Cmake vs waf driver#
Cmake vs waf full#
Cmake vs waf code#

pyproject.toml-based builds) to interact with pip and other PythonĪn RFC titled 'switch to Meson as a build system',Īnd after positive feedback started working on this project. One of the goals of this exercise is to get away from distutils/setuptoolsĬompletely and use only a modern build system and Python packaging standards Also, scikit-build still depends on setuptools and Larger user base, as well as better Python integration than Meson via Projects using either CMake or Meson, good docs and an easy to understandĬode base are very important - we're going to have to contribute! CMake has a Has great documentation (unlike CMake), and it'sĮasy to contribute to (it is ~25,000 lines of clean Python code, compared to After someĮxperimentation, I had a strong preference for Meson for two main reasons: it Really there were only two viable candidates: CMake and Meson.

That left the question: which build system? Those things together made me realize that it's time to move to a betterīuild system - if we are forced to do a lot of work because of PEP 632Īnyway, then now is the time. Problem through four (!) different projects that all modified how extensions Earlier in the year I spent half a day hunting a build Finally, debugging build issues is quite difficult - distutilsĭoesn't really have a design, so every other extension just monkey patches Of other maintainers have told me it's a blocker for them to work on newįeatures.

Cmake vs waf code#

Working on compiled code is cumbersome and slow, to the point where a couple CI systems frequently time out because build plus test suite run takesĬlose to 60 minutes (the limit on several free-for-open-source CI offerings). The growing amount of C++ and Cython code has increased build SciPy contains C, C++, Fortran and Cython code, exports C and CythonĪPIs, has complex dependencies (see diagram below) and does a lot of code

Cmake vs waf driver#

The other driver was that SciPy development has become more painful over Painful process - after almost a year, the vendored distutils still hasn't It has become clear though that this will be a really slow and Parts of numpy.distutils into setuptools, like Fortran support (see Setuptools maintainers still plan to do, with a cleaned up API) and then Instinct was to wait until setuptools integrated distutils (which the That PEP was written almost a year ago, and at the time my first SciPy - and even more on numpy.distutils, which directly extendsĭistutils. Of distutils from the Python standard library would have a major impact on Uses numpy.distutils and setuptools as its build system, so the removal In Python 3.10 distutils will be deprecated and in Python 3.12 it will be So the 1min 50 sec build time is close to optimal - the only ways to improve it are major surgery on that C++ code, or buying a faster CPU. ( bsr.cxx, which is one of SciPy's sparse matrix formats) takes over 90 Visualization created with ninjatracing and Perfetto.Īs you can see from the tracing results, building a single C++ file Profiling result of a parallel build (12 jobs) of SciPy with Meson. So how fast is it? Currently the build takesĪbout 1min 50s (a ~4x improvement) on my 3 year old 12-core Intel CPU

This is a pretty exciting milestone, and good news for SciPy maintainers andĬontributors - they can look forward to much faster builds and a more

Cmake vs waf full#

Meson on Linux, and the full test suite passes! Let's start with an announcement: SciPy now builds with