Formatting for the Masses Part II
summary
Demonstrates two concrete formatters: a grammar-based system using Codebuff + ANTLR that trains on token vectors computed from 21 grammar-derived features and uses k nearest neighbors with k=11 neighbors, and a Deep Learning prototype that learns whitespace and linebreak decisions from token windows. Both integrate with Eclipse/Xtext, the ML model reaches about 95–98% on Google Guava after hours of one‑time training, and the prototypes work on Java and DSL samples but remain experimental, not production-ready.
AI-written summary · about
chapters (5)
- 0:00Start
- 1:31But there are already formatters out there!
- 2:24Manual Formatting FTW
- 2:56Xtext Rich API to Define Formatters
- 22:36Let's try it out
related talks
20mAn Approach to Generate Text-Based IDEs for Syntax Correction based on Syntax Specification
14mAutomating the Synthesis of Recommender Systems for Modelling Languages
53mDomain Specifc Languages
01h 29mScalafix Workshop
56mBuilding My Own Clojure Tools
34mTerrible tokenizer troubles in large language models
31mMake your own code formatter in Swift
27mDefault Disambiguation in Online Parsers