You are not logged in.
The next release of pacman (after pacman 5.1.3-1.3) will set the architecture to "pentium4" if you have
Architecture = autoin your /etc/pacman.conf and your machine supports sse2 instructions.
This requires reinstallation (and if necessary a downgrade) of all installed packages, e.g. by running
pacman -Syy
pacman -Qqn | pacman -S -Note, that this is only possible and necessary after pacman > 5.1.3-1.3 gets installed.
Also note, that it's not a good idea to stick with i686 if your machine supports sse2, as more and more packages require sse2 and will be dropped from the i686 repositories, thus.
Edit: There seems to be a flaw in the sse2 detection logic of pacman 5.1.3-1.4 - it correctly switches to pentium4 on my test vm, but not on my test real hardware (both with sse2). You can check with
pacman-conf Architectureif it changes to pentium4 for you - and apply the above steps only if it does so.
I'm working on it ... stay tuned.
Offline
Upgraded to pacman 5.1.3-1.4 on my EeePC, and reverted my pacman.conf to Architecture=auto. pacman-conf Architecture still returned pentium4 and pacman -Syu correctly reported nothing-to-do. So, I think it's all working for me on real hardware.
Architecture: pentium4, Testing repos: Yes, Hardware: EeePC 901+2GB RAM+OS half on the SD card.
Offline
What happens with the i686 repositories (Pentium 3 without SSE2, others) ... Can it be updated safely? ... Is it time to abandon those PCs?
Architecture: i686, Hardware: Pentium 3, Coppermine, 1GHz + 256MB RAM, WM: BSPWM/2BWM.
Offline
Until something you absolutely need to use on that computer gets dropped you can continue to use archlinux32 with it. The auto setting in your pacman.conf should equate to i686 on that machine, and the i686 repos will still get updated with the applications that can get built for it.
This does make me wonder what debian is doing however. They target i386 for their 32-bit builds and given some stuff requires sse2 to build, I'd expect more to require things like mmx.
But as of now, update your system, run the pacman-conf command listed above, and check it comes back with i686.
Architecture: pentium4, Testing repos: Yes, Hardware: EeePC 901+2GB RAM+OS half on the SD card.
Offline
After the last update, pacman 5.1.3-1.4 had already been installed and I had never changed the Architecture-entry from "auto". "pacman-conf Architecture" responded with "pentium4", so I started with the full switch to the new architecture (my EeePC 1000H was still on i686).
All went well, however, after the update, "pacman-conf Architecture" suddenly responded with "i686". I then found out that the architecture switch had downgraded pacman back to 5.1.3-1.3 and there seems to be no update candidate for pentium4. Hence, I changed pacman.conf to "Architecture=pentium4" for the time being.
Is that supposed to happen or did I mess something up? Other than that (and the fact that a number of packages are still i686), the system is running just fine...
Offline
That could well be the errant behaviour that deep42thought referred to, but thanks for proving it can happen in the real world.
Actually for you I wonder if there's something up with your package mirrors in fact. I'm on 5.1.3-1.4 and on pentium4, which I guess came from the testing repo. There's now a -1.5 in testing, but core is still on -1.3.
Looking in i686, that has -1.3 and -1.4 in core. This seems to be a discrepancy that will catch anyone not using the testing repos, although if you only use one -u to your pacman -Sy invocation you might avoid the mess.
Architecture: pentium4, Testing repos: Yes, Hardware: EeePC 901+2GB RAM+OS half on the SD card.
Offline
Actually for you I wonder if there's something up with your package mirrors in fact. I'm on 5.1.3-1.4 and on pentium4, which I guess came from the testing repo. There's now a -1.5 in testing, but core is still on -1.3.
Looking in i686, that has -1.3 and -1.4 in core. This seems to be a discrepancy that will catch anyone not using the testing repos, although if you only use one -u to your pacman -Sy invocation you might avoid the mess.
Just to confirm: I'm indeed not using the testing repos, so that would make sense
Offline
Okay, that's good to know. So the fix for this is to make the pentium4 repos like the i686 repos by either promoting pacman 5.1.3-1.4 to core in pentium 4, or demoting the same back to testing in i686. If you leave it as is there's a possibility others will be left in this to-and-fro quandry.
All I can say is that as a pentium4 user, -1.4 has worked fine for me. I'm not sure what's in -1.5 yet as I'm kind of waiting for a new kernel before doing another update.
Architecture: pentium4, Testing repos: Yes, Hardware: EeePC 901+2GB RAM+OS half on the SD card.
Offline
The next release of pacman (after pacman 5.1.3-1.3) will set the architecture to "pentium4" if you have
more and more packages require sse2 and will be dropped from the i686 repositories
Just wondering (and this may be OT) but given some packages somehow/seem to already require SSE2 (e.g. Firefox*), might a middle-ground be to build these "i686" packages with `-march=pentium4` ?
They may break on some hardware but it might be that dependent packages might still work, and the net effect is essentially the same.
(* This may not be the case any more, but I think I remember that although Firefox was compiled `-march=i686` it still segfaulted on non-SSE2 hardware)
Offline
I don't have any packages installed that depend on firefox. What sort of thing are you thinking of that requires firefox to be installed but doesn't really use it. If necessary, I think it better to define a basically empty firefox package in that case at this stage, fwiw, rather than installing duff binaries on people's systems.
Architecture: pentium4, Testing repos: Yes, Hardware: EeePC 901+2GB RAM+OS half on the SD card.
Offline
I was thinking more along the lines of `webkit2gtk`, though I just found 2.24.2 compiled with `-march=i686`...
I suppose it doesn't really matter if SSE2-capable machines are migrated to `pentium4` and `i686` drops the packages. It's the same end result - working packages on `pentium4`, intended lack of working packages on `i686`.
Last edited by jonathon (2019-05-29 22:13:28)
Offline
Ah yes, that makes more sense. I have shotwell installed which depends on that both directly and via libgdata (and gnome-online-accounts), for reasons I've not yet discovered. I connect my camera to my computer over usb, import photos, then edit them via gimp, so I have no idea which part of its usability needs a web client.
But shotwell is a bit on the heavy side for me. I certainly wouldn't be using it on my old pentium3 machine which I couldn't even run a graphical server on (although that probably has something to do with the fact I was trying to use the on-board graphics rather than installing a card and having driver woes). However, there are probably lighter things that would be more suitable for such old hardware that somehow depend on webkit2gtk. If dillo went for example, it would take out claws-mail, and there's probably something else fairly lightweight which depends on webkit2gtk (neither of which are actually going from i486 yet, FWIW). For the example of claws-mail, shipping an entirely empty web client package would probably work; so long as you don't install the dillo html viewer plugin it should all work just fine.
Architecture: pentium4, Testing repos: Yes, Hardware: EeePC 901+2GB RAM+OS half on the SD card.
Offline