Well, the set of all patches runs passes the tests on flute when run in
valgrind (other than the dna read which fails for a virgin svn copy).
And all pass (without valgrind) on my mac. I suspect the problem is
changes you made to the patches I submitted, but it is also possible
that the patches simply expose bugs created by the problems in attribute
table that a later patch fixed.
The first was the correct one - I missed the IMP_ADD_OBJECT(s) that you
added to IMP_CONTAINER_SWIG. Should be fixed now though.
Such things are really delicate in IMP and there are, I'm sure, still a
whole bunch of places where the thisown things are missing. I wish we
had a more robust solution. One possibility is that if you can put a
hook into swig on each member function which checks if each argument is
either an Object or a vector of objects and calls thisown appropriately.
I can write C++ code which tests if an object should have thisown called.