Precompilation setup?

[juliohm]

I am experimenting with the package for the first time, and noticed that the precompilation time is quite large:

julia> @time average(gt); # first execution
 26.373887 seconds (206.77 M allocations: 9.999 GiB, 3.68% gc time, 99.96% compilation time)

julia> @time average(gt); # second execution
  0.014517 seconds (330.02 k allocations: 18.696 MiB)

Should we add a precompilation script to get a better first-time experience?

[mikeingold]

Do you have a MWE for the above result? I've noticed some suboptimal compilation times before but 26 seconds for a first-use is definitely not a great experience.

I'd considered working on precompilation before but didn't know enough about it to have a good sense of what would actually be beneficial, e.g. will the precompiled code remain valid/useful for non-identical integrand functions?

[juliohm]

Hi @mikeingold, thank you for looking into it.

Below is a MWE with a single triangle:

using Meshes
using MeshIntegrals

function integrand(tri::Triangle, f)
  p -> let
    # retrieve vertices
    v = vertices(tri)

    # linear system for barycentric coordinates
    A = [(v[2] - v[1]) (v[3] - v[1])]
    b = (p - v[1])

    # normalize by maximum absolute coordinate
    m = maximum(abs, A)
    Aₘ = A / m
    bₘ = b / m

    # solve for barycentric coordinates
    w₂, w₃ = Aₘ \ bₘ

    # linear interpolation inside triangle
    f[1] + w₂ * (f[2] - f[1]) + w₃ * (f[3] - f[1])
  end
end

t = rand(Triangle)

@time integral(integrand(t, [1, 2, 3]), t, GaussLegendre(2));

The actual example repeats this call for all triangles in a mesh. You can find the test here and the definition of the average function here.

[JoshuaLampert]

This is much better with the version in Meshes.jl v0.57, right?

[juliohm]

Didn't have a chance to test it yet. Hopefully it is much faster now. We should transfer the issue to Meshes.jl otherwise. Em qua., 8 de abr. de 2026, 15:25, Joshua Lampert ***@***.***> escreveu:

…

*JoshuaLampert* left a comment (JuliaGeometry/MeshIntegrals.jl#217) <#217 (comment)> This is much better with the version in Meshes.jl v0.57, right? — Reply to this email directly, view it on GitHub <#217 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAZQW3LDMXWX3EQJRAMK4NT4U2KRVAVCNFSM6AAAAACTTLVFBCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DEMBYGUZTIMRTHE> . You are receiving this because you authored the thread.Message ID: ***@***.***>