allow time_imports to return julia data that can be sorted and examined more easily

[anandijain]

@time_imports is really great, but I was a bit surprised that you can't do things like sum the list of times to get the total time, ie #46850, nor sort them to see the longest packages to import.

I briefly looked at the code for @time_imports but it got to ccalls and didn't see an easy way to expose the data generated.

doing this would allow us to track using times for big ecosystem packages across julia versions and package versions.
is there a way to do this that I don't know about?

my temporary solution is to run julia --project --startup=no -E 'using InteractiveUtils; @time_imports using DifferentialEquations' and then just parse it sloppily

[anandijain]

# julia --project --startup=no -E 'using InteractiveUtils; @time_imports using DifferentialEquations' > timings.txt
using CSV, DataFrames
fn = "timings.txt"
ls = strip.(readlines(fn))
cols = split.(ls, "  ")
last.(cols)
df = DataFrame(time=Float64[], unit=String[], pkg=String[], comp=Union{Missing, String}[])
for (i, col) in enumerate(cols)
    time = first(col)
    time, unit = split(time, " ")
    time = parse(Float64, time)
    pkg_and_comp = last(col)
    foo = split(pkg_and_comp, " ";limit=2)
    length(foo) == 1 ? (pkg, comp) = (foo[1], missing) : (pkg, comp) = foo
    row = vec([time unit pkg comp])
    push!(df, row)
end

sort!(df, :time; rev=true)
sum(df.time)

This is somewhat useful

[jmanthony3]

# julia --project --startup=no -E 'using InteractiveUtils; @time_imports using DifferentialEquations' > timings.txt
using CSV, DataFrames
fn = "timings.txt"
ls = strip.(readlines(fn))
cols = split.(ls, "  ")
last.(cols)
df = DataFrame(time=Float64[], unit=String[], pkg=String[], comp=Union{Missing, String}[])
for (i, col) in enumerate(cols)
    time = first(col)
    time, unit = split(time, " ")
    time = parse(Float64, time)
    pkg_and_comp = last(col)
    foo = split(pkg_and_comp, " ";limit=2)
    length(foo) == 1 ? (pkg, comp) = (foo[1], missing) : (pkg, comp) = foo
    row = vec([time unit pkg comp])
    push!(df, row)
end

sort!(df, :time; rev=true)
sum(df.time)

This is somewhat useful

Thanks for this little scriplet! It has been rather handy. In case it were relevant, I added some capability with Unitful.jl to get the returned load times into the same set of units (preferrably seconds).

function timings_results(cols)
    load_time, df = 0. * u"s", DataFrame(time=Float64[], unit=String[], pkg=String[], comp=Union{Missing, String}[])
    for (i, col) ∈ enumerate(cols)
        time, unit      = split(first(col), " ")
        time            = parse(Float64, time)
        pkg_and_comp    = last(col)
        foo = split(pkg_and_comp, " "; limit=2)
        length(foo) == 1 ? (pkg, comp) = (foo[1], missing) : (pkg, comp) = foo
        row = vec([time unit pkg comp])
        push!(df, row)
        load_time += upreferred(time*uparse(unit))
    end
    @printf("Total load time: %.3f s", uconvert(NoUnits, load_time * u"1/s"))
    return nothing
end