(I am Cc: debian-science, since this is probably specific to science) On 30.04.2014 03:45, Thomas Goirand wrote: > This is the exact same thing that happens with .so libraries, and > this should happen *only* when there's an API change (why would you > want to keep an older version otherwise?).
Because the *behaviour* changes. As an example: Consider that someone does a spectral analysis and needs the spectra in a certain binning. For this, he uses a some (debianized :-) ) software. However he finds that the resulting spectra are somehow shifted from what he expected. In this case, many scientists I know would *not* go by and debug the rebinning software. They would take the software as a black box and just investigate how to shift the spectrum back to its expected position. This is then tested on known data, and finally used for the analysis. And if he wrote his own piece of software, this correction goes into the source (and even may be packaged on Debian). If now the rebinning software is updated and the bug is corrected, the analysis of these scientist will not anymore be correct. Therefore, many scientists fear updates of their operating systems, and especially of the scientific libraries they use. I don't think that this behaviour is really correct (in a scientific sense), and I am not sure whether Debian should support this -- but the acceptance by scientists raises if they can (co-)install certain versions, even if they are API compatible. CERN did this (at least) with their own CERNLIB, which was installed (far before FHS) on /cern/<version> with links /cern/pro /cern/new /cern/old to the current, the next (to be tested) and the previous release. ESO pipelines (cpl-plugin-* on Debian) also have the version number in their installation path. The main reason is that an average scientist (PhD student which a short-term contract) just does not have the experience nor the resources to investigate the software he depends on to actually find the bug (or even to find out which software is responsible). Best regards Ole -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org Archive: https://lists.debian.org/5360a5da.60...@liska.ath.cx
