On Mon, Aug 02, 2010 at 09:37:14AM -0400, Lucas Nussbaum wrote: > severity 591346 serious > thanks > > On 02/08/10 at 13:55 +0200, Sylvestre Ledru wrote: > > Package: hdf5 > > Severity: important > > > > Hello, > > > > More and more packages are using hdf5 packages. > > Some of them (for example, Code Saturne) are using hdf5 mpi by default while > > others, like Scilab, needs the serial one. > > > > This is causing also some issues at build time. See bug #591164. > > > > To reproduce it: > > aptitude install libhdf5-openmpi-1.8.4 libhdf5-serial-1.8.4 > > > > By the way, if the ABI are compatible, #586149 might be the solution. > > > > Thanks for considering this. > > > > If you need help on this, don't hesitate to ask. > > Sylvestre > > This bug blocks 2 RC bugs, and I don't see any other resolution path for > those RC than fixing this bug. > So I'm upgrading this bug to severity:serious too, to make sure it stays > on the radar. >
WTF is the reason to both depend on serial and parallel flavors at building time instead of the parallel version libhdf5-mpi-dev? Note that *if the ABI are compatible* is a wrong guess AFAIK: serial and parallel versions have different shlib sub-dependendecies. There are also many limitations in parallel edition: e.g. parallel edition does not support pluggable compression in writing and does not support variable length records, cannot coexists with threadsafe and C++, and so on. I think the only safe and clean approach would be having different lib names for MPI flavors, but upstream think differently at the moment. I'm not intentioned to diverge about that, which would require rdepends changing their building snippets forever in Debian, and adopt other dirty tricks. -- Francesco P. Lovergine -- To UNSUBSCRIBE, email to debian-bugs-rc-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
