timed precompilation of package and dependents

[anandijain]

https://julialang.slack.com/archives/C6A044SQH/p1674223081428299

I want to build something that allows me to plot how long precompilation takes for a package and all of it's dependencies across package and julia versions.

the end goal being something like https://perf.rust-lang.org/

the usage I had in mind was

using Pkg

timings = Pkg.precompile_timed("DifferentialEquations";recompile=true) # recompile to delete caches
timings # Vector{Pair{PackageSpec, Float64}}`

i'd love to take a swing at this if someone wants to give me some pointers on how to implement it

[timholy]

I want to build something that allows me to plot how long precompilation takes for a package and all of it's dependencies across package and julia versions.

for juliabin in ("julia-1.6", "julia-1.7", "julia-1.8", "julia-1.9")   # customize these to the versions you have and how you launch them
    rm(expanduser("~/.julia/compiled/"); recursive=true)   # clear all precompiled files
    run(`$juliabin -e 'using Pkg; @time Pkg.precompile("ThePkgIWantToPrecompile")'`)
end

and do something to harvest the output of @time. You could, e.g., use @elapsed instead and write the returned time to a file.

(Script untested, but I'll leave that to you 😉 ).

[anandijain]

I think @IanButterworth was thinking of adding something to Pkg that gives not just a single package, but all dependents too.

Current progress is at the https://github.com/anandijain/JuliaBenchmarker.jl

[anandijain]

Can we reopen? That script is helpful but doesn't give the breakdown that I'm looking for. It would be much more valuable to have the functionality akin to @time_imports that actually describes the time each took.

I realize I could just do this for each node in the dependency graph, but that doesn't seem like the right solution.