On Tue, Jun 19, 2018 at 05:36:22PM +0200, KatolaZ wrote: > On Mon, Jun 18, 2018 at 03:41:40PM +0200, Adam Borowski wrote: > > On Mon, Jun 18, 2018 at 02:11:33PM +0200, KatolaZ wrote: > > > + amd64: 61% > > > + i386 24% > > > > This is troubling. There's a strong reason to deprecate i386, and relegate > > it to a level akin to mips or s390x -- ie, with security support but > > requiring a conscious decision to install. There are people with hardware > > that genuinely lacks the NX bit, but it's only them who should be running > > i386. You can read one of Linus' rants when discussing techniques needed > > for mitigating melted spectrum on 32-bit kernels on 64-bit capable hardware > > (TL;DR: it really sucks performance wise, compared to vulnerable > > (out-of-order) but 32-bit only CPUs), that's just one of reasons why. > > > > I guess you make download links for i386 installer way too prominent, > > without offering an explanation. > > uh? How are the i386 links "way too prominent"? The isos are just > there, and there is no blinking link. If people download those images, > it means that they need them. Why should Devuan "hide" them away?
The problem is that people are not told why they should away from i386, and when faced with a choice they don't understand they often make a bad decision. > I mean, there are not many distributions out there still offering > support for i686 hardware, and we have actually received many emails > of users who thank Devuan for supporting i686 and letting them > continue using their "old" hardware that you would like to see > discontinued. I'm not suggesting dropping support -- to the contrary, shifting back to 586 would be a good idea! My point is, there's no way 28% of x86 users are on pre-2004 hardware, thus the vast majority of them got the wrong version that's insecure (no Meltdown/Spectre mitigation, no NX, worse ASLR, etc), has issues with "large" memory (even phones these days tend to have 4+GB RAM), can't run an increasing portion of software, and gets weak or no support from many upstreams. All of those would be better served with amd64. > I know that the technologically sound solution in our western (rich) > world would be to buy newer hardware as soon as it is available, but > you don't always have that option, as strange as that migh sound to > you. Used hardware is drastically cheaper, yeah. But only to a point -- once you get to machines so old to require i386, you start getting ridiculous electricity usage. I still have a Pentium 4 desktop I sometimes power on to test stuff -- it draws so much juice that you can buy three i3 laptops for the money it takes to keep that P4 running for a year. Thus, Core 2 Duo would be the weakest hardware that's common and not counterproductive to use. You can get heaps of these going to landfills. > And the recently discovered flaws in newer CPUs cast some doubt > about the technological soundness of the consumerist "let's get the > latest on the market" approach anyway. Right, in-order hardware is safe here. I like my 2017 Pinebook I bought for $89. It would be nifty if you could run Devuan on such i386 machines, right? Right?!? Oh well, did you patch gcc defaults to re-enable older CPUs and rebuilt the world? Because if not, they just happen to sit right on the line cutting off any in-order traditional machines (I say "traditional" as there was a short run of 32-bit Atoms a bit later). Thus, this argument doesn't hold water. > I remain strongly convinced that maintaining i686 support as long as > possible should remain an ongoing objective for Devuan. For people who actually need it -- I fully agree with you. But for folks on amd64-capable hardware: no way! Right now, you promote i386 as strongly as amd64; compare to https://debian.org where you need to actually look if you want i386 installer. Meow! -- ⢀⣴⠾⠻⢶⣦⠀ There's an easy way to tell toy operating systems from real ones. ⣾⠁⢰⠒⠀⣿⡁ Just look at how their shipped fonts display U+1F52B, this makes ⢿⡄⠘⠷⠚⠋⠀ the intended audience obvious. It's also interesting to see OSes ⠈⠳⣄⠀⠀⠀⠀ go back and forth wrt their intended target. _______________________________________________ Dng mailing list Dng@lists.dyne.org https://mailinglists.dyne.org/cgi-bin/mailman/listinfo/dng
