use the new SnoopPrecompile

[SimonDanisch]

Since it's not registered yet and actually a pretty small macro, I've just inlined it into Makie for now...

[MakieBot]

Compile Times benchmark

Note, that these numbers may fluctuate on the CI servers, so take them with a grain of salt. All benchmark results are based on the mean time and negative percent mean faster than the base branch. Note, that GLMakie + WGLMakie run on an emulated GPU, so the runtime benchmark is much slower. Results are from running:

using_time = @ctime using Backend
# Compile time
create_time = @ctime fig = scatter(1:4; color=1:4, colormap=:turbo, markersize=20, visible=true)
display_time = @ctime Makie.colorbuffer(display(fig))
# Runtime
create_time = @benchmark fig = scatter(1:4; color=1:4, colormap=:turbo, markersize=20, visible=true)
display_time = @benchmark Makie.colorbuffer(display(fig))

	using	create	display	create	display
GLMakie	14.42s (13.72, 15.43) 0.76+-	18.25s (17.44, 19.52) 0.93+-	16.22s (15.14, 17.32) 0.77+-	17.96ms (17.39, 19.18) 0.79+-	78.43ms (76.28, 80.90) 1.62+-
master	11.83s (11.21, 12.72) 0.70+-	22.54s (20.99, 23.90) 1.38+-	18.45s (17.46, 19.47) 0.96+-	24.55ms (23.18, 25.78) 1.20+-	88.02ms (82.95, 92.74) 3.02+-
evaluation	+17.94%, 2.59s slower❌ (3.53d, 0.00p, 0.73std)	-23.54%, -4.3s faster✅ (-3.65d, 0.00p, 1.15std)	-13.74%, -2.23s faster✅ (-2.56d, 0.00p, 0.87std)	-36.71%, -6.59ms faster✅ (-6.50d, 0.00p, 0.99std)	-12.23%, -9.59ms faster✅ (-3.96d, 0.00p, 2.32std)
CairoMakie	17.49s (17.38, 17.57) 0.07+-	28.18s (27.87, 28.41) 0.18+-	3.52s (3.46, 3.57) 0.03+-	36.25ms (35.74, 36.61) 0.30+-	32.67ms (32.27, 33.31) 0.37+-
master	13.40s (13.25, 13.73) 0.16+-	33.49s (33.21, 33.61) 0.14+-	4.81s (4.70, 4.92) 0.09+-	30.77ms (30.27, 31.10) 0.34+-	31.38ms (31.00, 31.92) 0.31+-
evaluation	+23.34%, 4.08s slower❌ (34.20d, 0.00p, 0.11std)	-18.84%, -5.31s faster✅ (-32.89d, 0.00p, 0.16std)	-36.38%, -1.28s faster✅ (-19.65d, 0.00p, 0.06std)	+15.12%, 5.48ms slower❌ (17.12d, 0.00p, 0.32std)	+3.95%, 1.29ms slower X (3.74d, 0.00p, 0.34std)
WGLMakie	16.98s (16.81, 17.12) 0.11+-	32.39s (32.01, 32.67) 0.22+-	45.48s (44.64, 47.36) 0.89+-	24.89ms (24.38, 25.25) 0.31+-	1.72s (1.68, 1.78) 0.03+-
master	13.62s (13.52, 13.70) 0.06+-	33.85s (33.46, 34.44) 0.30+-	54.41s (53.55, 55.31) 0.64+-	23.57ms (22.88, 24.16) 0.41+-	1.70s (1.65, 1.74) 0.03+-
evaluation	+19.77%, 3.36s slower❌ (37.65d, 0.00p, 0.09std)	-4.51%, -1.46s faster ✓ (-5.54d, 0.00p, 0.26std)	-19.63%, -8.93s faster✅ (-11.53d, 0.00p, 0.77std)	+5.27%, 1.31ms slower❌ (3.63d, 0.00p, 0.36std)	+1.15%, 0.02s invariant (0.61d, 0.28p, 0.03std)

[SimonDanisch]

The using times slow down are expected, since this now caches precompiles for much more than just scatter....
The speed up and one slow down of run time must be noise, although the GLMakie one seems to be real, since it has been showing up pretty consistently.
Must be the change in screen.jl (maybe closeall(screen_ref)?).
Pretty happy about the improvement in general (thanks @timholy, this macro is really amazing and has been on my bucket list for quite some time :) ).