Ivan Kochurkin

> @mike-lischke I don't think fragment rules are visible to the parser. They are only meant as helper rules for other Lexer rules. Exactly. BTW, we have a feature request...

@parrt it's also ready. Tool tests will be fine after #3779

> Can you help me understand the prime motivation here? There's a lot to review and a lot of changes that could affect allow users, because it touches the tool...

> Are you certain of your benchmark? The benefits seem to small compared to what it should be. 3 times performance improvement is small? You can try the benchmark yourself....

Interesting, it looks like literals much more effective than refs, and maybe more effective than bitwise check: ``` check_by_if_test: 436 ns check_by_range_test: 650 ns check_by_range_with_literals_test: 176 ns check_by_bitwise_test: 203 ns...

> Well, yes. Compared to 15×2 = 30 dereferences, I was expecting one or two orders of magnitude, not just 1/3 of the time. Maybe dereference is very fast operation...

Anyway `110` and `100` is not a big difference and usually there are not so many tokens (up to 20). Constants are used not only in this places but in...

Please elaborate or suggest a pull request. ANTLR is compatible with non-Windows machines. `net45` for Windows was added for better compatibility with .NET Framework: https://github.com/antlr/antlr4/issues/3212

Got it. It looks like C# runtime should always build on Windows machines.

I haven't touched DFA yet, but it's interesting to look into.