On Fri, Oct 16, 2026 at 09:38:57PM +0200, Francois Mazen wrote: > Hi Adrian,
Hi Francois, > Yes there was a technical reason to limit the architecture because the code > did > use specific intel SIMD instruction not available everywhere. that's what NO_SIMD=ON in debian/rules seems to handle. > You may look at > the Salsa CI pipeline history for information [1] > > However, it was back to the 1.12 version and the initial packaging as an > OSPRay > dependency. Maybe with recent 1.15 version, the intel limitations are not > valid > anymore? It might just be bad luck that i386 is the second architecture in salsa. > Feel free to check other architectures and if successful I'll happy to change > to > "Architecture: any" i386 was the other architecture where I tried building before submitting the bug, and I got an error there that looked similar to [1]. Without much debugging this looks like an i386-only problem to me, something like tbb trying to use SSE headers. blender had already become 32-bit-only, for that it does not matter whether rkcommon is available on i386. I don't know if i386 is the only archtecture where rkcommon fails to build, but at least on riscv64 the NO_SIMD=ON codepath works and I'd be optimistic that it works on more architectures. > Best, > François >... cu Adrian [1] https://salsa.debian.org/science-team/rkcommon/-/jobs/5085752
