Create a Database Index for Jedi

[davidhalter]

For a lot of things (especially usages) jedi's completely lazy approach is not good enough. It is probably better to use a database index cache. The index will basically be a graph that saves all the type inference findings.

This is just an issue for discussion and collection of possible ideas.

[hajs]

Recently I wrote my own source indexer using jedi.
I used a sqlite databases with four tables: file, name, definition and reference.
Indexing stdlib took about four minutes using multiprocessing on four cores,

It would be great if the index could be exposed to the public api in some way.
Besides finding all usages of a definition, a database could be used to offer auto imports and fast fuzzy auto-complete.

[davidhalter]

That's actually pretty fast. Did you index all the subfolders (asyncio, multiprocessing, json, etc)?

Also can you post the script? I wonder if it's "complete".

[davidhalter]

@hajs I would still be interested :)