Skip to content

Latest commit

 

History

History
104 lines (68 loc) · 4.55 KB

TODO.md

File metadata and controls

104 lines (68 loc) · 4.55 KB

TODOs

Short Term

Client modules

In Python, Node.js, the usual suspects. Even cooler would be a node.js/Gremlin bridge that gave you the graph object.

Response wrapper details

Query time, statistics, that sort of thing.

Better run-iterator centralization

It's everywhere now, with subtly different semantics. Unify and do cool things (like abort).

More test coverage

Always good. Break out test_utils and compare text and Javascript outputs.

More documentation

Also always good.

Anything marked "TODO" in the code.

Usually something that should be taken care of.

Bootstraps

Start discussing bootstrap quads, things that make the database self-describing, if they exist (though they need not). Talk about sameAs and indexing and type systems and whatnot.

Better surfacing of Label

It exists, it's indexed, but it's basically useless right now

Optimize HasA Iterator

There are some simple optimizations that can be done there. And was the first one to get right, this is the next one. A simple example is just to convert the HasA to a fixed (next them out) if the subiterator size is guessable and small.

Gremlin features

Mid-query Limit

A way to limit the number of subresults at a point, without even running the query. Essentially, much as GetLimit() does for the end, be able to do the same in between

"Up" and "Down" traversals

Getting to the predicates from a node, or the nodes from a predicate, or some odd combinations thereof. Ditto for label.

Value comparison

Expose the value-comparison iterator in the language

MQL features

See also bootstrapping. Things like finding "name" predicates, and various schema or type enforcement.

An important failure of MQL before was that it was never well-specified. Let's not fall in that trap again, and be able to document what everything means.

New Iterators

Limit Iterator

The necessary component to make mid-query limit work. Acts as a limit on Next(), a passthrough on Contains(), and a limit on NextPath()

Medium Term

Direct JSON-LD loading

Because it's useful markup, and JSON is easy for Go to deal with. The NQuads library is nicely self-contained, there's no reason more formats can't be supported.

Value indexing

Since I have value comparison. It works, it's just not fast today. That could be improved.

Postgres Backend

It'd be nice to run on SQL as well. It's a big why not?

Generalist layout

Notionally, this is a simple quad table with a number of indices. Iterators and iterator optimization (ie, rewriting SQL queries) is the 'fun' part

"Short Schema" Layout?

This one is the crazy one. Suppose a world where we actually use the table schema for predicates, and update the table schema as we go along. Yes, it sucks when you add a new predicate (and the cell values are unclear) but for small worlds (or, "short schemas") it may (or may not) be interesting.

New Iterators

Predicate Iterator

Really, this is just the generalized value comparison iterator, across strings and dates and such.

Longer Term (and fuzzy)

SPARQL and more traditional RDF

There's a whole body of work there, and a lot of interested researchers. They're the choir who already know the sermon of graph stores. Once ease-of-use gets people in the door, supporting extensions that make everyone happy seems like a win. And because we're query-language agnostic, it's a cleaner win. See also bootstrapping, which is the first goal toward this (eg, let's talk about sameAs, and index it appropriately.)

Replication

Technically it works now if you piggyback on someone else's replication, but that's cheating. We speak HTTP, we can send quad sets over the wire to some other instance. Bonus points for a way to apply morphisms first -- massive graph on the backend, important graph on the frontend.

Related services

Eg, topic service, recon service -- whether in Cayley itself or as part of the greater project.

New languages

Javascript is nice for first-timers but experienced graph folks may want something more. Experiment with new languages, including but not limited to things that feel a lot like Datalog.

Meta-stores

Imagine an in-memory graph cache wrapped around another store.

New Iterators

Cross Iterator

This is the nutty one that's undefined and slow. The notion is to make graph-shapes more efficient, but it's unclear how that'll happen.

All sorts of backends:

Git?

Can we access git in a meaningful fashion, giving a history and rollbacks to memory/flat files?

ElasticSearch

Cassandra

Redis