Better than forcing it

Posted Oct 31, 2025 16:24 UTC (Fri) by nim-nim (subscriber, #34454)
In reply to: Better than forcing it by archaic
Parent article: Ubuntu introduces architecture variants

RHEL (and Fedora) have been able to do what Ubuntu announced for ages, the decision not to do it was deliberate, building for multiple x86 variants is a massive drain on mirrors and build infra, it unearths exotic assembly bugs, the pesky users mix and match binaries so your support matrix takes a direct hit, etc. So people deemed it was not worth the pain and a clean break was better for everyone involved this time.

Of course sucks to be the proud owner of legacy hardware when it passes into the museum category (as Linus would say).

Better than forcing it

Posted Oct 31, 2025 17:41 UTC (Fri) by carlos.odonell (subscriber, #99737) [Link]

Exposing the choice to end users was also the original UX problem with CPU features that the levels e.g. x86-64-v2 attempted to solve by creating a baseline that could be advanced.

We already have to consider library-based multilibs e.g. glibc-hwcaps delivery mechanism, and function multi-versioning e.g. __attribute__((target_clones(...))) in testing, and this adds to the qualification costs.

Something that I don't see talked about is that container image creation systems make static decisions at container build time and so you also end up with the entire combinatorial matrix as built containers if you, as an ISV or IHV, can't pin down what your customers are going to require.

In summary, I agree that raising the baseline is the simplest engineering solution to a complex problem, but the level depends on the needs of your community, your resources, and many other factors.

For example, see the recent "Architecture baseline for Forky" discussion in Debian: https://lists.debian.org/debian-release/2025/10/msg00471....