Add local db for performance reasons

[Panaetius]

Instead of saving a flat list as a .json or .yaml file, it'd be nice to have a local db (key-value-sture) that saves objects by IRI.

this could be done with https://docs.python.org/3/library/dbm.html#module-dbm or https://github.com/coleifer/unqlite-python or https://github.com/RaRe-Technologies/sqlitedict (this doesn't seem to be actively supported). Also check if there's other solutions that could fit our needs better.

[Panaetius]

After some more thought, adding a database to calamus is not really needed,we can use rdflib for that, since it already supports a multitude of DB backends through plugins.

Instead, we can just support an RDFLib graph object, since that already allows reading/writing triples, and RDFLib supports several backends Including SPARQL endpoints, solving #9 ). Then we don't need to worry about how to store things, and instead we just need to interact with rdflib.

The thing needed for this is to be able to deserialize rdflib triples to calamus objects and vice versa.

An easy implementation would be to just use rdflib-jsonld plugin to convert python objects <---> jsonld <---> rdflib representation. But that plugin does not support JSONLD 1.1 yet and using RDFLib types directly would likely increase performance.

So in addition to JsonLd, it'd be great if we could serialize to rdflib triples as well.

Proposal for an interface:

import rdflib

graph = Graph('Sleepycat', identifier='mygraph')

graph.open('/home/user/data/myRDFLibStore', create = True)

dataset = DatasetSchema(graph=graph).load_by_id("https://example.com/1")


DatasetSchema(graph=graph).store(dataset)

query = graph.query(
    """SELECT DISTINCT ?a
       WHERE {
          ?a rdf:type schema:Dataset .
       }""")

datasets = DatasetSchema().load_triples(query)

We could also think about supporting multiple graph, especially together with lazy:

local_graph = Graph('Sleepycat', identifier='mygraph')

local_graph.open('/home/user/data/myRDFLibStore', create = True)

remote_graph = Graph(store=SPARQLStore('http://dbpedia.org/sparql'))
remote_graph.open()

graph = calamus.FallbackGraph(local_graph, remote_graph)

DatasetSchema(graph=graph, lazy=True).load_by_id("https://example.com/1")

Which would try to access objects in local_graph but if they're not found, try the remote_graph. But this should probably be done in a followup ticket.

[rokroskar]

Graph above is rdflib.Graph?

I wonder if it could be made even simpler, it still reads maybe a little verbose to me but I really like the general idea.

I was just thinking that having a simple way to define what you want (e.g. a Person with a Name and a Birthplace), providing a really clean interface to getting that information from some valid endpoint, and followed with a means to then transform that data (into e.g. a pandas dataframe?), would be really valuable beyond the immediate context in which we happen to envision using this. For an example, have a look at this blogpost and note how complicated (ugly?) the "Retrieving SPARQL queries with Python" gets. And this is a really simple use-case.

[rokroskar]

This seems to be exactly what we need: https://github.com/RDFLib/rdflib-hdt

The interface seems to fit exactly into the use-case you sketched above.

Extra bonus: it seems currently supported unlike 99% of RDF tools out there