Why SciPy builds for Python 3.12 on Windows are a minor miracle (especially in conda-forge)

For a change of pace from standards discussions, I thought I’d share the resolution/recap of a packaging story that I’ve been following for a few years now. In the thread announcing conda-forge support for 3.12, I had noted:

In case you want to know just how outsized that impact was, I wrote up the story of how the removal of distutils caused a complete overhaul in the scipy build system (moving to meson), and how that put us in an extremely tight spot in conda-forge w.r.t. lack of usable Fortran compilers.

Things could have ended up being stuck quite badly – and for a long time it looked like that’s exactly what was going to happen – but in the end, things worked out juuuuust in the nick of time. :sweat_smile: