On 26. 09. 22, 17:54, Lubos Kocman wrote:
sorry to yet again for the thread but this time I think this time we have some really good news!
1. Community (and mostly -v2 internal infra) was heard ALP will have minimal architecture baselines set to x86_64-v2, according to
Fine for ALP, I don't care, but:
DimStar Factory will adopt the change first!
NACK (from a person who was taking care about tumbleweed with Greg at the beginning)
Provided one of the non-avx procs is still rather wide-spread Core 2 Duo (correct me if I am wrong), you cannot simply cut them all off.
SUSE plans to investigate hwcaps for v3 and potentially v4 support. Users of v1 can still use our 32 bit Intel images/rpms.
You must be kidding.
1) i586 support is literally _NONE_. At least from the kernel side, but I am sure every other package is 2nd class citizen on 32bit. Ask also x86 kernel maintainers like Boris what they think before you suggest something like this.
2) You are going to support old 32 sh*t and not modern 64bit processors? Who is going to do that and how? Noone cares (in both meanings of the word) about 32bit nowadays. Are you going to hire devs to support 32bit? I doubt that. You are just going to throw the passengers overboard.
3) If you suggest people running 32bit kernel on a HW which already supports 64bit, I will stop providing 32bit kernel for TW (as it's merely provided, not supported). With that, I will also remove the x86_64-v2 flags from the x86_64 kernel build, so that it still runs on procs like Core 2 Duo (and with your suggested 32bit userspace). There are several reasons for that. From the main ones: using more than 4G of RAM without stupid remapping all the time; no limit of 3G VM (or 800M of vmap) space etc.
I personally run some machines with Core 2 Duo to support drivers in the kernel for devices on the (now legacy) PCI bus. I see no single reason to discontinue that. Core 2 Duo are modern enough processors even nowadays, IMNSHO. You'd have to wait more years before
This also means that we won't have to figure out how to do a reasonable benchmark in between v1 and v3! :)
You'd have to. You have to persuade people why you are doing the step in the first place. It makes no sense to me.
So again, NACK for tumbleweed. Do whatever you want in your ALP experiment.