1. 18

    1. 3

      This is a shining example of the points I was making on the Ninja thread the other day

      • Build systems have a lot of diversity – Fortran is an example of that. (The R language is also an interpreted language around linear algebra routines written in Fortran.)
      • Diverse build problems in FOSS are often harder to solve than monorepo problems
      • Windows is a big problem, a special case that “ruins” otherwise good solutions. But it still needs to work, for almost all major projects.


      This is a nice and detailed post, explaining the problem, which is very complex. It’s kind of amazing that it was done at all, and it doesn’t surprise me that it took funding.

      We are also very grateful that this work was supported in part by a grant from NASA to SciPy, scikit-learn, pandas and NumPy under the NASA ROSES 2020 program.

      Also I like this perspective:

      Depending on where you draw the line, hundreds or even thousands of person years of effort went into the ingredients that were necessary for us to achieve the final result (SciPy, Meson, Flang, LLVM, LAPACK etc.).

    2. 1

      I would have loved to see the reaction of whoever was the first to witness those 55k tests pass with no failures.