-------- Message transféré -------- Sujet : Re: cernlib (20061220+dfsg3-4.3build1) misbehaves on x86_64 Date : Wed, 13 Feb 2019 09:58:12 +0100 De : Jacek M. Holeczek <jacek.m.holec...@gmail.com> Pour : Matthias Klose <d...@ubuntu.com>, Gilles Filippini <p...@debian.org> Dear Sirs, these are very old problems which nobody cares about. There is a bug report from 2014: https://bugs.launchpad.net/ubuntu/+source/paw/+bug/1354692 I myself was notified about these problems (by my colleague) in 2017. In that time, I found that your cernlib was broken on "Ubuntu 16.04 / x86_64 " and "Ubuntu 16.10 / x86_64". However, the forthcoming (in that time) "Ubuntu 17.10 / x86_64" seemed to be free from these problems. So I started to advice people to move to this newer distribution (in "beta" in that time), even though, I myself use the "monolithic" CERNLIB 2005 distribution. It turns out, however, that I was wrong and these problems are still there in "Ubuntu 18.04 / x86_64". In general, the current status (for your cernlib compiled with "-O3" or "-O2") is that there are severe problems with "ntuples". One cannot properly create "ntuples", neither "memory resident" nor "disk resident", in paw. In both cases, only the first variable has proper values, the remaining variables are always set to 0 (well, usually, sometimes another value appears). Old "disk resident ntuples" (created with another properly working cernlib versions) are sometimes retrieved properly, but sometimes also this fails (that's what the old bug report from 2014 is talking about). So, I tried to leave your "cernlib-20061220" libraries compiled with the default "-O3" and rebuild your "paw-2.14.04" with "-O0" but it did not help (i.e. the problems really originate in the "cernlib-20061220" package, not in "paw-2.14.04"). Note that the original cernlib was always built with "-O". More aggressive compiler optimizations were known to break it. You can find my instructions here: https://www-zeuthen.desy.de/linear_collider/cernlib/new/cernlib_2005.html Last, but not least. It seems I cannot file a bug report "anonymously". Your bug reporting system tries to force me to create and register a new account in order to file a bug report. I'm sick of such behavior. I already have tens of account somewhere there. I won't create another one just for a simple bug report (when nobody cares about 5 years old bugs anyhow). Best regards, Jacek.
signature.asc
Description: OpenPGP digital signature
