Bill Allombert: > On Tue, Sep 06, 2016 at 06:31:28PM +0200, Samuel Lelièvre wrote: >> I think PARI has a particular convention about even vs odd minor version >> numbers, >> so 2.7.x and 2.8.x don't work the same. >> >> See >> http://pari.math.u-bordeaux.fr/faq.html#versioninfo > > Exactly so. Old timers might remember that the Linux kernel used the > opposite convention up to the release of Linux 2.6.0 > > What it mean for PARI is that all pari-2.7.x have the same ABI and > sonames while the pari-2.6.x had different ABI and sonames. > > The soname of pari 2.7.x is libpari-gmp-tls.so.4, the soname of > pari 2.9.x will be libpari-gmp-tls.so.5 but the soname of > pari 2.8.x is libpari-gmp-tls-2.8.so.x (which depends on x). > > So the package name 'libpari-gmp-tls4' is not appropriate for > pari-2.8.x >
Understood, thank you both for the explanation. This also explains why we have to currently do this https://anonscm.debian.org/cgit/debian-science/packages/sagemath.git/tree/debian/README.Debian#n6 to make the Sage docbuild work; it seems that they are still using pari-2.8 as if it had the same sonames and paths as pari-2.7. Perhaps then a more long-term solution would be something like: 1. Ask pari upstream to fix installation paths so that pari-stable and pari-testing are co-installable. 2. Package pari-2.8.0 as its own Debian package. 3. Fix Sage's pari paths to point to the pari-testing paths. Does this sound feasible to you? (To be clear this is a longer term goal; I still think we should just Ignore This Problem in the meantime for debian-sagemath, for the reasons I mentioned earlier.) X -- GPG: ed25519/56034877E1F87C35 GPG: rsa4096/1318EFAC5FBBDBCE https://github.com/infinity0/pubkeys.git
