Στις 08-03-2017, ημέρα Τετ, και ώρα 17:34 +0100, ο/η John Paul Adrian Glaubitz έγραψε: > I'm not aware of any of such -altivec packages and stuff like atlas > should be compiled from source on the target machine anyway for > performance reasons. That could be applied to many/most/all? packages, so why bother packaging them in the first place? Sorry, that's not a very good argument. > I did a quick search and I could only find ardour-altivec. Any > others? Interesting, atlas-altivec is not released anymore, well, either it's not supported for powerpc or they do runtime detection. Regardless, we used to have more -altivec tagged packages, so either we have better runtime detection or we're running seriously underperforming packages. > You shouldn't make such statement without actually mentioning the > section of the Debian Policy which states that Altivec has to be > turned off by default or which makes a generic statement regarding > this. I did not mean to sound offensive, but it's just a fact. If a package is built with altivec, it will fail on all non-altivec CPUs, so that by definition makes the package unusable, hence a grave bug report. By all means, let's change the minimum architecture requirements so that it's not, and we set Altivec as mandatory but until we do that (and I actually would vote for that with both hands), it is a grave bug. No need to even quote policy for that, as right now powerpc port page states those machines as supported. > Also, powerpc is not a release architecture anymore, so I don't even > know how relevant such statements from the Debian Policy would be > nowadays. Again I don't disagree there, but since it's not a release architecture, we could just move on to newer CPUs and change the minimum specs -which would IMHO be a good thing, as it would lower the number of supported platforms to a much smaller and more manageable set. I have enough PowerPC boxes here, but none without Altivec, so even if I wanted to, I couldn't even test lack of the feature, and I don't care enough to go into the trouble of doing that. Regards Konstantinos
Attachment:
signature.asc
Description: This is a digitally signed message part