//! The string content type. //! //! **String** is a limited [text][] like content type which only allows //! character escapes and character references. //! It exists in things such as identifiers (media references, definitions), //! titles, URLs, code (fenced) info and meta parts. //! //! The constructs found in string are: //! //! * [Character escape][crate::construct::character_escape] //! * [Character reference][crate::construct::character_reference] //! //! [text]: crate::content::text use crate::construct::{ character_escape::start as character_escape, character_reference::start as character_reference, }; use crate::tokenizer::{Code, State, StateFnResult, TokenType, Tokenizer}; // To do: line endings? /// Before string. /// /// ```markdown /// |& /// |\& /// |qwe /// ``` pub fn start(tokenizer: &mut Tokenizer, code: Code) -> StateFnResult { match code { Code::None => (State::Ok, None), _ => tokenizer.attempt_2(character_reference, character_escape, |ok| { Box::new(if ok { start } else { before_data }) })(tokenizer, code), } } /// Before string, not at a character reference or character escape. /// /// We’re at data. /// /// ```markdown /// |qwe /// ``` fn before_data(tokenizer: &mut Tokenizer, code: Code) -> StateFnResult { if let Code::None = code { (State::Ok, None) } else { tokenizer.enter(TokenType::Data); tokenizer.consume(code); (State::Fn(Box::new(in_data)), None) } } /// In data. /// /// ```markdown /// q|w|e /// ``` fn in_data(tokenizer: &mut Tokenizer, code: Code) -> StateFnResult { match code { // To do: line endings. Code::None => { tokenizer.exit(TokenType::Data); (State::Ok, None) } // To do: somehow get these markers from constructs. Code::Char('&' | '\\') => { tokenizer.exit(TokenType::Data); start(tokenizer, code) } _ => { tokenizer.consume(code); (State::Fn(Box::new(in_data)), None) } } }