Reduce invalidations by defining isless and ==

[jaakkor2]

using SnoopCompileCore
invs = @snoop_invalidations using ProgressLogging
using SnoopCompile
length(uinvalidated(invs))

improves

• 1.10.10 (lts) 312->136
• 1.11.6 (release) 112->107
• 1.12.0-rc2 276->147

Ref: JuliaLang/julia#59504

[jaakkor2]

This PR improves TTFP of GLMakie on Julia 1.12.0-rc2, when ProgressLogging has been loaded

Julia v1.10.10 (LTS)

For reference

> julia +lts --startup-file=no --banner=no
julia> using ProgressLogging; @time using GLMakie; @time display(plot(rand(10)))
  3.676409 seconds (4.05 M allocations: 267.888 MiB, 5.89% gc time, 2.88% compilation time: 30% of which was recompilation)
  0.507509 seconds (128.96 k allocations: 13.300 MiB, 41.98% compilation time)

Julia v1.12.0-rc2

Without ProgressLogging

> julia +rc --startup-file=no --banner=no
julia> @time using GLMakie; @time display(plot(rand(10)))
  3.207026 seconds (5.28 M allocations: 302.216 MiB, 8.76% gc time, 16.19% compilation time: 91% of which was recompilation)
  0.530239 seconds (270.22 k allocations: 17.278 MiB, 42.26% compilation time: <1% of which was recompilation)
GLMakie.Screen(...)

With ProgressLogging v0.1.5

> julia +rc --startup-file=no --banner=no
julia> using ProgressLogging; @time using GLMakie; @time display(plot(rand(10)))
  3.180291 seconds (5.25 M allocations: 298.694 MiB, 9.25% gc time, 15.66% compilation time: 90% of which was recompilation)
  6.690593 seconds (30.71 M allocations: 1.538 GiB, 7.30% gc time, 95.38% compilation time: 83% of which was recompilation)

This PR

> julia +rc --startup-file=no --banner=no
julia> using ProgressLogging; @time using GLMakie; @time display(plot(rand(10)))
  3.141481 seconds (5.27 M allocations: 301.756 MiB, 8.74% gc time, 15.76% compilation time: 90% of which was recompilation)
  1.750266 seconds (4.12 M allocations: 216.110 MiB, 2.43% gc time, 82.57% compilation time: 67% of which was recompilation)

In summary, with this PR, TTFP goes down from 6.7 s to 1.8 s, clear improvement, but not yet the level of 0.5 s with LTS.

[jaakkor2]

Removing the line

Base.convert(::Type{T}, str::ProgressString) where {T<:AbstractString} =
    convert(T, str.progress.name)

gets rid of the remaining invalidations. Tests of ProgressLogging.jl and TerminalLoggers.jl pass locally.

> julia +rc --startup-file=no --banner=no # Julia 1.12.0-rc2
julia> using ProgressLogging; @time using GLMakie; @time display(plot(rand(10)))
  3.253941 seconds (5.18 M allocations: 301.034 MiB, 8.93% gc time, 15.09% compilation time: 91% of which was recompilation)
  0.497792 seconds (258.49 k allocations: 16.721 MiB, 41.59% compilation time: <1% of which was recompilation)
GLMakie.Screen(...)

> julia +pr59648 --startup-file=no --banner=no # Development version 1.12.0-rc3.0
julia> using ProgressLogging; @time using GLMakie; @time display(plot(rand(10)))
  3.198127 seconds (5.18 M allocations: 300.921 MiB, 8.96% gc time, 15.47% compilation time: 89% of which was recompilation)
  0.509152 seconds (259.15 k allocations: 16.688 MiB, 42.73% compilation time: <1% of which was recompilation)
GLMakie.Screen(...)

[pfitzseb]

Can you check whether progress logging still works in VS Code with this patch please? I don't think we rely on ProgressString there, but would be better to be sure.

[jaakkor2]

Can you check whether progress logging still works in VS Code with this patch please? I don't think we rely on ProgressString there, but would be better to be sure.

using ProgressLogging
@progress for i in 1:10
    sleep(1)
end

shows at the bottom of the VS Code window

[jaakkor2]

Works in Pluto.jl