yade-dev team mailing list archive

[Václav Šmilauer <eu@xxxxxxxx>]

Confirming this is a "bug" (?) with boost-serialization 1.42 (default
version on maverick). Rebuilding with 1.40 (can be installed, though
take care to packages for 1.42. Interestingly enough, building with
chunkSize=1 does not exhibit the problem (with the exception of
GlShapeFunctor), which is why we were able to build (by accident)
packages for maverick.

See http://lists.boost.org/boost-users/2010/03/57775.php for an
explanation that is not yet completely clear to me. It seems to be not
bug, but a chnge in behavior of boost::serialization to properly support
shared libs (which is exactly what goes broken here).

** Changed in: yade
       Status: Confirmed => In Progress

-- 
unregistered class when built at maverick
https://bugs.launchpad.net/bugs/666876
You received this bug notification because you are a member of Yade
developers, which is subscribed to Yade.

Status in Yet Another Dynamic Engine: In Progress

Bug description:
The problems with "unregistered class" seem to re-surface at maverick. The packaged version (bzr2460) runs the attached script just fine, but manually-compiled versions fail:

This is what classes fail to save, a g++ build: ['FrictMat', 'PeriodicEngine', 'BodyCallback', 'GlIGeomFunctor', 'IntrCallback', 'Collider', 'BoundDispatcher', 'IGeomFunctor', 'ElastMat', 'LawFunctor', 'FrictPhys', 'ScGeom', 'TriaxialStressController', 'GlIPhysFunctor', 'BoundaryController', 'GenericSpheresContact', 'Dem3DofGeom', 'BoundFunctor']

This is with clang: ['PeriodicEngine', 'Recorder', 'BodyCallback', 'GlIGeomFunctor', 'FrictPhys', 'TriaxialStressController', 'GlIPhysFunctor', 'BoundaryController', 'GenericSpheresContact', 'KinemSimpleShearBox', 'Dem3DofGeom', 'Sphere']

Everything is fine with lucid.

I added these tests to the regression suite: starting from r2511, running --test will also say if some classes fail to be saved (and loaded).