On 8/18/25 09:44, Samuel Thibault wrote:
Hello,
Andreas Beckmann, le lun. 18 août 2025 03:47:39 +0200, a ecrit:
I'd recommend switching from the whitelist to a shorter blacklist
excluding the architecures where papi does not build.
Mmm. The list would indeed be shorter, but the build failures are not
due to bugs, but due to missing porting, i.e. "by default" for a new
arch, papi would fail to build. So I'm not sure we'd want to use a
blacklist since by default an arch would be on it. But maybe papi is
usually very quick to add an architecture, and it's only the old or
niche ones (s390x) that they don't care about?
upstream 7.2.0 adds support for loongarch64 and riscv64, the latest
architectures added to Debian. papi support for new architectures
probably depends on the hardware vendors to provide documentation and
patches. loongarch64 support was backported to the 7.1 Debian packages
since someone had filed a bug and sent a patch ;-)
So I think we are fine assuming that new architectures get support
within reasonable time. And a new architecture FTBFS would serve better
as a hint to go looking for a patch while silent success doesn't remind
to update the architecture list. Regardless how the architecure lists
are written, we both would probably miss if at some point an "old"
architecture gains papi support.
I tried to binNMU starpu-contrib, but that didn't work because it misses
the 1.4.8+dfsg-2+ changes (and therefore has a B-D in experimental).
Andreas
