On 2026-01-06 2:08 p.m., Theo Buehler wrote:
Now that we have a modern gcc in our tree, we can update this port. I
thought I'd try 2.4.0 first. If there are concerns that this is too
recent, I can provide a diff for 2.3.5.
The releease notes are huge, as usual:
https://numpy.org/doc/stable/release/2.4.0-notes.html
for older releases (2.3.0 had a giant number of expired deprecations):
https://numpy.org/doc/stable/release.html
Remove DWZ workaround: the DWARF problems have been resolved one way or
the other. The riscv64 CPU feature detection was upstreamed by Brad
https://github.com/numpy/numpy/pull/29224
https://github.com/numpy/numpy/pull/29221
I also enabled the CPU feature detection for AArch64 and PowerPC64.
I ran regress on the following architectures and just started an amd64
bulk with the diff. Boost built, so I don't expect too much fallout.
Tests on other archs and runtime tests by actual users would be nice.
amd64:
= 40 failed, 47092 passed, 1094 skipped, 2817 deselected, 33 xfailed, 1
xpassed, 3 warnings in 505.25s (0:08:25) =
arm64:
= 99 failed, 46324 passed, 395 skipped, 2817 deselected, 33 xfailed, 1 xpassed,
65 warnings in 750.99s (0:12:30) =
i386:
= 27 failed, 45615 passed, 1144 skipped, 2817 deselected, 32 xfailed, 1
xpassed, 15 warnings in 899.33s (0:14:59) =
sparc64
= 87 failed, 45597 passed, 1134 skipped, 2817 deselected, 33 xfailed, 1
xpassed, 13 warnings in 2108.73s (0:35:08) =
