Re: py-scipy build failure on NetBSD 9.3

[Greg Troxel <gdt%lexort.com@localhost>]

"Dr. Thomas Orgis" <thomas.orgis%uni-hamburg.de@localhost> writes:

> Am Mon, 6 May 2024 18:55:10 +0200
> schrieb Thomas Klausner <wiz%gatalith.at@localhost>:
>
>> > Ah, the BLAS_C_INTERFACE was missing. This was a subtle difference …
>> > scipy used not to use CBLAS, now it does.  
>> 
>> No, it was before, but the line was removed during the last update -
>> probably because pkglint warns about it being set somewhere else too,
>> but it seems, too late.
>
> (sorry for me simply not remembering the exact situation, the mind
> tries to escape the hurt from dealing with BLAS and the pythons)
>
> Oh … something to fix in pkglint or in the BLAS machinery, then?
>
> NOTE: Makefile:64: Definition of BLAS_C_INTERFACE is redundant because of ../../math/py-numpy/buildlink3.mk:25.
>
> It clearly is not redundant and order of definitions seems to matter.
> Can I change that in blas.buildlink3.mk?

Also, if scipy uses the C interface itself, rather than just relying
that numpy does, then it needs to define it.   What if numpy tomorrow
worked w/o blas?  If scipy still needs it, it's not redundant.  This
smells like a pkglint bug to me.