Add some precompiles to help loading time

[IanButterworth]

A tiny PR that just shaves about 10% off Plots load time (1.1s -> 0.95s) by adding some precompiles that JLL init's hit.

I thought I'd take the opportunity to benchmark, on a M2 Pro macbook:

1.10.9

julia> @time Pkg.precompile("Plots")
Precompiling Plots finished.
  146 dependencies successfully precompiled in 45 seconds. 6 already precompiled.
 45.247241 seconds (2.59 M allocations: 196.504 MiB, 0.06% gc time, 1.81% compilation time)

julia> @time using Plots
  0.999457 seconds (1.01 M allocations: 67.811 MiB, 1.81% compilation time)

julia> @time display(plot(rand(3), rand(3)))
  0.667132 seconds (96.95 k allocations: 6.709 MiB, 20.10% compilation time: 5% of which was recompilation)

1.11.5

julia> @time Pkg.precompile("Plots")
Precompiling Plots...
  147 dependencies successfully precompiled in 53 seconds. 35 already precompiled.
 52.805236 seconds (2.44 M allocations: 167.030 MiB, 0.07% gc time, 3 lock conflicts, 0.85% compilation time: 56% of which was recompilation)

julia> @time using Plots
  1.170431 seconds (1.22 M allocations: 79.566 MiB, 3.49% gc time, 2.23% compilation time: 43% of which was recompilation)

julia> @time display(plot(rand(3), rand(3)))
  0.384659 seconds (204.01 k allocations: 10.608 MiB, 42.04% compilation time: 13% of which was recompilation)

1.12.0-beta3

julia> @time Pkg.precompile("Plots")
Precompiling Plots finished.
  147 dependencies successfully precompiled in 56 seconds. 36 already precompiled.
 55.739992 seconds (2.01 M allocations: 160.495 MiB, 0.26% gc time, 5 lock conflicts, 0.39% compilation time: 66% of which was recompilation)

julia> @time using Plots
  1.246677 seconds (1.88 M allocations: 105.135 MiB, 3.32% gc time, 4.70% compilation time)

julia> @time display(plot(rand(3), rand(3)))
  0.376974 seconds (208.82 k allocations: 10.938 MiB, 38.16% compilation time: 18% of which was recompilation)

Master 43ead47

julia> @time Pkg.precompile("Plots")
Precompiling Plots finished.
  147 dependencies successfully precompiled in 59 seconds. 35 already precompiled.
 59.410907 seconds (1.92 M allocations: 155.733 MiB, 0.24% gc time, 5 lock conflicts, 0.40% compilation time: 49% of which was recompilation)

julia> @time using Plots
  1.108766 seconds (1.54 M allocations: 87.170 MiB, 11.34% gc time, 6.27% compilation time)

julia> @time display(plot(rand(3), rand(3)))
  0.403667 seconds (202.36 k allocations: 10.240 MiB, 44.11% compilation time: 16% of which was recompilation)

PR

julia> @time Pkg.precompile("Plots")
Precompiling Plots finished.
  147 dependencies successfully precompiled in 58 seconds. 35 already precompiled.
 58.074903 seconds (1.91 M allocations: 155.245 MiB, 0.21% gc time, 3 lock conflicts, 0.33% compilation time: 61% of which was recompilation)

julia> @time using Plots
  0.951054 seconds (1.23 M allocations: 71.972 MiB, 4.11% gc time)

julia> @time display(plot(rand(3), rand(3)))
  0.404166 seconds (202.13 k allocations: 10.213 MiB, 43.13% compilation time: 16% of which was recompilation)

And with this PR almost all loading compilation is gone

julia> @trace_compile @time_imports using Plots
#=    4.6 ms =# precompile(Tuple{typeof(Base.vect), Array{String, 1}, Vararg{Array{String, 1}}})
#=    3.3 ms =# precompile(Tuple{typeof(Base.iterate), Array{Array{String, 1}, 1}})
      1.4 ms  Statistics
               ┌ 93.5 ms SuiteSparse_jll.__init__()
     94.6 ms  SuiteSparse_jll
      1.6 ms  Serialization
               ┌ 1.2 ms SparseArrays.CHOLMOD.__init__()
    123.1 ms  SparseArrays
      1.8 ms  #=    7.5 ms =# precompile(Tuple{typeof(Base.Filesystem.joinpath), NTuple{7, String}})
Statistics → SparseArraysExt
      1.5 ms  Preferences
      1.0 ms  PrecompileTools
      1.0 ms  Reexport
      1.3 ms  Scratch
      1.4 ms  RelocatableFolders
      3.8 ms  RecipesBase
     14.9 ms  FixedPointNumbers
               ┌ 0.0 ms ColorTypes.__init__()
     15.7 ms  ColorTypes
     24.3 ms  Colors
      2.2 ms  TensorCore
               ┌ 0.0 ms ColorVectorSpace.__init__()
     22.0 ms  ColorVectorSpace
      4.4 ms  ColorSchemes
      1.7 ms  StableRNGs
     39.1 ms  PlotUtils
      5.2 ms  PlotThemes
               ┌ 2.9 ms OpenLibm_jll.__init__()
      4.3 ms  OpenLibm_jll
      1.4 ms  NaNMath
     10.7 ms  RecipesPipeline
               ┌ 0.0 ms Requires.__init__()
      1.6 ms  Requires
      2.0 ms  UnicodeFun
      1.2 ms  ColorTypes → StyledStringsExt
      1.6 ms  DataAPI
      1.3 ms  Compat
      1.2 ms  Compat → CompatLinearAlgebraExt
      4.7 ms  OrderedCollections
     26.2 ms  DataStructures
      1.7 ms  SortingAlgorithms
      4.1 ms  Missings
               ┌ 0.0 ms DocStringExtensions.__init__()
      1.9 ms  DocStringExtensions
      6.5 ms  IrrationalConstants
      1.6 ms  LogExpFunctions
      1.6 ms  StatsAPI
      1.5 ms  PtrArrays
      2.2 ms  AliasTables
     10.2 ms  StatsBase
      1.5 ms  Showoff
      3.1 ms  Unzip
      1.5 ms  JLLWrappers
#=    0.0 ms =# precompile(Tuple{typeof(JLLWrappers.get_julia_libpaths)})
               ┌ 0.1 ms fzf_jll.__init__()
      1.5 ms  fzf_jll
      1.4 ms  JLFzf
      1.8 ms  Mmap
     11.9 ms  Parsers
      3.5 ms  JSON
      3.5 ms  Measures
               ┌ 2.2 ms Bzip2_jll.__init__()
      3.8 ms  Bzip2_jll
#=    0.0 ms =# precompile(Tuple{typeof(Bzip2_jll.eager_mode)})
               ┌ 2.2 ms FreeType2_jll.__init__()
      3.9 ms  FreeType2_jll
               ┌ 2.1 ms FriBidi_jll.__init__()
      3.7 ms  FriBidi_jll
               ┌ 4.1 ms Libiconv_jll.__init__()
      5.8 ms  Libiconv_jll
               ┌ 2.1 ms Libffi_jll.__init__()
      3.8 ms  Libffi_jll
#=    0.0 ms =# precompile(Tuple{typeof(Libiconv_jll.eager_mode)})
               ┌ 2.9 ms XML2_jll.__init__()
      4.7 ms  XML2_jll
#=    0.0 ms =# precompile(Tuple{typeof(XML2_jll.eager_mode)})
               ┌ 3.4 ms Gettext_jll.__init__()
      5.2 ms  Gettext_jll
               ┌ 4.4 ms PCRE2_jll.__init__()
      6.1 ms  PCRE2_jll
#=    0.0 ms =# precompile(Tuple{typeof(Libffi_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(Gettext_jll.eager_mode)})
               ┌ 12.5 ms Glib_jll.__init__()
     14.3 ms  Glib_jll
               ┌ 3.8 ms LLVMOpenMP_jll.__init__()
      5.7 ms  LLVMOpenMP_jll
#=    0.0 ms =# precompile(Tuple{typeof(LLVMOpenMP_jll.eager_mode)})
               ┌ 2.8 ms Pixman_jll.__init__()
      4.7 ms  Pixman_jll
               ┌ 2.5 ms libpng_jll.__init__()
      4.3 ms  libpng_jll
      1.8 ms  Libuuid_jll
               ┌ 2.4 ms Expat_jll.__init__()
      4.4 ms  Expat_jll
#=    0.0 ms =# precompile(Tuple{typeof(FreeType2_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(Expat_jll.eager_mode)})
               ┌ 2.4 ms Fontconfig_jll.__init__()
      4.5 ms  Fontconfig_jll
#=    0.0 ms =# precompile(Tuple{typeof(Glib_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(Pixman_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(libpng_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(Fontconfig_jll.eager_mode)})
               ┌ 7.5 ms Cairo_jll.__init__()
      9.5 ms  Cairo_jll
               ┌ 2.7 ms Graphite2_jll.__init__()
      4.7 ms  Graphite2_jll
#=    0.0 ms =# precompile(Tuple{typeof(Cairo_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(Graphite2_jll.eager_mode)})
               ┌ 7.8 ms HarfBuzz_jll.__init__()
      9.9 ms  HarfBuzz_jll
#=    0.0 ms =# precompile(Tuple{typeof(FriBidi_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(HarfBuzz_jll.eager_mode)})
               ┌ 2.9 ms libass_jll.__init__()
      5.0 ms  libass_jll
               ┌ 2.7 ms libfdk_aac_jll.__init__()
      4.8 ms  libfdk_aac_jll
               ┌ 2.9 ms LAME_jll.__init__()
      4.9 ms  LAME_jll
               ┌ 2.2 ms Ogg_jll.__init__()
      4.4 ms  Ogg_jll
#=    0.0 ms =# precompile(Tuple{typeof(Ogg_jll.eager_mode)})
               ┌ 8.0 ms libvorbis_jll.__init__()
     10.3 ms  libvorbis_jll
               ┌ 3.0 ms libaom_jll.__init__()
      5.3 ms  libaom_jll
               ┌ 2.9 ms x264_jll.__init__()
      5.1 ms  x264_jll
               ┌ 3.7 ms x265_jll.__init__()
      5.9 ms  x265_jll
               ┌ 2.9 ms Opus_jll.__init__()
      5.3 ms  Opus_jll
#=    0.0 ms =# precompile(Tuple{typeof(libass_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(libfdk_aac_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(LAME_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(libvorbis_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(libaom_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(x264_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(x265_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(Opus_jll.eager_mode)})
               ┌ 17.1 ms FFMPEG_jll.__init__()
     19.8 ms  FFMPEG_jll
      2.3 ms  FFMPEG
               ┌ 3.0 ms GLFW_jll.__init__()
      5.5 ms  GLFW_jll
               ┌ 6.2 ms JpegTurbo_jll.__init__()
      8.9 ms  JpegTurbo_jll
               ┌ 3.2 ms LERC_jll.__init__()
      5.8 ms  LERC_jll
               ┌ 2.9 ms XZ_jll.__init__()
      5.6 ms  XZ_jll
               ┌ 3.1 ms Zstd_jll.__init__()
      5.6 ms  Zstd_jll
#=    0.0 ms =# precompile(Tuple{typeof(JpegTurbo_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(LERC_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(XZ_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(Zstd_jll.eager_mode)})
               ┌ 3.2 ms Libtiff_jll.__init__()
      5.7 ms  Libtiff_jll
      2.5 ms  libinput_jll
      2.5 ms  Xorg_libXext_jll
      2.7 ms  Xorg_libxcb_jll
      2.5 ms  Xorg_xcb_util_wm_jll
      2.5 ms  Xorg_xcb_util_cursor_jll
      2.6 ms  Xorg_xcb_util_image_jll
      4.2 ms  Xorg_xcb_util_keysyms_jll
      2.7 ms  Xorg_xcb_util_renderutil_jll
      2.6 ms  Xorg_libXrender_jll
      2.7 ms  Xorg_libSM_jll
      2.6 ms  xkbcommon_jll
      2.7 ms  Libglvnd_jll
               ┌ 3.6 ms Vulkan_Loader_jll.__init__()
      6.4 ms  Vulkan_Loader_jll
#=    0.0 ms =# precompile(Tuple{typeof(Vulkan_Loader_jll.eager_mode)})
               ┌ 58.1 ms Qt6Base_jll.__init__()
     61.1 ms  Qt6Base_jll
#=    0.0 ms =# precompile(Tuple{typeof(FFMPEG_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(GLFW_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(Libtiff_jll.eager_mode)})
#=    0.0 ms =# precompile(Tuple{typeof(Qt6Base_jll.eager_mode)})
               ┌ 0.1 ms GR_jll.__init__()
      3.4 ms  GR_jll
               ┌ 0.1 ms GR.GRPreferences.__init__()
      9.3 ms  GR
               ┌ 0.5 ms Plots.__init__()
    339.5 ms  Plots

Notes

Linux is a bit faster than macOS on this M2, and I think it's #58409, which #58405 works to avoid. i.e. If I remove the dlpath calls in SuiteSparse_jll alone I get

julia> @time using Plots
  0.866295 seconds (1.22 M allocations: 71.202 MiB, 4.40% gc time)

Also, I think we should change these zero trace-compiles to indicate more clearly that they're being inferred, not compiled, AFAIU. i.e. #= 0.0 ms =# precompile(Tuple{typeof(FFMPEG_jll.eager_mode)})