On Fri 22/10/2021 12:16, Dima Pasechnik wrote: > On Fri, Oct 22, 2021 at 09:55:19AM +0200, Martin Reindl wrote: > > Am 22.10.2021 um 09:49 schrieb Stuart Henderson: > > > Only issue in the i386 build was math/py-h5py which specifically wants > > > 1.16.5. > > > > > > > Yes, this is one patch which can go away with newer numpy. > > numpy by itself doesn't really use Fortran - it only sets up a configuration, > used by other Python packages, > mainly SciPy. So a good test of this being ready is to build and test SciPy > using the NumPy.
I ran scipy's regression tests using both numpy-1.16.5 and numpy-1.19.5. No new failing tests. logs attached. Using numpy-1.16.5: = 5 failed, 36305 passed, 2226 skipped, 102 xfailed, 7 xpassed, 2 warnings in 2787.49 seconds = Using numpy-1.16.5: = 5 failed, 36319 passed, 2212 skipped, 102 xfailed, 7 xpassed, 12 warnings in 3222.45 seconds =
test.log-numpy-1.16.5.log.gz
Description: application/gunzip
test.log-numpy-1.19.5.log.gz
Description: application/gunzip