 9.79% gc time, 0.65% compilation time)

and likely some of these will help (I don't know how I would get the same effect with a "workload"):

precompile(Tuple{typeof(Base.first), Array{Any, 1}})
precompile(Tuple{typeof(Base.setindex!), Base.EnvDict, Bool, String})

maybe even just the first one, or maybe some of these too:

precompile(Tuple{typeof(Base.Threads.atomic_add!), Base.Threads.Atomic{Int64}, Int64})
precompile(Tuple{typeof(Base.print), Base.TTY, String})
precompile(Tuple{typeof(Base.println), Base.TTY})
precompile(Tuple{typeof(Core.kwcall), NamedTuple{(:bold, :italic, :underline, :blink, :reverse, :hidden, :color), Tuple{Bool, Bool, Bool, Bool, Bool, Bool, Symbol}}, typeof(Base.printstyled), Base.TTY, String})
precompile(Tuple{Base.var"##printstyled#1118", Bool, Bool, Bool, Bool, Bool, Bool, Symbol, typeof(Base.printstyled), String, Vararg{String}})
precompile(Tuple{typeof(Core.kwcall), NamedTuple{(:bold, :italic, :underline, :blink, :reverse, :hidden, :color), Tuple{Bool, Bool, Bool, Bool, Bool, Bool, Symbol}}, typeof(Base.printstyled), Base.TTY, String, Vararg{String}})
precompile(Tuple{Base.var"##printstyled#1117", Bool, Bool, Bool, Bool, Bool, Bool, Symbol, typeof(Base.printstyled), Base.TTY, String, Vararg{String}})
precompile(Tuple{typeof(Core.kwcall), NamedTuple{(:bold, :italic, :underline, :blink, :reverse, :hidden), NTuple{6, Bool}}, typeof(Base.with_output_color), Function, Symbol, Base.TTY, String, Vararg{String}})
precompile(Tuple{Base.var"##with_output_color#1116", Bool, Bool, Bool, Bool, Bool, Bool, typeof(Base.with_output_color), Function, Symbol, Base.TTY, String, Vararg{String}})

precompile(Tuple{typeof(Base.first), Array{Any, 1}})

precompile(Tuple{typeof(Base.Threads.atomic_sub!), Base.Threads.Atomic{Int64}, Int64})

Another 46x regression:

@time using NetworkOptions
  0.021625 seconds (6.27 k allocations: 436.031 KiB)

may need some or all of these:

precompile(Tuple{typeof(Base.setindex!), Base.EnvDict, Bool, String})
precompile(Tuple{typeof(Base.Threads.lock_profiling), Bool})
precompile(Tuple{typeof(Base.getindex), Base.Threads.Atomic{Int64}})
precompile(Tuple{typeof(Base.gc_num)})
precompile(Tuple{typeof(Base.cumulative_compile_timing), Bool})
precompile(Tuple{typeof(Base.cumulative_compile_time_ns)})
precompile(Tuple{typeof(Base.Broadcast.broadcasted), typeof(Base.:(-)), Tuple{UInt64, UInt64}, Tuple{UInt64, UInt64}})
precompile(Tuple{typeof(Base.Broadcast.materialize), Base.Broadcast.Broadcasted{Base.Broadcast.Style{Tuple}, Nothing, typeof(Base.:(-)), Tuple{Tuple{UInt64, UInt64}, Tuple{UInt64, UInt64}}}})
precompile(Tuple{Type{Base.GC_Diff}, Base.GC_Num, Base.GC_Num})
precompile(Tuple{typeof(Base.getproperty), Base.GC_Diff, Symbol})
precompile(Tuple{Type{NamedTuple{(:value, :time, :bytes, :gctime, :gcstats, :lock_conflicts, :compile_time, :recompile_time), T} where T<:Tuple}, Tuple{Nothing, Float64, Int64, Float64, Base.GC_Diff, Int64, Float64, Float64}})
precompile(Tuple{typeof(Base.getproperty), NamedTuple{(:value, :time, :bytes, :gctime, :gcstats, :lock_conflicts, :compile_time, :recompile_time), Tuple{Nothing, Float64, Int64, Float64, Base.GC_Diff, Int64, Float64, Float64}}, Symbol})
precompile(Tuple{typeof(Base.:(*)), Float64, Float64})
precompile(Tuple{typeof(Base.gc_alloc_count), Base.GC_Diff})
precompile(Tuple{typeof(Core.kwcall), NamedTuple{(:msg,), Tuple{Nothing}}, typeof(Base.time_print), Base.TTY, Float64, Int64, Int64, Int64, Int64, Float64, Float64, Bool})
precompile(Tuple{Base.var"#1109#1110"{Nothing, Float64, Int64, Int64, Int64, Float64, Float64, Bool, String}, Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}})
precompile(Tuple{typeof(Base.prettyprint_getunits), Int64, Int64, Int64})
precompile(Tuple{typeof(Base.Ryu.writefixed), Float64, Int64})
precompile(Tuple{typeof(Base.print), Base.GenericIOBuffer{GenericMemory{:not_atomic, UInt8, Core.AddrSpace{Core}(0x00)}}, String, String, Vararg{Any}})

julia> @time precompile(Tuple{typeof(Base.in), String, Array{String, 1}})
  0.013533 seconds (331 allocations: 14.688 KiB, 99.87% compilation time)

[KristofferC]

Pkg also has a 1543x regression:

Pkg is not in the sysimage in 1.11. Can we please stay focused on one thing here.

[PallHaraldsson]

This is ready to review/merge, feel free to change the text, add more representative Markdown markup (and drop mentioning Pkg, see Kristoffer's (not outdated) review.

This passed CI with md"" so most likely will with whatever text in there, before my most recent additional (trivial?) commit(s).

[PallHaraldsson]

Pkg is not in the sysimage in 1.11. Can we please stay focused on one thing here.

Ok that part is here: #55721 (comment) though it's also seemingly useful for Markdown as an add-on for the PR, so please merge here first, if not combining merging as one.

[PallHaraldsson]

I can no longer amend by adding one commit, despite restoring the branch I previously deleted. Not sure if that was the problem or if it can't be done after merging. I can also just open another PR for the trivial fix of changing to printing also (do devnull.. ?!), since that was also a regression I overlooked (see the closed issue, that probably should be reopened).