Discussed in the community meeting today. While NumPy 1.26 is EOL according to Spec0, and SciPy has dropped it from main, they won't be releasing for a bit longer. @rgommers has mentioned:
scientific-python/specs#386 (comment)
The recommendation is definitely the right thing to do, so I'd say don't drop NumPy 1.26 support until mid 2026. SPEC 0 is already very agressive, and any gaps in a regular release schedule or something out of the ordinary like an ABI break exacerbate that (and numpy 2.0 has both of those, it was also 6 months late).
I believe the way forward is to keep 1.26 in pandas 3.0, and look to drop it at the appropriate time in pandas 3.1 or 3.2 depending on when those get released.
cc @jorisvandenbossche @mroeschke @jbrockmendel @Dr-Irv
Discussed in the community meeting today. While NumPy 1.26 is EOL according to Spec0, and SciPy has dropped it from main, they won't be releasing for a bit longer. @rgommers has mentioned:
scientific-python/specs#386 (comment)
I believe the way forward is to keep 1.26 in pandas 3.0, and look to drop it at the appropriate time in pandas 3.1 or 3.2 depending on when those get released.
cc @jorisvandenbossche @mroeschke @jbrockmendel @Dr-Irv