//! This module contains all pre-defined strategies provided by this crate. All
//! of them are re-exported in [crate::solver::strategy], so you should not
//! have to `use` anything from this module directly.

use crate::constraint::Constraint;
use crate::solver::strategy::{Strategy, SudokuInfo};
use crate::util::USizeSet;

use std::collections::HashSet;

/// A [Strategy] which detects naked singles, that is, cells which only have
/// one possible value, and enters them into the Sudoku.
///
/// As a small example, take a look at the following grid:
///
/// ```text
/// ╔═══╤═══╦═══╤═══╗
/// ║ X │   ║   │ 2 ║
/// ╟───┼───╫───┼───╢
/// ║   │ 1 ║   │   ║
/// ╠═══╪═══╬═══╪═══╣
/// ║   │   ║   │   ║
/// ╟───┼───╫───┼───╢
/// ║ 3 │   ║   │   ║
/// ╚═══╧═══╩═══╧═══╝
/// ```
///
/// The cell marked with X cannot be a 1 because of the 1 in its block, nor a 2
/// because of the 2 in its row, and also cannot be a 3 because of the 3 in its
/// column. Consequently, it can only be a 4. This would be detected by this
/// strategy.
#[derive(Clone)]
pub struct NakedSingleStrategy;

impl Strategy for NakedSingleStrategy {

    fn apply(&self, sudoku_info: &mut SudokuInfo<impl Constraint + Clone>)
            -> bool {
        let size = sudoku_info.size();
        let mut changed = false;

        for row in 0..size {
            for column in 0..size {
                let options = sudoku_info.get_options(column, row).unwrap();

                if sudoku_info.get_cell(column, row).unwrap() == None &&
                        options.len() == 1 {
                    let option = options.iter().next().unwrap();
                    sudoku_info.enter_cell_no_update(column, row, option).unwrap();
                    changed = true;
                }
            }
        }

        sudoku_info.invalidate();
        changed
    }
}

#[derive(Clone)]
enum Location {
    None,
    One(usize, usize),
    Multiple
}

impl std::fmt::Display for Location {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        match self {
            Location::None => f.write_str("None"),
            Location::One(a, b) => f.write_str(format!("One({}, {})", a, b).as_str()),
            Location::Multiple => f.write_str("Multiple")
        }
    }
}

impl Location {
    fn union(&self, column: usize, row: usize) -> Location {
        match self {
            Location::None => Location::One(column, row),
            Location::One(_, _) => Location::Multiple,
            Location::Multiple => Location::Multiple
        }
    }
}

/// A [Strategy] which detects situations in which a digit can only go in one
/// cell of a group.
///
/// As a visualization, the cell marked with X in the following example is the
/// only one in its block that can be a 2 (using classic Sudoku rules).
///
/// ```text
/// ╔═══╤═══╦═══╤═══╗
/// ║   │   ║   │ 2 ║
/// ╟───┼───╫───┼───╢
/// ║ X │ 1 ║   │   ║
/// ╠═══╪═══╬═══╪═══╣
/// ║   │   ║   │   ║
/// ╟───┼───╫───┼───╢
/// ║   │   ║   │   ║
/// ╚═══╧═══╩═══╧═══╝
/// ```
#[derive(Clone)]
pub struct OnlyCellStrategy;

impl Strategy for OnlyCellStrategy {

    fn apply(&self, sudoku_info: &mut SudokuInfo<impl Constraint + Clone>)
            -> bool {
        let size = sudoku_info.size();
        let grid = sudoku_info.sudoku().grid();
        let groups = sudoku_info.sudoku().constraint().get_groups(grid);
        let mut changed = false;

        for group in groups {
            if group.len() < size {
                // For smaller groups, there is no guarantee that all digits
                // are present.
                continue;
            }

            let mut locations = vec![Location::None; size + 1];

            for (column, row) in group {
                if sudoku_info.get_cell(column, row).unwrap().is_some() {
                    continue;
                }

                let options = sudoku_info.get_options(column, row).unwrap();

                for option in options.iter() {
                    let location = &locations[option];
                    locations[option] = location.union(column, row);
                }
            }

            for (number, location) in locations.into_iter().enumerate() {
                if let Location::One(column, row) = location {
                    sudoku_info.enqueue_enter_cell(column, row, number)
                        .unwrap();
                    changed = true;
                }
            }
        }

        sudoku_info.invalidate();
        changed
    }
}

/// A [Strategy] which searches groups for tuples, that is, 2 or more cells
/// that in total have as many options as there are cells in the tuple. It then
/// excludes all of these options from all cells in the group which are not a
/// part of the tuple.
///
/// As an example, consider the following configuration (with standard Sudoku
/// rules):
///
/// ```text
/// ╔═══╤═══╤═══╦═══╤═══╤═══╦═══╤═══╤═══╗
/// ║ A │ A │ A ║ 4 │ 5 │ 6 ║ 7 │ 8 │ 9 ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║ B │ B │ B ║ 1 │ 2 │ 3 ║ 4 │ 5 │ 6 ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║   │   │ X ║   │   │   ║   │   │   ║
/// ╠═══╪═══╪═══╬═══╪═══╪═══╬═══╪═══╪═══╣
/// ║   │   │ 4 ║   │   │   ║   │   │   ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║   │   │ 5 ║   │   │   ║   │   │   ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║   │   │   ║   │   │   ║   │   │   ║
/// ╠═══╪═══╪═══╬═══╪═══╪═══╬═══╪═══╪═══╣
/// ║   │   │   ║   │   │   ║   │   │   ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║   │   │   ║   │   │   ║   │   │   ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║   │   │   ║   │   │   ║   │   │   ║
/// ╚═══╧═══╧═══╩═══╧═══╧═══╩═══╧═══╧═══╝
/// ```
///
/// Because the first row already contains the digits 4-9, the cells marked
/// with A must contain the digits 1-3, meaning they are a triple (3-tuple).
/// Similarly, the cells marked with B must contain the digits 7-9. This
/// excludes the options 1-3 and 7-9 from the cell marked with X. The 4 and 5
/// in the third column then fix it to 6.
///
/// When creating a tuple strategy using [TupleStrategy::new], the maximum size
/// of tuples that are considered can be defined.
#[derive(Clone)]
pub struct TupleStrategy<F: Fn(usize) -> usize> {
    max_size_computer: F
}

impl<F: Fn(usize) -> usize> TupleStrategy<F> {

    /// Creates a new tuple strategy that considers tuples up to the size
    /// defined by `max_size_computer`. This closure receives the size of the
    /// grid and outputs the maximum size of tuples that this strategy shall
    /// consider.
    pub fn new(max_size_computer: F) -> TupleStrategy<F> {
        TupleStrategy {
            max_size_computer
        }
    }
}

#[derive(Clone)]
struct Tuple {
    cells: HashSet<(usize, usize)>,
    options: USizeSet
}

impl Tuple {
    fn new(size: usize) -> Tuple {
        Tuple {
            cells: HashSet::new(),
            options: USizeSet::new(1, size).unwrap()
        }
    }

    fn add_cell(&mut self, options: &USizeSet, column: usize, row: usize) {
        self.cells.insert((column, row));
        self.options |= options;
    }

    fn is_full(&self) -> bool {
        // Note: |options| < |cells| can only be the case if the Sudoku is
        // impossible.
        // TODO add a shortcut for returning impossible if a tuple with too
        // many cells is detected

        let options_len = self.options.len();
        options_len >= 2 && options_len <= self.cells.len()
    }
}

fn find_tuples_rec(sudoku_info: &SudokuInfo<impl Constraint + Clone>,
        group_rest: &[(usize, usize)], max_size: usize, mut curr_tuple: Tuple,
        accumulator: &mut Vec<Tuple>) {
    if curr_tuple.options.len() > max_size {
        return;
    }

    if curr_tuple.is_full() {
        accumulator.push(curr_tuple);
        return;
    }

    if let Some((next_column, next_row)) = group_rest.iter().cloned().next() {
        let next_options =
            sudoku_info.get_options(next_column, next_row).unwrap();
        let next_rest = &group_rest[1..];

        if next_options.len() > 1 {
            find_tuples_rec(sudoku_info, next_rest, max_size,
                curr_tuple.clone(), accumulator);
            curr_tuple.add_cell(next_options, next_column, next_row);
            find_tuples_rec(sudoku_info, next_rest, max_size, curr_tuple,
                accumulator);
        }
        else {
            find_tuples_rec(sudoku_info, next_rest, max_size, curr_tuple,
                accumulator);
        }
    }
}

fn find_tuples(sudoku_info: &SudokuInfo<impl Constraint + Clone>,
        group: &[(usize, usize)], max_size: usize) -> Vec<Tuple> {
    let mut result = Vec::new();
    find_tuples_rec(&sudoku_info, group, max_size,
        Tuple::new(sudoku_info.size()), &mut result);
    result
}

impl<F: Fn(usize) -> usize> Strategy for TupleStrategy<F> {

    fn apply(&self, sudoku_info: &mut SudokuInfo<impl Constraint + Clone>)
            -> bool {
        let mut changed = false;
        let grid = sudoku_info.sudoku().grid();
        let groups = sudoku_info.sudoku().constraint().get_groups(grid);
        let max_size = (self.max_size_computer)(sudoku_info.size());

        for group in groups {
            let tuples = find_tuples(&sudoku_info, &group, max_size);
            
            for tuple in tuples {
                for (column, row) in group.iter().cloned() {
                    if sudoku_info.get_cell(column, row).unwrap() == None &&
                            !tuple.cells.contains(&(column, row)) {
                        let mut cell_options =
                            sudoku_info.get_options_mut(column, row).unwrap();
                        let before_len = cell_options.len();
                        cell_options -= &tuple.options;
                        changed |= before_len != cell_options.len();
                    }
                }
            }
        }

        changed
    }
}

/// A [Strategy] which looks for cells with few options (up to a specified
/// maximum) and tries all of them. It then uses a wrapped strategy to find
/// deductions in all paths. If any of those deductions hold for all options,
/// they are stored in the metadata.
///
/// As an example, consider the following situation.
///
/// ```text
/// ╔═══╤═══╤═══╦═══╤═══╤═══╦═══╤═══╤═══╗
/// ║ 1 │ A │ 2 ║ 3 │ 4 │ 5 ║ 6 │ B │ 7 ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║   │   │   ║   │   │   ║   │   │   ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║   │   │   ║   │   │   ║   │   │   ║
/// ╠═══╪═══╪═══╬═══╪═══╪═══╬═══╪═══╪═══╣
/// ║ 2 │ 3 │ C ║   │   │   ║   │ Z │   ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║ 4 │   │ 1 ║   │   │   ║   │   │   ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║ 5 │ 6 │ 7 ║   │   │   ║   │   │   ║
/// ╠═══╪═══╪═══╬═══╪═══╪═══╬═══╪═══╪═══╣
/// ║   │   │   ║   │   │   ║   │   │   ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║   │   │   ║   │   │   ║   │   │   ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║   │   │   ║   │   │   ║   │   │   ║
/// ╚═══╧═══╧═══╩═══╧═══╧═══╩═══╧═══╧═══╝
/// ```
///
/// In this case, if A is an 8, then B must be a 9 by the [OnlyCellStrategy],
/// so Z cannot be a 9. If A is a 9, then C must be a 9 by the same strategy,
/// and consequently Z cannot be a 9 aswell. So, this strategy with an options
/// bound of at least 2 (since A can be 8 or 9), an [OnlyCellStrategy] as the
/// continuation strategy, and at least 1 application, can conclude that Z
/// cannot be a 9.
#[derive(Clone)]
pub struct BoundedOptionsBacktrackingStrategy<FO, FA, S>
where
    FO: Fn(usize) -> usize,
    FA: Fn(usize) -> Option<usize>,
    S: Strategy
{
    max_options_computer: FO,
    max_applications_computer: FA,
    continuation_strategy: S
}

impl<FO, FA, S> BoundedOptionsBacktrackingStrategy<FO, FA, S>
where
    FO: Fn(usize) -> usize,
    FA: Fn(usize) -> Option<usize>,
    S: Strategy
{

    /// Creates a new bounded options backtracking strategy.
    ///
    /// # Arguments
    ///
    /// * `max_options_computer`: A closure that, given the grid size, computes
    /// the maximum number of options that may be present in a cell for this
    /// strategy to consider all of them.
    /// * `max_applications_computer`: A closure that, given the grid size,
    /// computes the maximum number of times the continuation strategy may be
    /// applied for each considered option before no further inference is done.
    /// If no limit is desired, this may return `None`.
    /// * `continuation_strategy`: The [Strategy] with which each considered
    /// option is developed to find any inferences.
    pub fn new(max_options_computer: FO, max_applications_computer: FA,
            continuation_strategy: S)
            -> BoundedOptionsBacktrackingStrategy<FO, FA, S> {
        BoundedOptionsBacktrackingStrategy {
            max_options_computer,
            max_applications_computer,
            continuation_strategy
        }
    }
}

/// Applies `continuation_strategy` to `sudoku_info` until the fixed point is
/// reached, but at most `max_applications` times, if it is given.
fn apply_continuation(max_applications: Option<usize>,
        continuation_strategy: &impl Strategy,
        sudoku_info: &mut SudokuInfo<impl Constraint + Clone>) {
    match max_applications {
        None => {
            while continuation_strategy.apply(sudoku_info) { }
        },
        Some(max) => {
            for _ in 0..max {
                if !continuation_strategy.apply(sudoku_info) {
                    break;
                }
            }
        }
    }
}

/// Makes deductions in `sudoku_info` under the assumption that one of the
/// sudoku infos in `results` has to be true - i.e. it contains a complete case
/// distinction. This function returns `true` if `sudoku_info` changed.
fn collapse_results<C: Constraint + Clone>(sudoku_info: &mut SudokuInfo<C>,
        results: Vec<SudokuInfo<C>>) -> bool {
    if results.is_empty() {
        return false;
    }

    let mut results_iter = results.into_iter();
    let first = results_iter.next().unwrap();
    let union = results_iter.fold(first,
        |mut acc, x| {
            acc.union_assign(&x).unwrap();
            acc
        });
    sudoku_info.intersect_assign(&union).unwrap()
}

impl<FO, FA, S> Strategy for BoundedOptionsBacktrackingStrategy<FO, FA, S>
where
    FO: Fn(usize) -> usize,
    FA: Fn(usize) -> Option<usize>,
    S: Strategy
{
    fn apply(&self, sudoku_info: &mut SudokuInfo<impl Constraint + Clone>)
            -> bool {
        let size = sudoku_info.size();
        let max_options = (self.max_options_computer)(size);
        let max_applications = (self.max_applications_computer)(size);
        let mut changed = false;

        for column in 0..size {
            for row in 0..size {
                if sudoku_info.get_cell(column, row).unwrap().is_some() {
                    continue;
                }

                let options = sudoku_info.get_options(column, row).unwrap();

                if options.len() > max_options {
                    continue;
                }

                let mut results = Vec::new();

                for option in options.iter() {
                    let mut sudoku_info = sudoku_info.clone();
                    sudoku_info.enter_cell(column, row, option).unwrap();
                    apply_continuation(max_applications,
                        &self.continuation_strategy, &mut sudoku_info);
                    results.push(sudoku_info);
                }

                changed |= collapse_results(sudoku_info, results);
            }
        }

        changed
    }
}

/// A [Strategy] which looks for groups in which some number can only occur in
/// a limited number of cells (up to a specified maximum) and tries all of
/// them. It then uses a wrapped strategy to find deductions in all paths. If
/// any of those deductions hold for all options, they are stored in the
/// metadata.
///
/// As an example, consider the following situation:
///
/// ```text
/// ╔═══╤═══╤═══╦═══╤═══╤═══╦═══╤═══╤═══╗
/// ║   │   │   ║   │ 5 │   ║   │   │   ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║ 1 │ 2 │ 3 ║   │   │   ║   │   │   ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║ 4 │   │   ║   │   │   ║ X │   │   ║
/// ╠═══╪═══╪═══╬═══╪═══╪═══╬═══╪═══╪═══╣
/// ║ 2 │   │   ║   │   │   ║ X │   │   ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║   │   │   ║   │   │ 5 ║   │   │   ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║ 3 │ 1 │ 4 ║   │   │   ║   │   │   ║
/// ╠═══╪═══╪═══╬═══╪═══╪═══╬═══╪═══╪═══╣
/// ║   │ Y │ Y ║   │   │   ║   │   │   ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║   │ Y │ Y ║   │   │   ║   │   │   ║
/// ╟───┼───┼───╫───┼───┼───╫───┼───┼───╢
/// ║   │ Y │ Y ║   │   │   ║   │   │   ║
/// ╚═══╧═══╧═══╩═══╧═══╧═══╩═══╧═══╧═══╝
/// ```
///
/// In this configuration, a bounded cells backtracking strategy without any
/// wrapped stratey (i.e. a [NoStrategy]) with a maximum of 2 cells to consider
/// would find that the cells marked with X cannot be a 5. This is because both
/// in the top-left and the top-central box, there are two places for 5s each,
/// both in the same rows, thus excluding a 5 from the X-cells. Furthermore, if
/// an [OnlyCellStrategy] with at least 1 application is used as the
/// continuation strategy, the bounded cells backtracking strategy would be
/// able to deduce that fives must always be in columns 2 and 3 in the top-left
/// and top-central box and thus all cells marked with Y cannot be a 5.
///
/// Note that this strategy contains some common strategies with different
/// names. For example a bounded cells backtracking strategy with a limit of 2
/// cells and an [OnlyCellStrategy] with 1 application would find X-Wings.
#[derive(Clone)]
pub struct BoundedCellsBacktrackingStrategy<FC, FA, S>
where
    FC: Fn(usize) -> usize,
    FA: Fn(usize) -> Option<usize>,
    S: Strategy
{
    max_cells_computer: FC,
    max_applications_computer: FA,
    continuation_strategy: S
}

impl<FC, FA, S> BoundedCellsBacktrackingStrategy<FC, FA, S>
where
    FC: Fn(usize) -> usize,
    FA: Fn(usize) -> Option<usize>,
    S: Strategy
{

    /// Creates a new bounded cells backtracking strategy.
    ///
    /// # Arguments
    ///
    /// * `max_cells_computer`: A closure that, given the grid size, computes
    /// the maximum number of cells in a group in which a number can be for
    /// this strategy to consider all of them. 
    /// * `max_applications_computer`: A closure that, given the grid size,
    /// computes the maximum number of times the continuation strategy may be
    /// applied for each considered cell before no further inference is done.
    /// If no limit is desired, this may return `None`.
    /// * `continuation_strategy`: The [Strategy] with which each considered
    /// cell is developed to find any inferences.
    pub fn new(max_cells_computer: FC, max_applications_computer: FA,
            continuation_strategy: S)
            -> BoundedCellsBacktrackingStrategy<FC, FA, S> {
        BoundedCellsBacktrackingStrategy {
            max_cells_computer,
            max_applications_computer,
            continuation_strategy
        }
    }
}

impl<FC, FA, S> Strategy for BoundedCellsBacktrackingStrategy<FC, FA, S>
where
    FC: Fn(usize) -> usize,
    FA: Fn(usize) -> Option<usize>,
    S: Strategy
{
    fn apply(&self, sudoku_info: &mut SudokuInfo<impl Constraint + Clone>)
            -> bool {
        let size = sudoku_info.size();
        let max_cells = (self.max_cells_computer)(size);
        let max_applications = (self.max_applications_computer)(size);
        let mut changed = false;
        let grid = sudoku_info.sudoku().grid();
        let groups = sudoku_info.sudoku().constraint().get_groups(grid);

        for group in groups {
            // We must assume that the group contains all numbers for the case
            // distinction to be valid.
            if group.len() < size {
                continue;
            }

            let mut number_locations: Vec<Vec<(usize, usize)>> =
                vec![Vec::new(); size + 1];

            for (column, row) in group {
                if sudoku_info.get_cell(column, row).unwrap().is_some() {
                    continue;
                }

                let options = sudoku_info.get_options(column, row).unwrap();

                for option in options.iter() {
                    number_locations[option].push((column, row));
                }
            }

            let number_locations_iter = number_locations.into_iter().skip(1);

            for (number, locations) in number_locations_iter.enumerate() {
                let number = number + 1;

                if locations.is_empty() || locations.len() > max_cells {
                    continue;
                }

                let mut results = Vec::new();

                for (column, row) in locations {
                    let mut sudoku_info = sudoku_info.clone();
                    sudoku_info.enter_cell(column, row, number).unwrap();
                    apply_continuation(max_applications,
                        &self.continuation_strategy, &mut sudoku_info);
                    results.push(sudoku_info);
                }

                changed |= collapse_results(sudoku_info, results);
            }
        }

        changed
    }
}

/// A [Strategy] which does nothing. This is to be used in backtracking
/// strategies to define that no further logic shall be applied after trying an
/// option.
#[derive(Clone)]
pub struct NoStrategy;

impl Strategy for NoStrategy {
    fn apply(&self, _: &mut SudokuInfo<impl Constraint + Clone>) -> bool {
        false
    }
}

/// A [Strategy] which uses two strategies by first applying one and then the
/// other on the output of the first one. If any child changed the state, this
/// strategy is defined to have changed the state aswell.
pub struct CompositeStrategy<S1: Strategy, S2: Strategy> {
    s1: S1,
    s2: S2
}

impl<S1: Strategy, S2: Strategy> CompositeStrategy<S1, S2> {

    /// Creates a new composite strategy from the two children strategies.
    ///
    /// # Arguments
    ///
    /// * `s1`: The strategy which is applied first.
    /// * `s2`: The strategy which is applied second.
    pub fn new(s1: S1, s2: S2) -> CompositeStrategy<S1, S2> {
        CompositeStrategy {
            s1,
            s2
        }
    }
}

impl<S1: Strategy, S2: Strategy> Strategy for CompositeStrategy<S1, S2> {
    fn apply(&self, sudoku_info: &mut SudokuInfo<impl Constraint + Clone>)
            -> bool {
        self.s1.apply(sudoku_info) | self.s2.apply(sudoku_info)
    }
}

impl<S1: Strategy + Clone, S2: Strategy + Clone> Clone for CompositeStrategy<S1, S2> {
    fn clone(&self) -> Self {
        CompositeStrategy::new(self.s1.clone(), self.s2.clone())
    }
}

#[cfg(test)]
mod tests {

    use super::*;

    use crate::Sudoku;
    use crate::constraint::DefaultConstraint;
    use crate::solver::strategy::SudokuInfo;

    fn apply<C: Constraint + Clone, S: Strategy>(strategy: S,
            sudoku_info: &mut SudokuInfo<C>, apply_once: bool) {
        while strategy.apply(sudoku_info) {
            if apply_once {
                break;
            }
        }
    }

    /// Tests that the `weak_strategy` cannot find deductions that make `test`
    /// true, but `strong_strategy` can. Also asserts that the resulting Sudoku
    /// is correct.
    fn test_strategy_stronger_and_sound<C, W, S>(sudoku: Sudoku<C>,
        weak_strategy: W, strong_strategy: S, apply_once: bool,
        test: impl Fn(&SudokuInfo<C>) -> bool)
    where
        C: Constraint + Clone,
        W: Strategy,
        S: Strategy
    {
        let mut sudoku_info = SudokuInfo::from_sudoku(sudoku.clone());
        apply(weak_strategy, &mut sudoku_info, apply_once);
        assert!(!test(&sudoku_info));

        let mut sudoku_info = SudokuInfo::from_sudoku(sudoku);
        apply(strong_strategy, &mut sudoku_info, apply_once);
        assert!(test(&sudoku_info));

        assert!(sudoku_info.sudoku().is_valid());
    }

    #[test]
    fn naked_single_strategy_finds_digit() {
        let sudoku = Sudoku::parse("2x2;\
            1, , , ,\
             , , ,2,\
             , , , ,\
             ,3, , ", DefaultConstraint).unwrap();
        let mut sudoku_info = SudokuInfo::from_sudoku(sudoku);
        
        assert!(NakedSingleStrategy.apply(&mut sudoku_info));
        assert_eq!(Some(4), sudoku_info.get_cell(1, 1).unwrap());
    }

    #[test]
    fn only_cell_strategy_finds_digits() {
        let sudoku = Sudoku::parse("2x2;\
            1, , , ,\
             , , ,2,\
             , , , ,\
             ,3, , ", DefaultConstraint).unwrap();
        let mut sudoku_info = SudokuInfo::from_sudoku(sudoku);
        
        assert!(OnlyCellStrategy.apply(&mut sudoku_info));
        assert_eq!(Some(1), sudoku_info.get_cell(2, 1).unwrap());
        assert_eq!(Some(1), sudoku_info.get_cell(1, 2).unwrap());
        assert_eq!(Some(2), sudoku_info.get_cell(1, 0).unwrap());
        assert_eq!(Some(3), sudoku_info.get_cell(0, 1).unwrap());
    }

    #[test]
    fn tuple_strategy_helps_naked_single_strategy() {
        // In this Sudoku, the cell in column 2, row 2 must be a 6, but that
        // can only be recognized once the options 1, 2, 7, and 8 have been
        // excluded by the tuple strategy.
        // Only tuples of size 2 need to be considered.

        let sudoku = Sudoku::parse("3x3;\
             , ,3,4,5,6,7,8,9,\
             , ,9,1,2,3,4,5,6,\
             , , , , , , , , ,\
             , ,4, , , , , , ,\
             , ,5, , , , , , ,\
             , , , , , , , , ,\
             , , , , , , , , ,\
             , , , , , , , , ,\
             , , , , , , , , ", DefaultConstraint).unwrap();
        let weak_strategy = NakedSingleStrategy;
        let strong_strategy = CompositeStrategy::new(
            TupleStrategy::new(|_| 2), NakedSingleStrategy);
        
        test_strategy_stronger_and_sound(sudoku, weak_strategy,
            strong_strategy, false, |s| s.get_cell(2, 2).unwrap() == Some(6));
    }

    #[test]
    fn tuple_strategy_does_not_consider_too_large_tuples() {
        // This Sudoku is equivalent to the one above, but missing the 3 and 9
        // in column 2. This means that tuples of size 3 are necessary to
        // deduce the 6.

        let sudoku = Sudoku::parse("3x3;\
             , , ,4,5,6,7,8,9,\
             , , ,1,2,3,4,5,6,\
             , , , , , , , , ,\
             , ,4, , , , , , ,\
             , ,5, , , , , , ,\
             , , , , , , , , ,\
             , , , , , , , , ,\
             , , , , , , , , ,\
             , , , , , , , , ", DefaultConstraint).unwrap();
        let weak_strategy = CompositeStrategy::new(
            TupleStrategy::new(|_| 2), NakedSingleStrategy);
        let strong_strategy = CompositeStrategy::new(
            TupleStrategy::new(|_| 3), NakedSingleStrategy);
        
        test_strategy_stronger_and_sound(sudoku, weak_strategy,
            strong_strategy, false, |s| s.get_cell(2, 2).unwrap() == Some(6));
    }

    #[test]
    fn bounded_options_backtracking_deduces_impossible_option() {
        let sudoku = Sudoku::parse("3x3;\
            1, ,2,3,4,5,6, ,7,\
             , , , , , , , , ,\
             , , , , , , , , ,\
            2,3, , , , , , , ,\
            4, ,1, , , , , , ,\
            5,6,7, , , , , , ,\
             , , , , , , , , ,\
             , , , , , , , , ,\
             , , , , , , , , ", DefaultConstraint).unwrap();
        let strategy =
            BoundedOptionsBacktrackingStrategy::new(|_| 2, |_| Some(1),
                OnlyCellStrategy);
        let mut sudoku_info = SudokuInfo::from_sudoku(sudoku);
        
        assert!(strategy.apply(&mut sudoku_info));
        assert!(!sudoku_info.get_options(7, 3).unwrap().contains(8));
        assert!(!sudoku_info.get_options(7, 3).unwrap().contains(9));
    }

    fn has_option<C: Constraint + Clone>(sudoku_info: &SudokuInfo<C>,
            column: usize, row: usize, option: usize) -> bool {
        sudoku_info.get_options(column, row).unwrap().contains(option)
    }

    #[test]
    fn bounded_options_backtracking_respects_application_limit() {
        let sudoku = Sudoku::parse("3x3;\
            1, ,2,3,4,5,6, ,7,\
             , , , , , , , , ,\
             , , , , , , , , ,\
            2,1, , , , , , , ,\
            3, ,6, , , , , , ,\
            4,5,7, , , , , , ,\
             , , , , , , , , ,\
             , , , , , , , , ,\
             , , , , , , , , ", DefaultConstraint).unwrap();
        let weak_strategy =
            BoundedOptionsBacktrackingStrategy::new(|_| 2, |_| Some(1),
                NakedSingleStrategy);
        let strong_strategy =
            BoundedOptionsBacktrackingStrategy::new(|_| 2, |_| Some(2),
                NakedSingleStrategy);
        test_strategy_stronger_and_sound(sudoku, weak_strategy,
            strong_strategy, true, |s|
                !has_option(s, 7, 3, 9) && !has_option(s, 7, 3, 9))
    }

    #[test]
    fn bounded_options_backtracking_respects_option_limit() {
        let sudoku = Sudoku::parse("3x3;\
            1, ,2,3, ,4,5, ,6,\
             , , , , , , , , ,\
             , , ,7, , ,8, , ,\
            2, , ,1, , , , , ,\
            3, ,1,4, ,5, , , ,\
            4, ,5,2, ,3, , , ,\
             , , , , , , , , ,\
             , , , , , , , , ,\
             , , , , , , , , ", DefaultConstraint).unwrap();
        let weak_strategy =
            BoundedOptionsBacktrackingStrategy::new(|_| 2, |_| None,
                OnlyCellStrategy);
        let strong_strategy =
            BoundedOptionsBacktrackingStrategy::new(|_| 3, |_| None,
                OnlyCellStrategy);
        test_strategy_stronger_and_sound(sudoku, weak_strategy,
            strong_strategy, true, |s| !has_option(s, 7, 3, 9));
    }

    #[test]
    fn bounded_cells_backtracking_detects_impossible_option() {
        let sudoku = Sudoku::parse("3x3;\
             , , , ,5, , , , ,\
            1,2,3, , , , , , ,\
            4, , , , , , , , ,\
            2, , , , , , , , ,\
            3,1,4, , , , , , ,\
             , , , , ,5, , , ,\
             , , , , , , , , ,\
             , , , , , , , , ,\
             , , , , , , , , ", DefaultConstraint).unwrap();
        let strategy =
            BoundedCellsBacktrackingStrategy::new(|_| 2, |_| Some(0),
                NoStrategy);
        let mut sudoku_info = SudokuInfo::from_sudoku(sudoku);

        assert!(strategy.apply(&mut sudoku_info));
        assert!(!sudoku_info.get_options(6, 2).unwrap().contains(5));
        assert!(!sudoku_info.get_options(6, 3).unwrap().contains(5));
    }

    #[test]
    fn bounded_cells_backtracking_respects_application_limit() {
        let sudoku = Sudoku::parse("3x3;\
             , , , ,5, , , , ,\
            1,2,3, , , , , , ,\
            4, , , , , , , , ,\
            2, , , , , , , , ,\
            3,1,4, , , , , , ,\
             , , , , ,5, , , ,\
             , , , , , , , , ,\
             , , , , , , , , ,\
             , , , , , , , , ", DefaultConstraint).unwrap();
        let weak_strategy =
            BoundedCellsBacktrackingStrategy::new(|_| 2, |_| Some(0),
                OnlyCellStrategy);
        let strong_strategy =
            BoundedCellsBacktrackingStrategy::new(|_| 2, |_| Some(1),
                OnlyCellStrategy);
        test_strategy_stronger_and_sound(sudoku, weak_strategy,
            strong_strategy, true, |s|
                !has_option(s, 1, 6, 5) && !has_option(s, 2, 6, 5));
    }

    #[test]
    fn bounded_cells_backtracking_respects_cell_limit() {
        let sudoku = Sudoku::parse("3x3;\
             , , , ,5, , , , ,\
            1,2,3, , , , , , ,\
            4, , , , , , , , ,\
            2,1, , , , , , , ,\
            3, , , , , , , , ,\
             , , , , ,5, , , ,\
             , , , , , , , , ,\
             , , , , , , , , ,\
             , , , , , , , , ", DefaultConstraint).unwrap();
        let weak_strategy =
            BoundedCellsBacktrackingStrategy::new(|_| 2, |_| Some(1),
                OnlyCellStrategy);
        let strong_strategy =
            BoundedCellsBacktrackingStrategy::new(|_| 3, |_| Some(1),
                OnlyCellStrategy);
        test_strategy_stronger_and_sound(sudoku, weak_strategy,
            strong_strategy, true, |s| !has_option(s, 2, 6, 5));
    }
}