Re:Linux Distributions You Can Rely on for Your Ancient 32-B...
by cbmuser on Thursday November 05, 2020 @12:14PM
(#24999921)
If the librsvg package in Debian i386 crashes with SIGILL on a Pentium II or Pentium III, it’s a baseline violation and should be reported as a bug against src:librsvg with severity serious.
What you describe here is a Debian Policy violation.
Re:Linux Distributions You Can Rely on for Your Ancient 32-B...
by merb on Thursday November 05, 2020 @12:49PM
(#25000263)
tbf. I still hope that at some point in time the i386 code will be completly removed in linux, even on debian.
microsoft won't remove such code, they even keep some 16-bit stuff in windows even if it just does not work anymore.
I mean yeah it's cool to run an old machine with top up to date linux, but at some point we still should move on.
Re:Linux Distributions You Can Rely on for Your Ancient 32-B...
by gnu8 on Thursday November 05, 2020 @12:16PM
(#24999949)
I completely understand the desire to use the newer instructions and get the most performance possible out of the latest hardware, but we also should make it a priority to keep a portable C codepath in all of these libraries. Otherwise we’re going to find there are only two architectures left in the world because none of the software runs on anything else.
Re:Linux Distributions You Can Rely on for Your Ancient 32-B...
by CJefferson on Thursday November 05, 2020 @12:37PM
(#25000149)
There is a clean codepath, this is about setting the minimum CPU required for x86 -- there is a trade-off, we could make executables that will run on a 386, but it would be slow (particularly if multithreaded) due to later added instructions. Also, such old systems are little tested.
Re:Linux Distributions You Can Rely on for Your Ancient 32-B...
by zozbot234 on Thursday November 05, 2020 @05:00PM
(#25002516)
> As a consequence of this issue, I had to manually downgrade librsvg on a older machine in order to get Buster functional.
How old is your hardware? AIUI, this should only affect Pentium Pro to Pentium III hardware, which is often limited by other factors anyway (particularly RAM). Even old netbook CPU's ought to support SSE2.
In fact, that bug report was closed with a suggestion to use i586 as a build target if you have an early i686 machine - would that be satisfactory for your use case?
https://github.com/rust-lang/rust/issues/14441
As a consequence of this issue, I had to manually downgrade librsvg on a older machine in order to get Buster functional.