/* This Source Code Form is subject to the terms of the Mozilla Public
 * License, v. 2.0. If a copy of the MPL was not distributed with this
 * file, You can obtain one at http://mozilla.org/MPL/2.0/. */

use crate::preferences::{Pref, PrefValue, Preferences};
use std::borrow::Borrow;
use std::borrow::Cow;
use std::char;
use std::error::Error;
use std::io::{self, Write};

use std::str;
use thiserror::Error;

impl PrefReaderError {
    fn new(message: String, position: Position, parent: Option<Box<dyn Error>>) -> PrefReaderError {
        PrefReaderError {
            message,
            position,
            parent,
        }
    }
}

impl From<io::Error> for PrefReaderError {
    fn from(err: io::Error) -> PrefReaderError {
        PrefReaderError::new("IOError".into(), Position::new(), Some(err.into()))
    }
}

#[derive(Copy, Clone, Debug, PartialEq)]
enum TokenizerState {
    Junk,
    CommentStart,
    CommentLine,
    CommentBlock,
    FunctionName,
    AfterFunctionName,
    FunctionArgs,
    FunctionArg,
    DoubleQuotedString,
    SingleQuotedString,
    Number,
    Bool,
    AfterFunctionArg,
    AfterFunction,
    Error,
}

#[derive(Copy, Clone, Debug, Default, PartialEq)]
pub struct Position {
    line: u32,
    column: u32,
}

impl Position {
    pub fn new() -> Position {
        Position { line: 1, column: 0 }
    }
}

#[derive(Copy, Clone, Debug, PartialEq)]
pub enum TokenType {
    None,
    PrefFunction,
    UserPrefFunction,
    StickyPrefFunction,
    CommentBlock,
    CommentLine,
    CommentBashLine,
    Paren,
    Semicolon,
    Comma,
    String,
    Int,
    Bool,
    Error,
}

#[derive(Debug, PartialEq)]
pub enum PrefToken<'a> {
    PrefFunction(Position),
    UserPrefFunction(Position),
    StickyPrefFunction(Position),
    CommentBlock(Cow<'a, str>, Position),
    CommentLine(Cow<'a, str>, Position),
    CommentBashLine(Cow<'a, str>, Position),
    Paren(char, Position),
    Semicolon(Position),
    Comma(Position),
    String(Cow<'a, str>, Position),
    Int(i64, Position),
    Bool(bool, Position),
    Error(String, Position),
}

impl<'a> PrefToken<'a> {
    fn position(&self) -> Position {
        match *self {
            PrefToken::PrefFunction(position) => position,
            PrefToken::UserPrefFunction(position) => position,
            PrefToken::StickyPrefFunction(position) => position,
            PrefToken::CommentBlock(_, position) => position,
            PrefToken::CommentLine(_, position) => position,
            PrefToken::CommentBashLine(_, position) => position,
            PrefToken::Paren(_, position) => position,
            PrefToken::Semicolon(position) => position,
            PrefToken::Comma(position) => position,
            PrefToken::String(_, position) => position,
            PrefToken::Int(_, position) => position,
            PrefToken::Bool(_, position) => position,
            PrefToken::Error(_, position) => position,
        }
    }
}

#[derive(Debug, Error)]
#[error("{message} at line {}, column {}", .position.line, .position.column)]
pub struct PrefReaderError {
    message: String,
    position: Position,
    #[source]
    parent: Option<Box<dyn Error>>,
}

struct TokenData<'a> {
    token_type: TokenType,
    complete: bool,
    position: Position,
    data: Cow<'a, str>,
    start_pos: usize,
}

impl<'a> TokenData<'a> {
    fn new(token_type: TokenType, position: Position, start_pos: usize) -> TokenData<'a> {
        TokenData {
            token_type,
            complete: false,
            position,
            data: Cow::Borrowed(""),
            start_pos,
        }
    }

    fn start(&mut self, tokenizer: &PrefTokenizer, token_type: TokenType) {
        self.token_type = token_type;
        self.position = tokenizer.position;
        self.start_pos = tokenizer.pos;
    }

    fn end(&mut self, buf: &'a [u8], end_pos: usize) -> Result<(), PrefReaderError> {
        self.complete = true;
        self.add_slice_to_token(buf, end_pos)
    }

    fn add_slice_to_token(&mut self, buf: &'a [u8], end_pos: usize) -> Result<(), PrefReaderError> {
        let data = match str::from_utf8(&buf[self.start_pos..end_pos]) {
            Ok(x) => x,
            Err(_) => {
                return Err(PrefReaderError::new(
                    "Could not convert string to utf8".into(),
                    self.position,
                    None,
                ));
            }
        };
        if self.data != "" {
            self.data.to_mut().push_str(data)
        } else {
            self.data = Cow::Borrowed(data)
        };
        Ok(())
    }

    fn push_char(&mut self, tokenizer: &PrefTokenizer, data: char) {
        self.data.to_mut().push(data);
        self.start_pos = tokenizer.pos + 1;
    }
}

pub struct PrefTokenizer<'a> {
    data: &'a [u8],
    pos: usize,
    cur: Option<char>,
    position: Position,
    state: TokenizerState,
    next_state: Option<TokenizerState>,
}

impl<'a> PrefTokenizer<'a> {
    pub fn new(data: &'a [u8]) -> PrefTokenizer<'a> {
        PrefTokenizer {
            data,
            pos: 0,
            cur: None,
            position: Position::new(),
            state: TokenizerState::Junk,
            next_state: Some(TokenizerState::FunctionName),
        }
    }

    fn make_token(&mut self, token_data: TokenData<'a>) -> PrefToken<'a> {
        let buf = token_data.data;
        let position = token_data.position;
        // Note: the panic! here are for cases where the invalid input is regarded as
        // a bug in the caller. In cases where `make_token` can legitimately be called
        // with invalid data we must instead return a PrefToken::Error
        match token_data.token_type {
            TokenType::None => panic!("Got a token without a type"),
            TokenType::PrefFunction => PrefToken::PrefFunction(position),
            TokenType::UserPrefFunction => PrefToken::UserPrefFunction(position),
            TokenType::StickyPrefFunction => PrefToken::StickyPrefFunction(position),
            TokenType::CommentBlock => PrefToken::CommentBlock(buf, position),
            TokenType::CommentLine => PrefToken::CommentLine(buf, position),
            TokenType::CommentBashLine => PrefToken::CommentBashLine(buf, position),
            TokenType::Paren => {
                if buf.len() != 1 {
                    panic!("Expected a buffer of length one");
                }
                PrefToken::Paren(buf.chars().next().unwrap(), position)
            }
            TokenType::Semicolon => PrefToken::Semicolon(position),
            TokenType::Comma => PrefToken::Comma(position),
            TokenType::String => PrefToken::String(buf, position),
            TokenType::Int => {
                return match buf.parse::<i64>() {
                    Ok(value) => PrefToken::Int(value, position),
                    Err(_) => PrefToken::Error(format!("Expected integer, got {}", buf), position),
                }
            }
            TokenType::Bool => {
                let value = match buf.borrow() {
                    "true" => true,
                    "false" => false,
                    x => panic!("Boolean wasn't 'true' or 'false' (was {})", x),
                };
                PrefToken::Bool(value, position)
            }
            TokenType::Error => panic!("make_token can't construct errors"),
        }
    }

    fn get_char(&mut self) -> Option<char> {
        if self.pos + 1 >= self.data.len() {
            self.cur = None;
            return None;
        };
        if self.cur.is_some() {
            self.pos += 1;
        }
        let c = self.data[self.pos] as char;
        if self.cur == Some('\n') {
            self.position.line += 1;
            self.position.column = 0;
        } else if self.cur.is_some() {
            self.position.column += 1;
        };
        self.cur = Some(c);
        self.cur
    }

    fn unget_char(&mut self) -> Option<char> {
        if self.pos == 0 {
            self.position.column = 0;
            self.cur = None
        } else {
            self.pos -= 1;
            let c = self.data[self.pos] as char;
            if c == '\n' {
                self.position.line -= 1;
                let mut col_pos = self.pos;
                while col_pos > 0 {
                    col_pos -= 1;
                    if self.data[col_pos] as char == '\n' {
                        break;
                    }
                }
                self.position.column = (self.pos - col_pos) as u32;
            } else {
                self.position.column -= 1;
            }
            self.cur = Some(c);
        }
        self.cur
    }

    fn is_space(c: char) -> bool {
        matches!(c, ' ' | '\t' | '\r' | '\n')
    }

    fn skip_whitespace(&mut self) -> Option<char> {
        while let Some(c) = self.cur {
            if PrefTokenizer::is_space(c) {
                self.get_char();
            } else {
                break;
            };
        }
        self.cur
    }

    fn consume_escape(&mut self, token_data: &mut TokenData<'a>) -> Result<(), PrefReaderError> {
        let pos = self.pos;
        let escaped = self.read_escape()?;
        if let Some(escape_char) = escaped {
            token_data.add_slice_to_token(self.data, pos)?;
            token_data.push_char(self, escape_char);
        };
        Ok(())
    }

    fn read_escape(&mut self) -> Result<Option<char>, PrefReaderError> {
        let escape_char = match self.get_char() {
            Some('u') => self.read_hex_escape(4, true)?,
            Some('x') => self.read_hex_escape(2, true)?,
            Some('\\') => '\\' as u32,
            Some('"') => '"' as u32,
            Some('\'') => '\'' as u32,
            Some('r') => '\r' as u32,
            Some('n') => '\n' as u32,
            Some(_) => return Ok(None),
            None => {
                return Err(PrefReaderError::new(
                    "EOF in character escape".into(),
                    self.position,
                    None,
                ))
            }
        };
        Ok(Some(char::from_u32(escape_char).ok_or_else(|| {
            PrefReaderError::new(
                "Invalid codepoint decoded from escape".into(),
                self.position,
                None,
            )
        })?))
    }

    fn read_hex_escape(&mut self, hex_chars: isize, first: bool) -> Result<u32, PrefReaderError> {
        let mut value = 0;
        for _ in 0..hex_chars {
            match self.get_char() {
                Some(x) => {
                    value <<= 4;
                    match x {
                        '0'..='9' => value += x as u32 - '0' as u32,
                        'a'..='f' => value += x as u32 - 'a' as u32,
                        'A'..='F' => value += x as u32 - 'A' as u32,
                        _ => {
                            return Err(PrefReaderError::new(
                                "Unexpected character in escape".into(),
                                self.position,
                                None,
                            ))
                        }
                    }
                }
                None => {
                    return Err(PrefReaderError::new(
                        "Unexpected EOF in escape".into(),
                        self.position,
                        None,
                    ))
                }
            }
        }
        if first && (0xD800..=0xDBFF).contains(&value) {
            // First part of a surrogate pair
            if self.get_char() != Some('\\') || self.get_char() != Some('u') {
                return Err(PrefReaderError::new(
                    "Lone high surrogate in surrogate pair".into(),
                    self.position,
                    None,
                ));
            }
            self.unget_char();
            let high_surrogate = value;
            let low_surrogate = self.read_hex_escape(4, false)?;
            let high_value = (high_surrogate - 0xD800) << 10;
            let low_value = low_surrogate - 0xDC00;
            value = high_value + low_value + 0x10000;
        } else if first && (0xDC00..=0xDFFF).contains(&value) {
            return Err(PrefReaderError::new(
                "Lone low surrogate".into(),
                self.position,
                None,
            ));
        } else if !first && !(0xDC00..=0xDFFF).contains(&value) {
            return Err(PrefReaderError::new(
                "Invalid low surrogate in surrogate pair".into(),
                self.position,
                None,
            ));
        }
        Ok(value)
    }

    fn get_match(&mut self, target: &str, separators: &str) -> bool {
        let initial_pos = self.pos;
        let mut matched = true;
        for c in target.chars() {
            if self.cur == Some(c) {
                self.get_char();
            } else {
                matched = false;
                break;
            }
        }

        if !matched {
            for _ in 0..(self.pos - initial_pos) {
                self.unget_char();
            }
        } else {
            // Check that the next character is whitespace or a separator
            if let Some(c) = self.cur {
                if !(PrefTokenizer::is_space(c) || separators.contains(c) || c == '/') {
                    matched = false;
                }
                self.unget_char();
            }
            // Otherwise the token was followed by EOF. That's a valid match, but
            // will presumably cause a parse error later.
        }

        matched
    }

    fn next_token(&mut self) -> Result<Option<TokenData<'a>>, PrefReaderError> {
        let mut token_data = TokenData::new(TokenType::None, Position::new(), 0);

        loop {
            let mut c = match self.get_char() {
                Some(x) => x,
                None => return Ok(None),
            };

            self.state = match self.state {
                TokenizerState::Junk => {
                    c = match self.skip_whitespace() {
                        Some(x) => x,
                        None => return Ok(None),
                    };
                    match c {
                        '/' => TokenizerState::CommentStart,
                        '#' => {
                            token_data.start(self, TokenType::CommentBashLine);
                            token_data.start_pos = self.pos + 1;
                            TokenizerState::CommentLine
                        }
                        _ => {
                            self.unget_char();
                            let next = match self.next_state {
                                Some(x) => x,
                                None => {
                                    return Err(PrefReaderError::new(
                                        "In Junk state without a next state defined".into(),
                                        self.position,
                                        None,
                                    ))
                                }
                            };
                            self.next_state = None;
                            next
                        }
                    }
                }
                TokenizerState::CommentStart => match c {
                    '*' => {
                        token_data.start(self, TokenType::CommentBlock);
                        token_data.start_pos = self.pos + 1;
                        TokenizerState::CommentBlock
                    }
                    '/' => {
                        token_data.start(self, TokenType::CommentLine);
                        token_data.start_pos = self.pos + 1;
                        TokenizerState::CommentLine
                    }
                    _ => {
                        return Err(PrefReaderError::new(
                            "Invalid character after /".into(),
                            self.position,
                            None,
                        ))
                    }
                },
                TokenizerState::CommentLine => match c {
                    '\n' => {
                        token_data.end(self.data, self.pos)?;
                        TokenizerState::Junk
                    }
                    _ => TokenizerState::CommentLine,
                },
                TokenizerState::CommentBlock => match c {
                    '*' => {
                        if self.get_char() == Some('/') {
                            token_data.end(self.data, self.pos - 1)?;
                            TokenizerState::Junk
                        } else {
                            TokenizerState::CommentBlock
                        }
                    }
                    _ => TokenizerState::CommentBlock,
                },
                TokenizerState::FunctionName => {
                    let position = self.position;
                    let start_pos = self.pos;
                    match c {
                        'u' => {
                            if self.get_match("user_pref", "(") {
                                token_data.start(self, TokenType::UserPrefFunction);
                            }
                        }
                        's' => {
                            if self.get_match("sticky_pref", "(") {
                                token_data.start(self, TokenType::StickyPrefFunction);
                            }
                        }
                        'p' => {
                            if self.get_match("pref", "(") {
                                token_data.start(self, TokenType::PrefFunction);
                            }
                        }
                        _ => {}
                    };
                    if token_data.token_type == TokenType::None {
                        // We didn't match anything
                        return Err(PrefReaderError::new(
                            "Expected a pref function name".into(),
                            position,
                            None,
                        ));
                    } else {
                        token_data.start_pos = start_pos;
                        token_data.position = position;
                        token_data.end(self.data, self.pos + 1)?;
                        self.next_state = Some(TokenizerState::AfterFunctionName);
                        TokenizerState::Junk
                    }
                }
                TokenizerState::AfterFunctionName => match c {
                    '(' => {
                        self.next_state = Some(TokenizerState::FunctionArgs);
                        token_data.start(self, TokenType::Paren);
                        token_data.end(self.data, self.pos + 1)?;
                        self.next_state = Some(TokenizerState::FunctionArgs);
                        TokenizerState::Junk
                    }
                    _ => {
                        return Err(PrefReaderError::new(
                            "Expected an opening paren".into(),
                            self.position,
                            None,
                        ))
                    }
                },
                TokenizerState::FunctionArgs => match c {
                    ')' => {
                        token_data.start(self, TokenType::Paren);
                        token_data.end(self.data, self.pos + 1)?;
                        self.next_state = Some(TokenizerState::AfterFunction);
                        TokenizerState::Junk
                    }
                    _ => {
                        self.unget_char();
                        TokenizerState::FunctionArg
                    }
                },
                TokenizerState::FunctionArg => match c {
                    '"' => {
                        token_data.start(self, TokenType::String);
                        token_data.start_pos = self.pos + 1;
                        TokenizerState::DoubleQuotedString
                    }
                    '\'' => {
                        token_data.start(self, TokenType::String);
                        token_data.start_pos = self.pos + 1;
                        TokenizerState::SingleQuotedString
                    }
                    't' | 'f' => {
                        self.unget_char();
                        TokenizerState::Bool
                    }
                    '0'..='9' | '-' | '+' => {
                        token_data.start(self, TokenType::Int);
                        TokenizerState::Number
                    }
                    _ => {
                        return Err(PrefReaderError::new(
                            "Invalid character at start of function argument".into(),
                            self.position,
                            None,
                        ))
                    }
                },
                TokenizerState::DoubleQuotedString => match c {
                    '"' => {
                        token_data.end(self.data, self.pos)?;
                        self.next_state = Some(TokenizerState::AfterFunctionArg);
                        TokenizerState::Junk
                    }
                    '\n' => {
                        return Err(PrefReaderError::new(
                            "EOL in double quoted string".into(),
                            self.position,
                            None,
                        ))
                    }
                    '\\' => {
                        self.consume_escape(&mut token_data)?;
                        TokenizerState::DoubleQuotedString
                    }
                    _ => TokenizerState::DoubleQuotedString,
                },
                TokenizerState::SingleQuotedString => match c {
                    '\'' => {
                        token_data.end(self.data, self.pos)?;
                        self.next_state = Some(TokenizerState::AfterFunctionArg);
                        TokenizerState::Junk
                    }
                    '\n' => {
                        return Err(PrefReaderError::new(
                            "EOL in single quoted string".into(),
                            self.position,
                            None,
                        ))
                    }
                    '\\' => {
                        self.consume_escape(&mut token_data)?;
                        TokenizerState::SingleQuotedString
                    }
                    _ => TokenizerState::SingleQuotedString,
                },
                TokenizerState::Number => match c {
                    '0'..='9' => TokenizerState::Number,
                    ')' | ',' => {
                        token_data.end(self.data, self.pos)?;
                        self.unget_char();
                        self.next_state = Some(TokenizerState::AfterFunctionArg);
                        TokenizerState::Junk
                    }
                    x if PrefTokenizer::is_space(x) => {
                        token_data.end(self.data, self.pos)?;
                        self.next_state = Some(TokenizerState::AfterFunctionArg);
                        TokenizerState::Junk
                    }
                    _ => {
                        return Err(PrefReaderError::new(
                            "Invalid character in number literal".into(),
                            self.position,
                            None,
                        ))
                    }
                },
                TokenizerState::Bool => {
                    let start_pos = self.pos;
                    let position = self.position;
                    match c {
                        't' => {
                            if self.get_match("true", ",)") {
                                token_data.start(self, TokenType::Bool)
                            }
                        }
                        'f' => {
                            if self.get_match("false", ",)") {
                                token_data.start(self, TokenType::Bool)
                            }
                        }
                        _ => {}
                    };
                    if token_data.token_type == TokenType::None {
                        return Err(PrefReaderError::new(
                            "Unexpected characters in function argument".into(),
                            position,
                            None,
                        ));
                    } else {
                        token_data.start_pos = start_pos;
                        token_data.position = position;
                        token_data.end(self.data, self.pos + 1)?;
                        self.next_state = Some(TokenizerState::AfterFunctionArg);
                        TokenizerState::Junk
                    }
                }
                TokenizerState::AfterFunctionArg => match c {
                    ',' => {
                        token_data.start(self, TokenType::Comma);
                        token_data.end(self.data, self.pos + 1)?;
                        self.next_state = Some(TokenizerState::FunctionArg);
                        TokenizerState::Junk
                    }
                    ')' => {
                        token_data.start(self, TokenType::Paren);
                        token_data.end(self.data, self.pos + 1)?;
                        self.next_state = Some(TokenizerState::AfterFunction);
                        TokenizerState::Junk
                    }
                    _ => {
                        return Err(PrefReaderError::new(
                            "Unexpected character after function argument".into(),
                            self.position,
                            None,
                        ))
                    }
                },
                TokenizerState::AfterFunction => match c {
                    ';' => {
                        token_data.start(self, TokenType::Semicolon);
                        token_data.end(self.data, self.pos)?;
                        self.next_state = Some(TokenizerState::FunctionName);
                        TokenizerState::Junk
                    }
                    _ => {
                        return Err(PrefReaderError::new(
                            "Unexpected character after function".into(),
                            self.position,
                            None,
                        ))
                    }
                },
                TokenizerState::Error => TokenizerState::Error,
            };
            if token_data.complete {
                return Ok(Some(token_data));
            }
        }
    }
}

impl<'a> Iterator for PrefTokenizer<'a> {
    type Item = PrefToken<'a>;

    fn next(&mut self) -> Option<PrefToken<'a>> {
        if let TokenizerState::Error = self.state {
            return None;
        }
        let token_data = match self.next_token() {
            Err(e) => {
                self.state = TokenizerState::Error;
                return Some(PrefToken::Error(e.message.clone(), e.position));
            }
            Ok(Some(token_data)) => token_data,
            Ok(None) => return None,
        };
        let token = self.make_token(token_data);
        Some(token)
    }
}

pub fn tokenize(data: &[u8]) -> PrefTokenizer {
    PrefTokenizer::new(data)
}

pub fn serialize_token<T: Write>(token: &PrefToken, output: &mut T) -> Result<(), PrefReaderError> {
    let mut data_buf = String::new();

    let data = match *token {
        PrefToken::PrefFunction(_) => "pref",
        PrefToken::UserPrefFunction(_) => "user_pref",
        PrefToken::StickyPrefFunction(_) => "sticky_pref",
        PrefToken::CommentBlock(ref data, _) => {
            data_buf.reserve(data.len() + 4);
            data_buf.push_str("/*");
            data_buf.push_str(data.borrow());
            data_buf.push('*');
            &*data_buf
        }
        PrefToken::CommentLine(ref data, _) => {
            data_buf.reserve(data.len() + 2);
            data_buf.push_str("//");
            data_buf.push_str(data.borrow());
            &*data_buf
        }
        PrefToken::CommentBashLine(ref data, _) => {
            data_buf.reserve(data.len() + 1);
            data_buf.push('#');
            data_buf.push_str(data.borrow());
            &*data_buf
        }
        PrefToken::Paren(data, _) => {
            data_buf.push(data);
            &*data_buf
        }
        PrefToken::Comma(_) => ",",
        PrefToken::Semicolon(_) => ";\n",
        PrefToken::String(ref data, _) => {
            data_buf.reserve(data.len() + 2);
            data_buf.push('"');
            data_buf.push_str(escape_quote(data.borrow()).borrow());
            data_buf.push('"');
            &*data_buf
        }
        PrefToken::Int(data, _) => {
            data_buf.push_str(&data.to_string());
            &*data_buf
        }
        PrefToken::Bool(data, _) => {
            if data {
                "true"
            } else {
                "false"
            }
        }
        PrefToken::Error(ref data, pos) => {
            return Err(PrefReaderError::new(data.clone(), pos, None))
        }
    };
    output.write_all(data.as_bytes())?;
    Ok(())
}

pub fn serialize_tokens<'a, I, W>(tokens: I, output: &mut W) -> Result<(), PrefReaderError>
where
    I: Iterator<Item = &'a PrefToken<'a>>,
    W: Write,
{
    for token in tokens {
        serialize_token(token, output)?;
    }
    Ok(())
}

fn escape_quote(data: &str) -> Cow<str> {
    // Not very efficient…
    if data.contains('"') || data.contains('\\') {
        Cow::Owned(data.replace('\\', r"\\").replace('"', r#"\""#))
    } else {
        Cow::Borrowed(data)
    }
}

#[derive(Debug, PartialEq)]
enum ParserState {
    Function,
    Key,
    Value,
}

struct PrefBuilder {
    key: Option<String>,
    value: Option<PrefValue>,
    sticky: bool,
}

impl PrefBuilder {
    fn new() -> PrefBuilder {
        PrefBuilder {
            key: None,
            value: None,
            sticky: false,
        }
    }
}

fn skip_comments<'a>(tokenizer: &mut PrefTokenizer<'a>) -> Option<PrefToken<'a>> {
    loop {
        match tokenizer.next() {
            Some(PrefToken::CommentBashLine(_, _))
            | Some(PrefToken::CommentBlock(_, _))
            | Some(PrefToken::CommentLine(_, _)) => {}
            Some(x) => return Some(x),
            None => return None,
        }
    }
}

pub fn parse_tokens(tokenizer: &mut PrefTokenizer<'_>) -> Result<Preferences, PrefReaderError> {
    let mut state = ParserState::Function;
    let mut current_pref = PrefBuilder::new();
    let mut rv = Preferences::new();

    loop {
        // Not just using a for loop here seems strange, but this restricts the
        // scope of the borrow
        let token = {
            match tokenizer.next() {
                Some(x) => x,
                None => break,
            }
        };
        // First deal with comments and errors
        match token {
            PrefToken::Error(msg, position) => {
                return Err(PrefReaderError::new(msg, position, None));
            }
            PrefToken::CommentBashLine(_, _)
            | PrefToken::CommentLine(_, _)
            | PrefToken::CommentBlock(_, _) => continue,
            _ => {}
        }
        state = match state {
            ParserState::Function => {
                match token {
                    PrefToken::PrefFunction(_) => {
                        current_pref.sticky = false;
                    }
                    PrefToken::UserPrefFunction(_) => {
                        current_pref.sticky = false;
                    }
                    PrefToken::StickyPrefFunction(_) => {
                        current_pref.sticky = true;
                    }
                    _ => {
                        return Err(PrefReaderError::new(
                            "Expected pref function".into(),
                            token.position(),
                            None,
                        ));
                    }
                }
                let next = skip_comments(tokenizer);
                match next {
                    Some(PrefToken::Paren('(', _)) => ParserState::Key,
                    _ => {
                        return Err(PrefReaderError::new(
                            "Expected open paren".into(),
                            next.map(|x| x.position()).unwrap_or(tokenizer.position),
                            None,
                        ))
                    }
                }
            }
            ParserState::Key => {
                match token {
                    PrefToken::String(data, _) => current_pref.key = Some(data.into_owned()),
                    _ => {
                        return Err(PrefReaderError::new(
                            "Expected string".into(),
                            token.position(),
                            None,
                        ));
                    }
                }
                let next = skip_comments(tokenizer);
                match next {
                    Some(PrefToken::Comma(_)) => ParserState::Value,
                    _ => {
                        return Err(PrefReaderError::new(
                            "Expected comma".into(),
                            next.map(|x| x.position()).unwrap_or(tokenizer.position),
                            None,
                        ))
                    }
                }
            }
            ParserState::Value => {
                match token {
                    PrefToken::String(data, _) => {
                        current_pref.value = Some(PrefValue::String(data.into_owned()))
                    }
                    PrefToken::Int(data, _) => current_pref.value = Some(PrefValue::Int(data)),
                    PrefToken::Bool(data, _) => current_pref.value = Some(PrefValue::Bool(data)),
                    _ => {
                        return Err(PrefReaderError::new(
                            "Expected value".into(),
                            token.position(),
                            None,
                        ))
                    }
                }
                let next = skip_comments(tokenizer);
                match next {
                    Some(PrefToken::Paren(')', _)) => {}
                    _ => {
                        return Err(PrefReaderError::new(
                            "Expected close paren".into(),
                            next.map(|x| x.position()).unwrap_or(tokenizer.position),
                            None,
                        ))
                    }
                }
                let next = skip_comments(tokenizer);
                match next {
                    Some(PrefToken::Semicolon(_)) | None => {}
                    _ => {
                        return Err(PrefReaderError::new(
                            "Expected semicolon".into(),
                            next.map(|x| x.position()).unwrap_or(tokenizer.position),
                            None,
                        ))
                    }
                }
                let key = current_pref.key.take();
                let value = current_pref.value.take();
                let pref = if current_pref.sticky {
                    Pref::new_sticky(value.unwrap())
                } else {
                    Pref::new(value.unwrap())
                };
                rv.insert(key.unwrap(), pref);
                current_pref.sticky = false;
                ParserState::Function
            }
        }
    }
    match state {
        ParserState::Key | ParserState::Value => {
            return Err(PrefReaderError::new(
                "EOF in middle of function".into(),
                tokenizer.position,
                None,
            ));
        }
        _ => {}
    }
    Ok(rv)
}

pub fn serialize<W: Write>(prefs: &Preferences, output: &mut W) -> io::Result<()> {
    let mut p: Vec<_> = prefs.iter().collect();
    p.sort_by(|a, b| a.0.cmp(b.0));
    for &(key, pref) in &p {
        let func = if pref.sticky {
            "sticky_pref("
        } else {
            "user_pref("
        }
        .as_bytes();
        output.write_all(func)?;
        output.write_all(b"\"")?;
        output.write_all(escape_quote(key).as_bytes())?;
        output.write_all(b"\"")?;
        output.write_all(b", ")?;
        match pref.value {
            PrefValue::Bool(x) => {
                output.write_all(if x { b"true" } else { b"false" })?;
            }
            PrefValue::Int(x) => {
                output.write_all(x.to_string().as_bytes())?;
            }
            PrefValue::String(ref x) => {
                output.write_all(b"\"")?;
                output.write_all(escape_quote(x).as_bytes())?;
                output.write_all(b"\"")?;
            }
        };
        output.write_all(b");\n")?;
    }
    Ok(())
}

pub fn parse(data: &[u8]) -> Result<Preferences, PrefReaderError> {
    let mut tokenizer = tokenize(data);
    parse_tokens(&mut tokenizer)
}