Make Pythran a required build dependency (original) (raw)

Pythran has been an opt-out build dependence since SciPy 1.7; we’ve been toying with the idea of removing the opt-out for a while (way longer than the RFC issue exists).

I propose to make this official for 1.14.

rgommers April 12, 2024, 8:02am 2

We haven’t had many issues with Pythran in the last 2 years, and the ones we did have were resolved pretty swiftly. So from that perspective it seems fine to make Pythran required.

I did have a quick search for use of -Duse-pythran=false, and found that two distros build with that flag set by default:

Before taking this decision, it should be investigated what the problem exactly is there.

Did anyone reach out to Termux/PTXdist? It looks like the authors of those flags are licy183 (Chongyun Lee) · GitHub and michaelolbrich (Michael Olbrich) · GitHub respectively.

stefanv April 6, 2025, 3:37pm 4

One potential problem: in the Windows Intel oneAPI job we set -Duse-pythran=False (scipy/.github/workflows/windows_intel_oneAPI.yml at 41641c2574363ccde8a9177004082e12e18aef71 · scipy/scipy · GitHub) because MSVC chokes on Pythran, and it sounded from MSVC throws errors in Pythran code of SciPy · Issue #2215 · serge-sans-paille/pythran · GitHub like that isn’t in the pipeline to change.

Is this a blocker, or is there any chance of moving away from MSVC for oneAPI on Windows? cc @czgdp1807 @rgommers

rgommers April 7, 2025, 12:56pm 7

I’m fine with the status quo actually. There isn’t that much to gain from making Pythran required, and being able to not use it has been a useful workaround for build problems multiple times in the past. And yes, the MSVC situation is an issue - we don’t use MSVC ourselves, but in principle we do allow users to build with MSVC + Intel Fortran.