Repository navigation
Test dependency accounts for almost all the load time #38
Description
This package loads two others, InverseFunctions and Test. Claude measured their load times for me (Julia 1.13.1):
┌──────────────────┬─────────────────────────────────────┬─────────────┐
│ Package │ Six fresh-process runs (ms) │ Median (ms) │
├──────────────────┼─────────────────────────────────────┼─────────────┤
│ InverseFunctions │ 8.9, 2.2, 11.4, 2.0, 4.8, 2.4 │ ~2–5 │
├──────────────────┼─────────────────────────────────────┼─────────────┤
│ Test │ 231.3, 72.3, 74.7, 83.3, 73.8, 98.1 │ ~79 │
├──────────────────┼─────────────────────────────────────┼─────────────┤
│ DensityInterface │ 84.5, 77.1, 84.9, 84.0, 67.8, 71.2 │ ~80 │
└──────────────────┴─────────────────────────────────────┴─────────────┘
This is a clear argument for splitting Test support out as an extension.
Activity
timholy commentedon Oct 15, 2026
Since this is a breaking change, a related question: when that's done, should we bump to 0.5 or 1.0?
devmotion commentedon Oct 15, 2026
Seems reasonable to me, I somehow thought it's an extension already - as in InverseFunctions. I wonder if some of the load time might be due to the surprisingly slow
__init__function in InverseFunctions: JuliaMath/InverseFunctions.jl#67timholy commentedon Oct 15, 2026
Huh, I would have thought that would have been included in Claude's measurements. Did you measure that with any other packages loaded? If so, my precompilation suggestion won't help if it's invalidation.
gdalle commentedon Oct 15, 2026
I think moving to an extension is reasonable
devmotion commentedon Oct 15, 2026
No, empty environment with
--startup-file=no. But it seems I was just measuring a@time_importsartifact: JuliaMath/InverseFunctions.jl#67 (comment)oschulz commentedon Oct 15, 2026
If it gives us a real benefit, fine with me as well. We also use an Test extension in InverseFunctions.jl and ChangesOfVariables.jl.
I guess it would be time for a 1.0 then?