I apparently never paid enough attention to how in the spec there's a clear defined list and I always assumed that a "parser" should handle all valid header*is

I apparently never paid enough attention to how in the spec there's a clear defined list and I always assumed that a "parser" should handle all valid header*ish looking pairs. Based on this consideration I was thinking that the ragel state machine would generate faster code for the non happy path (invalid non-ascii, or other types of error) at least in the GOTO version. When working on the full list it makes perfect sense to check the minimum amount of bytes for identifying headers, so thank you for the clarifications, very informative. :)