socialite - wanna be private-first social network
socialite is proof-of-concept social network that experiments various backend architectural choices.
socialite wants to proove that a complex application can be developed and operated more easily as a monolithic service using the right abstractions. That's why socialite use FoundationDB.
2018/09/30 - What Are The Civilian Applications
- Continous Integration [DONE]
- Basic Data Persistence [DONE]
- Example use of
- Baisc Feed Reader [TODO]
- Basic Task queue [TODO]
- Example Unit Test that mocks a coroutine [TODO]
- Deploy [TODO]
2018/10/XY - Pick a Culture ship at random
- Basic TODO
- Basic Wiki
- Basic Forum
- Basic Paste
- CSRF Protection
- Basic Search Engine with a crawler
Functions for the win
socialite use a lot of functions. There is nothing wrong with classes. In particular there is no Object Data Mapper (ODM) or Object Relational Mapper (ORM) abstraction, yet.
That said, socialite rely on
trafaret for data
validation which is built using classes. Also socialite make use of
SocialiteException class that you can inherit.
Socialite rely on FoundationDB (FDB) to persist data to disk. Becareful the default configuration use the in-memory backend. The goal with this choice is double:
Experiment with higher level database abstractions (called layers FDB jargon) on top the versatile ordered key-value store offered by FDB.
Experiment operations of FDB from development to deployement of single machine cluster to multiple machine clusters.
src/socialite/sparky.py offers an abstraction similar to rdf /
SPARQL. It implements a subset of the standard that should be very
easy to get started.
To get started you can read FDB's documentation about the Python client. Mind the fact that socialite rely on a fork of found that is asyncio driver for FDB based on cffi (which is the recommeded way to interop with C code by PyPy).
Of course it would be very nice to have well-thought, easy to use, with migration magics. socialite proceed step-by-step. Implement, use, gain knowledge, then build higher level abstractions. When things seem blurry, do not over think it and try something simple to get started.
sparky is small RDF-like layer which support a subset of SPARQL.
Simply said, it's a quad-store.
Let's try again.
Simply said, it stores a set of 4-tuples of primitive datatypes
dict is not
supported as-is)) most commonly described as:
(graph, subject, predicate, object)
But one might have an easier time mapping that machinery to the easier to the mind when you come from MongoDB:
(collection, uid, field, value)
The difference with a document store is that tuples are very unique! Which makes sense since it is a set ot tuples. Otherwise said, you can have the following three tuples in the same database:
("blog", "P4X432", "title", "hyperdev.fr") ("blog", "P4X432", "SeeAlso", "julien.danjou.info") ("blog", "P4X432", "SeeAlso", "blog.dolead.com")
This is not possible in document-store because the
Querying in RDF land happens via a language "similar" to SQL that is called SPARQL. Basically, it's pattern matching with bells and dragons... That being said, sparky implements only the pattern matching part which makes coding things like the following SQL query:
SELECT post.title FROM blog, post WHERE blog.title='hyperdev.fr' AND post.blog_id=blog.id
Here is the equivalent using sparky:
patterns = [ ('blog', sparky.var('blog'), 'title', 'hyperdev.fr'), ('post', sparky.var('post'), 'blog', sparky.var('blog')), ('post', sparky.var('post'), 'title', sparky.var('title')), ] out = await sparky.where(db, *patterns)
That is you can do regular
SELECT without joins or a
multiple joins in a single declarative statment. See the unit tests
See this superb tutorial on SPARQL at data.world.
The roadmap is to implement something like datomic without versioning.
The current implementation is very naive but the API is (almost) set.
Basically, going further means implementing a cost / statistic based
query planner. That said, for the time being we will not rely on
explicit indexing but instead index-all-the-things. Another machinery
might be put to good use that is inspired from Gremlin and
implemented in AjguDB, if one can make
async for in a transaction work without leaks.
Mind the fact, that since sparky use
fdb.pack for serialiazing a
tuple items, lexicographic ordering is preserved. That is, one can
defer complex indexing to upper layer namely the application ;]
Styles Style Guide
- Do no rely on LESS or SASS
- Only rely on classes and tags
- Avoid class when tag is sufficent to disambiguate
- Prefix class names with component name to avoid any leak
- Avoid cascade ie. all styles must appear in the class declaration (ie. it is not DRY)
- When it makes sens, be precise in the selector (most of the time it must start with