Re: Seems I have bad timing with Debian on SPARC.
On Sat, Oct 5, 2013 at 9:36 PM, Bernhard R. Link wrote: > * Jurij Smakov [131005 12:38]: > > That's the theory. In reality, maintainers of large and complex software > > projects (like mozilla/firefox) do not really care about fringe > > architectures, and I don't see why this situation would improve with > time. > > Large and complex software has many bugs so maintainers will not care > for all of them equally. Having people care for them because the hit > them on their architecture causes them to be fixed before they come back > to bite everyone. > > > A pragmatic (but less conceptually-correct) approach would be to convince > > sparc kernel maintainers to introduce unaligned memory access handling > for > > userspace programs. > > For me that would make sparc totally uninteresting. Without the ability > to find bugs (which sparc was always very good at, even though alignment > was even stricter on hppa), sparc would just be another architecture > hardly worth supporting at all, especially as the hardware is no more > found as commonly as in former times and there is no longer that much a > difference in quality so that using has become more a liability than > a stability boost. > I really doubt that at this point sparc (well, Linux on sparc) is doing anyone a service by finding bugs. Vast majority of problems we saw in the past are unaligned access problems, which are not really bugs on other architectures - "fixing" them will probably not make the binary run faster on x86. So, when we find and file them, typically nobody cares. One spectacular example is https://bugzilla.mozilla.org/show_bug.cgi?id=161826- it took over 7 *years* for this bug to be declared fixed. The fact that the current "iceweasel crashes on sparc" bug ( http://bugs.debian.org/674908) was open (with "grave" severity) for almost a year and was eventually tagged wheezy-ignore to prevent it from blocking the last release is an indication that Debian's release managers are adopting a similar attitude - and I don't blame them. Releasing Debian is a huge task, and expecting to delay the release because iceweasel is crashing for a few dozen people who bother running it on sparc is not reasonable. I don't want to discourage you (or anyone else), but I think that sparc as a Debian port is facing some serious problems, which can potentially lead to its demise in not-so-distant future, same way it happened to sparc32. Preventing binaries crashing on unaligned memory accesses would keep if afloat a bit longer (and you can make the behavior configurable, of course) - if I would still be a port maintainer, I would pursue this goal. > > Bernhard R. Link > > > -- > To UNSUBSCRIBE, email to debian-sparc-requ...@lists.debian.org > with a subject of "unsubscribe". Trouble? Contact > listmas...@lists.debian.org > Archive: http://lists.debian.org/20131005203607.ga2...@client.brlink.eu > > -- Jurij Smakov | ju...@wooyd.org | Key IDs: 43C30A7D/C99E03CC
Re: Seems I have bad timing with Debian on SPARC.
* Jurij Smakov [131005 12:38]: > That's the theory. In reality, maintainers of large and complex software > projects (like mozilla/firefox) do not really care about fringe > architectures, and I don't see why this situation would improve with time. Large and complex software has many bugs so maintainers will not care for all of them equally. Having people care for them because the hit them on their architecture causes them to be fixed before they come back to bite everyone. > A pragmatic (but less conceptually-correct) approach would be to convince > sparc kernel maintainers to introduce unaligned memory access handling for > userspace programs. For me that would make sparc totally uninteresting. Without the ability to find bugs (which sparc was always very good at, even though alignment was even stricter on hppa), sparc would just be another architecture hardly worth supporting at all, especially as the hardware is no more found as commonly as in former times and there is no longer that much a difference in quality so that using has become more a liability than a stability boost. Bernhard R. Link -- To UNSUBSCRIBE, email to debian-sparc-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/20131005203607.ga2...@client.brlink.eu
Re: Seems I have bad timing with Debian on SPARC.
On Sat, Oct 5, 2013 at 10:57 AM, Bernhard R. Link wrote: > * Howard Eisenberger [131004 20:20]: > > (2) "Bus error" with iceweasel and iceape. I believe it's been > > like this for a couple of years now. > > Those are big problems and getting them bug free can be a big task, > though sparc is really good to catch the bugs here. A valid C program > cannot cause a bus error on sparc and an invalid C program is likely > to fail also on other architectures whenever the compiler adds new > optimisations. So fixing those bugs will benefit everyone. > That's the theory. In reality, maintainers of large and complex software projects (like mozilla/firefox) do not really care about fringe architectures, and I don't see why this situation would improve with time. A pragmatic (but less conceptually-correct) approach would be to convince sparc kernel maintainers to introduce unaligned memory access handling for userspace programs. That would incur a penalty every time an unaligned access happens, but, in my opinion, it's better to have a slow-but-working binary than the one which crashes all the time. The code for handling unaligned memory accesses for kernel code already exists, so I don't think it would be too big of a challenge for someone who knows their way around kernel code. On a related note, I posted a patch to http://bugs.debian.org/674908 (was RC, but eventually got tagged wheezy-ignore) which allowed me to browse some javascript-heavy sites. > > Bernhard R. Link > > > -- > To UNSUBSCRIBE, email to debian-sparc-requ...@lists.debian.org > with a subject of "unsubscribe". Trouble? Contact > listmas...@lists.debian.org > Archive: http://lists.debian.org/20131005095745.ga2...@client.brlink.eu > > -- Jurij Smakov | ju...@wooyd.org | Key IDs: 43C30A7D/C99E03CC
Re: Seems I have bad timing with Debian on SPARC.
* Howard Eisenberger [131004 20:20]: > (2) "Bus error" with iceweasel and iceape. I believe it's been > like this for a couple of years now. Those are big problems and getting them bug free can be a big task, though sparc is really good to catch the bugs here. A valid C program cannot cause a bus error on sparc and an invalid C program is likely to fail also on other architectures whenever the compiler adds new optimisations. So fixing those bugs will benefit everyone. Bernhard R. Link -- To UNSUBSCRIBE, email to debian-sparc-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/20131005095745.ga2...@client.brlink.eu