On 2025/08/14 13:05, Kirill A. Korinsky wrote:
> On Sun, 10 Aug 2025 04:33:37 +0200,
> George Koehler <kernigh@gmail.com> wrote:
> > 
> > On Tue, 05 Aug 2025 17:37:20 +0200
> > Kirill A. Korinsky <kirill@korins.ky> wrote:
> > 
> > > Thanks, I'll run build on mips64 but I have no idea how long does it take.
> > > And it hasn't got numpy. So, just build two gcc. Yeah.
> > 
> > I did a powerpc build, with gas-2.45, numpy, and all their
> > dependencies.  (It took about 18 hours on a dual G5 at 2.0 GHz.)
> > 
> > I discovered that "make test" in numpy crashes quickly on powerpc.
> > The cause is clang bug 56023 "PowerPC ABI incompatibility with GNU on
> > complex arguments", SIGSEGV as npy_csqrtf compiled by gcc calls
> > csqrtf(3) compiled by clang.  (Our fortran module changes numpy's
> > C compiler to gcc.)
> > 
> > gas-2.45 didn't break numpy and looks good on powerpc.
> > 
> > I will probably not try gas-2.45 on powerpc64 before you commit it.
> 
> As far as I understand it hasn't broke anything.
> 
> Shall I go forward and commit it?

I think that's ok.

Any feedback, Pascal?