kysrpex
commented
3 years ago I had a closer look at the buffer mechanism, and I would say that indeed as I believe it makes no sense. Take a look at this file. It contains the only example in the OSP-core code of what being in each buffer actually means. I am looking at the version before I introduced the buffer performance updates to make sure that the bug is there as well, as this bug is 20-25 days old.
The get_triples method is also relevant.
def get_triples(self, include_neighbor_types=False):
"""Get the triples of the cuds object."""
o_set = set()
for s, p, o in self._graph.triples((self.iri, None, None)):
yield s, p, o
o_set.add(o)
if include_neighbor_types:
for o in o_set:
yield from self._graph.triples((o, RDF.type, None))Added buffer (link)
def _apply_added(self, root_obj, buffer):
# Perform the SQL-Statements to add the elements
# in the buffers to the DB.
for added in buffer.values():
triples = self._substitute_root_iri(added.get_triples())
self._add(*triples)In plain English, for every CUDS object in the added buffer, get all the triples (from memory, the CUDS only exists in memory) that have the CUDS IRI as subject. Put them in the backend.
In even simpler English: copy the CUDS from the memory to the backend.
Updated buffer (link)
def _apply_updated(self, root_obj, buffer):
# Perform the SQL-Statements to update the elements
# in the buffers in the DB.
for updated in buffer.values():
pattern = (updated.iri, None, None)
self._remove(next(self._substitute_root_iri([pattern])))
triples = self._substitute_root_iri(updated.get_triples())
self._add(*triples)In plain English, for every CUDS object in the added buffer, remove all triples that have the CUDS IRI as subject from the backend. Then get all the triples (from memory, the CUDS only exists in memory) that have the CUDS iri as subject. Put them in the backend.
In even simpler English: remove the CUDS from the backend. Copy the CUDS from the memory to the backend.
Deleted buffer (link)
def _apply_deleted(self, root_obj, buffer):
# Perform the SQL-Statements to delete the elements
# in the buffers in the DB.
for deleted in buffer.values():
pattern = (deleted.iri, None, None)
self._remove(next(self._substitute_root_iri([pattern])))In plain English, for every CUDS object in the added buffer, remove all triples that have the CUDS IRI as subject from the backend.
In even simpler English: remove the CUDS from the backend.
So, guess what happens if you put the wrapper in the added buffer? All its triples get added to the backend, while the old ones are NOT removed. Ideally, one would "load the wrapper from backend" and return the loaded wrapper instead of a new one, but there are several checks in place to prevent this and the solution is more complex.
Simply never putting wrappers in the added buffer seems to be good enough, since they are always supposed to be the root of the session, so their existence only makes sense when they are attached to something. And when they are attached to something, they go to the updated buffer, where what happens actually makes sense according to the code above.
paulzierep
Not putting the wrapper in the added buffer solves the problem.