Optimize linting single file with project cache about clj-kondo HOT 4 CLOSED

Options:

• https://github.com/luposlip/ndjson-db
  It seems like this tool is mostly optimized for reads, since for each write you would have to do a another re-index.

• http://jsondb.io/
  Reads entire database into memory, which is a disadvantage.

• Already tried nippy, but that doesn't work well with GraalVM yet. Codox needs nippy.

• https://db.apache.org/derby/
  Tried, didn't work with GraalVM.

• https://www.mcobject.com/perst/

• Fressian: tried it, but didn't compile with Clojure 1.10. It did work with Clojure 1.9. It's probably due to this issue: https://dev.clojure.org/jira/browse/CLJ-1472.

• Transit: worked well with GraalVM and had better performance than normal EDN.

• https://github.com/greglook/clj-cbor
  Haven't tried this one yet.

from clj-kondo.

For now I transitioned to Transit (see #28 and #29).

Thanks to https://github.com/clojure-goes-fast/clj-async-profiler … it's clear that about 2/3 of the time is spent in (de)serializing when linting a single file with clj-kondo. I'm optimizing for this use case (editor feedback). So that's definitely worth optimizing more.

What's left is:

• splitting the cache into files per language and namespace.

from clj-kondo.

Fixed with 9e93904

from clj-kondo.

When performance becomes a concern, we should probably also try jsonista.

from clj-kondo.