ANTLR4 to TextMate grammar translation

Here is an example of what would happen if we translated the given ANTLR4 lexer grammar naively into a TextMate syntax-highlighting grammar:

We call this the shadowing problem. The core issue is that ANTLR-generated lexers differ in behavior from TextMate-based syntax highlighters when the input matches multiple rules:

1. an ANTLR lexer picks the rule that matches the most characters from the input (→ longest-match)
2. a TextMate tokenizer picks the first rule that finds any match (→ leftmost-match)

This also applies to alternatives (e.g. ('a' | 'ab')) within a single rule due to the way the Oniguruma regex engine works.

Just reordering the rules doesn’t help:

// TODO: write about the solution