helix/helix-core/src/selection.rs

//! Selections are the primary editing construct. Even cursors are
//! defined as a selection range.
//!
//! All positioning is done via `char` offsets into the buffer.
use crate::{
    graphemes::{
        ensure_grapheme_boundary_next, ensure_grapheme_boundary_prev, next_grapheme_boundary,
        prev_grapheme_boundary,
    },
    line_ending::get_line_ending,
    movement::Direction,
    Assoc, ChangeSet, RopeGraphemes, RopeSlice,
};
use helix_stdx::rope::{self, RopeSliceExt};
use smallvec::{smallvec, SmallVec};
use std::{borrow::Cow, iter, slice};
use tree_sitter::Node;

/// A single selection range.
///
/// A range consists of an "anchor" and "head" position in
/// the text.  The head is the part that the user moves when
/// directly extending a selection.  The head and anchor
/// can be in any order, or even share the same position.
///
/// The anchor and head positions use gap indexing, meaning
/// that their indices represent the gaps *between* `char`s
/// rather than the `char`s themselves. For example, 1
/// represents the position between the first and second `char`.
///
/// Below are some examples of `Range` configurations.
/// The anchor and head indices are shown as "(anchor, head)"
/// tuples, followed by example text with "[" and "]" symbols
/// representing the anchor and head positions:
///
/// - (0, 3): `[Som]e text`.
/// - (3, 0): `]Som[e text`.
/// - (2, 7): `So[me te]xt`.
/// - (1, 1): `S[]ome text`.
///
/// Ranges are considered to be inclusive on the left and
/// exclusive on the right, regardless of anchor-head ordering.
/// This means, for example, that non-zero-width ranges that
/// are directly adjacent, sharing an edge, do not overlap.
/// However, a zero-width range will overlap with the shared
/// left-edge of another range.
///
/// By convention, user-facing ranges are considered to have
/// a block cursor on the head-side of the range that spans a
/// single grapheme inward from the range's edge.  There are a
/// variety of helper methods on `Range` for working in terms of
/// that block cursor, all of which have `cursor` in their name.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub struct Range {
    /// The anchor of the range: the side that doesn't move when extending.
    pub anchor: usize,
    /// The head of the range, moved when extending.
    pub head: usize,
    /// The previous visual offset (softwrapped lines and columns) from
    /// the start of the line
    pub old_visual_position: Option<(u32, u32)>,
}

impl Range {
    pub fn new(anchor: usize, head: usize) -> Self {
        Self {
            anchor,
            head,
            old_visual_position: None,
        }
    }

    pub fn point(head: usize) -> Self {
        Self::new(head, head)
    }

    pub fn from_node(node: Node, text: RopeSlice, direction: Direction) -> Self {
        let from = text.byte_to_char(node.start_byte());
        let to = text.byte_to_char(node.end_byte());
        Range::new(from, to).with_direction(direction)
    }

    /// Start of the range.
    #[inline]
    #[must_use]
    pub fn from(&self) -> usize {
        std::cmp::min(self.anchor, self.head)
    }

    /// End of the range.
    #[inline]
    #[must_use]
    pub fn to(&self) -> usize {
        std::cmp::max(self.anchor, self.head)
    }

    /// Total length of the range.
    #[inline]
    #[must_use]
    pub fn len(&self) -> usize {
        self.to() - self.from()
    }

    /// The (inclusive) range of lines that the range overlaps.
    #[inline]
    #[must_use]
    pub fn line_range(&self, text: RopeSlice) -> (usize, usize) {
        let from = self.from();
        let to = if self.is_empty() {
            self.to()
        } else {
            prev_grapheme_boundary(text, self.to()).max(from)
        };

        (text.char_to_line(from), text.char_to_line(to))
    }

    /// `true` when head and anchor are at the same position.
    #[inline]
    pub fn is_empty(&self) -> bool {
        self.anchor == self.head
    }

    /// `Direction::Backward` when head < anchor.
    /// `Direction::Forward` otherwise.
    #[inline]
    #[must_use]
    pub fn direction(&self) -> Direction {
        if self.head < self.anchor {
            Direction::Backward
        } else {
            Direction::Forward
        }
    }

    /// Flips the direction of the selection
    pub fn flip(&self) -> Self {
        Self {
            anchor: self.head,
            head: self.anchor,
            old_visual_position: self.old_visual_position,
        }
    }

    /// Returns the selection if it goes in the direction of `direction`,
    /// flipping the selection otherwise.
    pub fn with_direction(self, direction: Direction) -> Self {
        if self.direction() == direction {
            self
        } else {
            self.flip()
        }
    }

    /// Check two ranges for overlap.
    #[must_use]
    pub fn overlaps(&self, other: &Self) -> bool {
        // To my eye, it's non-obvious why this works, but I arrived
        // at it after transforming the slower version that explicitly
        // enumerated more cases.  The unit tests are thorough.
        self.from() == other.from() || (self.to() > other.from() && other.to() > self.from())
    }

    #[inline]
    pub fn contains_range(&self, other: &Self) -> bool {
        self.from() <= other.from() && self.to() >= other.to()
    }

    pub fn contains(&self, pos: usize) -> bool {
        self.from() <= pos && pos < self.to()
    }

    /// Map a range through a set of changes. Returns a new range representing
    /// the same position after the changes are applied. Note that this
    /// function runs in O(N) (N is number of changes) and can therefore
    /// cause performance problems if run for a large number of ranges as the
    /// complexity is then O(MN) (for multicuror M=N usually). Instead use
    /// [Selection::map] or [ChangeSet::update_positions].
    pub fn map(mut self, changes: &ChangeSet) -> Self {
        use std::cmp::Ordering;
        if changes.is_empty() {
            return self;
        }

        let positions_to_map = match self.anchor.cmp(&self.head) {
            Ordering::Equal => [
                (&mut self.anchor, Assoc::AfterSticky),
                (&mut self.head, Assoc::AfterSticky),
            ],
            Ordering::Less => [
                (&mut self.anchor, Assoc::AfterSticky),
                (&mut self.head, Assoc::BeforeSticky),
            ],
            Ordering::Greater => [
                (&mut self.head, Assoc::AfterSticky),
                (&mut self.anchor, Assoc::BeforeSticky),
            ],
        };
        changes.update_positions(positions_to_map.into_iter());
        self.old_visual_position = None;
        self
    }

    /// Extend the range to cover at least `from` `to`.
    #[must_use]
    pub fn extend(&self, from: usize, to: usize) -> Self {
        debug_assert!(from <= to);

        if self.anchor <= self.head {
            Self {
                anchor: self.anchor.min(from),
                head: self.head.max(to),
                old_visual_position: None,
            }
        } else {
            Self {
                anchor: self.anchor.max(to),
                head: self.head.min(from),
                old_visual_position: None,
            }
        }
    }

    /// Returns a range that encompasses both input ranges.
    ///
    /// This is like `extend()`, but tries to negotiate the
    /// anchor/head ordering between the two input ranges.
    #[must_use]
    pub fn merge(&self, other: Self) -> Self {
        if self.anchor > self.head && other.anchor > other.head {
            Range {
                anchor: self.anchor.max(other.anchor),
                head: self.head.min(other.head),
                old_visual_position: None,
            }
        } else {
            Range {
                anchor: self.from().min(other.from()),
                head: self.to().max(other.to()),
                old_visual_position: None,
            }
        }
    }

    // groupAt

    /// Returns the text inside this range given the text of the whole buffer.
    ///
    /// The returned `Cow` is a reference if the range of text is inside a single
    /// chunk of the rope. Otherwise a copy of the text is returned. Consider
    /// using `slice` instead if you do not need a `Cow` or `String` to avoid copying.
    #[inline]
    pub fn fragment<'a, 'b: 'a>(&'a self, text: RopeSlice<'b>) -> Cow<'b, str> {
        self.slice(text).into()
    }

    /// Returns the text inside this range given the text of the whole buffer.
    ///
    /// The returned value is a reference to the passed slice. This method never
    /// copies any contents.
    #[inline]
    pub fn slice<'a, 'b: 'a>(&'a self, text: RopeSlice<'b>) -> RopeSlice<'b> {
        text.slice(self.from()..self.to())
    }

    //--------------------------------
    // Alignment methods.

    /// Compute a possibly new range from this range, with its ends
    /// shifted as needed to align with grapheme boundaries.
    ///
    /// Zero-width ranges will always stay zero-width, and non-zero-width
    /// ranges will never collapse to zero-width.
    #[must_use]
    pub fn grapheme_aligned(&self, slice: RopeSlice) -> Self {
        use std::cmp::Ordering;
        let (new_anchor, new_head) = match self.anchor.cmp(&self.head) {
            Ordering::Equal => {
                let pos = ensure_grapheme_boundary_prev(slice, self.anchor);
                (pos, pos)
            }
            Ordering::Less => (
                ensure_grapheme_boundary_prev(slice, self.anchor),
                ensure_grapheme_boundary_next(slice, self.head),
            ),
            Ordering::Greater => (
                ensure_grapheme_boundary_next(slice, self.anchor),
                ensure_grapheme_boundary_prev(slice, self.head),
            ),
        };
        Range {
            anchor: new_anchor,
            head: new_head,
            old_visual_position: if new_anchor == self.anchor {
                self.old_visual_position
            } else {
                None
            },
        }
    }

    /// Compute a possibly new range from this range, attempting to ensure
    /// a minimum range width of 1 char by shifting the head in the forward
    /// direction as needed.
    ///
    /// This method will never shift the anchor, and will only shift the
    /// head in the forward direction.  Therefore, this method can fail
    /// at ensuring the minimum width if and only if the passed range is
    /// both zero-width and at the end of the `RopeSlice`.
    ///
    /// If the input range is grapheme-boundary aligned, the returned range
    /// will also be.  Specifically, if the head needs to shift to achieve
    /// the minimum width, it will shift to the next grapheme boundary.
    #[must_use]
    #[inline]
    pub fn min_width_1(&self, slice: RopeSlice) -> Self {
        if self.anchor == self.head {
            Range {
                anchor: self.anchor,
                head: next_grapheme_boundary(slice, self.head),
                old_visual_position: self.old_visual_position,
            }
        } else {
            *self
        }
    }

    //--------------------------------
    // Block-cursor methods.

    /// Gets the left-side position of the block cursor.
    #[must_use]
    #[inline]
    pub fn cursor(self, text: RopeSlice) -> usize {
        if self.head > self.anchor {
            prev_grapheme_boundary(text, self.head)
        } else {
            self.head
        }
    }

    /// Puts the left side of the block cursor at `char_idx`, optionally extending.
    ///
    /// This follows "1-width" semantics, and therefore does a combination of anchor
    /// and head moves to behave as if both the front and back of the range are 1-width
    /// blocks
    ///
    /// This method assumes that the range and `char_idx` are already properly
    /// grapheme-aligned.
    #[must_use]
    #[inline]
    pub fn put_cursor(self, text: RopeSlice, char_idx: usize, extend: bool) -> Range {
        if extend {
            let anchor = if self.head >= self.anchor && char_idx < self.anchor {
                next_grapheme_boundary(text, self.anchor)
            } else if self.head < self.anchor && char_idx >= self.anchor {
                prev_grapheme_boundary(text, self.anchor)
            } else {
                self.anchor
            };

            if anchor <= char_idx {
                Range::new(anchor, next_grapheme_boundary(text, char_idx))
            } else {
                Range::new(anchor, char_idx)
            }
        } else {
            Range::point(char_idx)
        }
    }

    /// The line number that the block-cursor is on.
    #[inline]
    #[must_use]
    pub fn cursor_line(&self, text: RopeSlice) -> usize {
        text.char_to_line(self.cursor(text))
    }

    /// Returns true if this Range covers a single grapheme in the given text
    pub fn is_single_grapheme(&self, doc: RopeSlice) -> bool {
        let mut graphemes = RopeGraphemes::new(doc.slice(self.from()..self.to()));
        let first = graphemes.next();
        let second = graphemes.next();
        first.is_some() && second.is_none()
    }

    /// Converts this char range into an in order byte range, discarding
    /// direction.
    pub fn into_byte_range(&self, text: RopeSlice) -> (usize, usize) {
        (text.char_to_byte(self.from()), text.char_to_byte(self.to()))
    }
}

impl From<(usize, usize)> for Range {
    fn from((anchor, head): (usize, usize)) -> Self {
        Self {
            anchor,
            head,
            old_visual_position: None,
        }
    }
}

/// A selection consists of one or more selection ranges.
/// invariant: A selection can never be empty (always contains at least primary range).
#[derive(Debug, Clone, PartialEq, Eq)]
pub struct Selection {
    ranges: SmallVec<[Range; 1]>,
    primary_index: usize,
}

#[allow(clippy::len_without_is_empty)] // a Selection is never empty
impl Selection {
    // eq

    #[inline]
    #[must_use]
    pub fn primary(&self) -> Range {
        self.ranges[self.primary_index]
    }

    #[inline]
    #[must_use]
    pub fn primary_mut(&mut self) -> &mut Range {
        &mut self.ranges[self.primary_index]
    }

    /// Ensure selection containing only the primary selection.
    pub fn into_single(self) -> Self {
        if self.ranges.len() == 1 {
            self
        } else {
            Self {
                ranges: smallvec![self.ranges[self.primary_index]],
                primary_index: 0,
            }
        }
    }

    /// Adds a new range to the selection and makes it the primary range.
    pub fn push(mut self, range: Range) -> Self {
        self.ranges.push(range);
        self.set_primary_index(self.ranges().len() - 1);
        self.normalize()
    }

    /// Removes a range from the selection.
    pub fn remove(mut self, index: usize) -> Self {
        assert!(
            self.ranges.len() > 1,
            "can't remove the last range from a selection!"
        );

        self.ranges.remove(index);
        if index < self.primary_index || self.primary_index == self.ranges.len() {
            self.primary_index -= 1;
        }
        self
    }

    /// Replace a range in the selection with a new range.
    pub fn replace(mut self, index: usize, range: Range) -> Self {
        self.ranges[index] = range;
        self.normalize()
    }

    /// Map selections over a set of changes. Useful for adjusting the selection position after
    /// applying changes to a document.
    pub fn map(self, changes: &ChangeSet) -> Self {
        self.map_no_normalize(changes).normalize()
    }

    /// Map selections over a set of changes. Useful for adjusting the selection position after
    /// applying changes to a document. Doesn't normalize the selection
    pub fn map_no_normalize(mut self, changes: &ChangeSet) -> Self {
        if changes.is_empty() {
            return self;
        }

        let positions_to_map = self.ranges.iter_mut().flat_map(|range| {
            use std::cmp::Ordering;
            range.old_visual_position = None;
            match range.anchor.cmp(&range.head) {
                Ordering::Equal => [
                    (&mut range.anchor, Assoc::AfterSticky),
                    (&mut range.head, Assoc::AfterSticky),
                ],
                Ordering::Less => [
                    (&mut range.anchor, Assoc::AfterSticky),
                    (&mut range.head, Assoc::BeforeSticky),
                ],
                Ordering::Greater => [
                    (&mut range.head, Assoc::AfterSticky),
                    (&mut range.anchor, Assoc::BeforeSticky),
                ],
            }
        });
        changes.update_positions(positions_to_map);
        self
    }

    pub fn ranges(&self) -> &[Range] {
        &self.ranges
    }

    /// Returns an iterator over the line ranges of each range in the selection.
    ///
    /// Adjacent and overlapping line ranges of the [Range]s in the selection are merged.
    pub fn line_ranges<'a>(&'a self, text: RopeSlice<'a>) -> LineRangeIter<'a> {
        LineRangeIter {
            ranges: self.ranges.iter().peekable(),
            text,
        }
    }

    pub fn primary_index(&self) -> usize {
        self.primary_index
    }

    pub fn set_primary_index(&mut self, idx: usize) {
        assert!(idx < self.ranges.len());
        self.primary_index = idx;
    }

    #[must_use]
    /// Constructs a selection holding a single range.
    pub fn single(anchor: usize, head: usize) -> Self {
        Self {
            ranges: smallvec![Range {
                anchor,
                head,
                old_visual_position: None
            }],
            primary_index: 0,
        }
    }

    /// Constructs a selection holding a single cursor.
    pub fn point(pos: usize) -> Self {
        Self::single(pos, pos)
    }

    /// Normalizes a `Selection`.
    ///
    /// Ranges are sorted by [Range::from], with overlapping ranges merged.
    fn normalize(mut self) -> Self {
        if self.len() < 2 {
            return self;
        }
        let mut primary = self.ranges[self.primary_index];
        self.ranges.sort_unstable_by_key(Range::from);

        self.ranges.dedup_by(|curr_range, prev_range| {
            if prev_range.overlaps(curr_range) {
                let new_range = curr_range.merge(*prev_range);
                if prev_range == &primary || curr_range == &primary {
                    primary = new_range;
                }
                *prev_range = new_range;
                true
            } else {
                false
            }
        });

        self.primary_index = self
            .ranges
            .iter()
            .position(|&range| range == primary)
            .unwrap();

        self
    }

    /// Replaces ranges with one spanning from first to last range.
    pub fn merge_ranges(self) -> Self {
        let first = self.ranges.first().unwrap();
        let last = self.ranges.last().unwrap();
        Selection::new(smallvec![first.merge(*last)], 0)
    }

    /// Merges all ranges that are consecutive.
    pub fn merge_consecutive_ranges(mut self) -> Self {
        let mut primary = self.ranges[self.primary_index];

        self.ranges.dedup_by(|curr_range, prev_range| {
            if prev_range.to() == curr_range.from() {
                let new_range = curr_range.merge(*prev_range);
                if prev_range == &primary || curr_range == &primary {
                    primary = new_range;
                }
                *prev_range = new_range;
                true
            } else {
                false
            }
        });

        self.primary_index = self
            .ranges
            .iter()
            .position(|&range| range == primary)
            .unwrap();

        self
    }

    // TODO: consume an iterator or a vec to reduce allocations?
    #[must_use]
    pub fn new(ranges: SmallVec<[Range; 1]>, primary_index: usize) -> Self {
        assert!(!ranges.is_empty());
        debug_assert!(primary_index < ranges.len());

        let selection = Self {
            ranges,
            primary_index,
        };

        selection.normalize()
    }

    /// Takes a closure and maps each `Range` over the closure.
    pub fn transform<F>(mut self, mut f: F) -> Self
    where
        F: FnMut(Range) -> Range,
    {
        for range in self.ranges.iter_mut() {
            *range = f(*range)
        }
        self.normalize()
    }

    /// Takes a closure and maps each `Range` over the closure to multiple `Range`s.
    pub fn transform_iter<F, I>(mut self, f: F) -> Self
    where
        F: FnMut(Range) -> I,
        I: Iterator<Item = Range>,
    {
        self.ranges = self.ranges.into_iter().flat_map(f).collect();
        self.normalize()
    }

    // Ensures the selection adheres to the following invariants:
    // 1. All ranges are grapheme aligned.
    // 2. All ranges are at least 1 character wide, unless at the
    //    very end of the document.
    // 3. Ranges are non-overlapping.
    // 4. Ranges are sorted by their position in the text.
    pub fn ensure_invariants(self, text: RopeSlice) -> Self {
        self.transform(|r| r.min_width_1(text).grapheme_aligned(text))
            .normalize()
    }

    /// Transforms the selection into all of the left-side head positions,
    /// using block-cursor semantics.
    pub fn cursors(self, text: RopeSlice) -> Self {
        self.transform(|range| Range::point(range.cursor(text)))
    }

    pub fn fragments<'a>(
        &'a self,
        text: RopeSlice<'a>,
    ) -> impl DoubleEndedIterator<Item = Cow<'a, str>> + ExactSizeIterator<Item = Cow<str>> + 'a
    {
        self.ranges.iter().map(move |range| range.fragment(text))
    }

    pub fn slices<'a>(
        &'a self,
        text: RopeSlice<'a>,
    ) -> impl DoubleEndedIterator<Item = RopeSlice<'a>> + ExactSizeIterator<Item = RopeSlice<'a>> + 'a
    {
        self.ranges.iter().map(move |range| range.slice(text))
    }

    #[inline(always)]
    pub fn iter(&self) -> std::slice::Iter<'_, Range> {
        self.ranges.iter()
    }

    #[inline(always)]
    pub fn len(&self) -> usize {
        self.ranges.len()
    }

    // returns true if self ⊇ other
    pub fn contains(&self, other: &Selection) -> bool {
        let (mut iter_self, mut iter_other) = (self.iter(), other.iter());
        let (mut ele_self, mut ele_other) = (iter_self.next(), iter_other.next());

        loop {
            match (ele_self, ele_other) {
                (Some(ra), Some(rb)) => {
                    if !ra.contains_range(rb) {
                        // `self` doesn't contain next element from `other`, advance `self`, we need to match all from `other`
                        ele_self = iter_self.next();
                    } else {
                        // matched element from `other`, advance `other`
                        ele_other = iter_other.next();
                    };
                }
                (None, Some(_)) => {
                    // exhausted `self`, we can't match the reminder of `other`
                    return false;
                }
                (_, None) => {
                    // no elements from `other` left to match, `self` contains `other`
                    return true;
                }
            }
        }
    }
}

impl<'a> IntoIterator for &'a Selection {
    type Item = &'a Range;
    type IntoIter = std::slice::Iter<'a, Range>;

    fn into_iter(self) -> std::slice::Iter<'a, Range> {
        self.ranges().iter()
    }
}

impl IntoIterator for Selection {
    type Item = Range;
    type IntoIter = smallvec::IntoIter<[Range; 1]>;

    fn into_iter(self) -> smallvec::IntoIter<[Range; 1]> {
        self.ranges.into_iter()
    }
}

impl From<Range> for Selection {
    fn from(range: Range) -> Self {
        Self {
            ranges: smallvec![range],
            primary_index: 0,
        }
    }
}

pub struct LineRangeIter<'a> {
    ranges: iter::Peekable<slice::Iter<'a, Range>>,
    text: RopeSlice<'a>,
}

impl<'a> Iterator for LineRangeIter<'a> {
    type Item = (usize, usize);

    fn next(&mut self) -> Option<Self::Item> {
        let (start, mut end) = self.ranges.next()?.line_range(self.text);
        while let Some((next_start, next_end)) =
            self.ranges.peek().map(|range| range.line_range(self.text))
        {
            // Merge overlapping and adjacent ranges.
            // This subtraction cannot underflow because the ranges are sorted.
            if next_start - end <= 1 {
                end = next_end;
                self.ranges.next();
            } else {
                break;
            }
        }

        Some((start, end))
    }
}

// TODO: checkSelection -> check if valid for doc length && sorted

pub fn keep_or_remove_matches(
    text: RopeSlice,
    selection: &Selection,
    regex: &rope::Regex,
    remove: bool,
) -> Option<Selection> {
    let result: SmallVec<_> = selection
        .iter()
        .filter(|range| regex.is_match(text.regex_input_at(range.from()..range.to())) ^ remove)
        .copied()
        .collect();

    // TODO: figure out a new primary index
    if !result.is_empty() {
        return Some(Selection::new(result, 0));
    }
    None
}

// TODO: support to split on capture #N instead of whole match
pub fn select_on_matches(
    text: RopeSlice,
    selection: &Selection,
    regex: &rope::Regex,
) -> Option<Selection> {
    let mut result = SmallVec::with_capacity(selection.len());

    for sel in selection {
        for mat in regex.find_iter(text.regex_input_at(sel.from()..sel.to())) {
            // TODO: retain range direction

            let start = text.byte_to_char(mat.start());
            let end = text.byte_to_char(mat.end());

            let range = Range::new(start, end);
            // Make sure the match is not right outside of the selection.
            // These invalid matches can come from using RegEx anchors like `^`, `$`
            if range != Range::point(sel.to()) {
                result.push(range);
            }
        }
    }

    // TODO: figure out a new primary index
    if !result.is_empty() {
        return Some(Selection::new(result, 0));
    }

    None
}

pub fn split_on_newline(text: RopeSlice, selection: &Selection) -> Selection {
    let mut result = SmallVec::with_capacity(selection.len());

    for sel in selection {
        // Special case: zero-width selection.
        if sel.from() == sel.to() {
            result.push(*sel);
            continue;
        }

        let sel_start = sel.from();
        let sel_end = sel.to();

        let mut start = sel_start;

        for line in sel.slice(text).lines() {
            let Some(line_ending) = get_line_ending(&line) else {
                break;
            };
            let line_end = start + line.len_chars();
            // TODO: retain range direction
            result.push(Range::new(start, line_end - line_ending.len_chars()));
            start = line_end;
        }

        if start < sel_end {
            result.push(Range::new(start, sel_end));
        }
    }

    // TODO: figure out a new primary index
    Selection::new(result, 0)
}

pub fn split_on_matches(text: RopeSlice, selection: &Selection, regex: &rope::Regex) -> Selection {
    let mut result = SmallVec::with_capacity(selection.len());

    for sel in selection {
        // Special case: zero-width selection.
        if sel.from() == sel.to() {
            result.push(*sel);
            continue;
        }

        let sel_start = sel.from();
        let sel_end = sel.to();
        let mut start = sel_start;

        for mat in regex.find_iter(text.regex_input_at(sel_start..sel_end)) {
            // TODO: retain range direction
            let end = text.byte_to_char(mat.start());
            result.push(Range::new(start, end));
            start = text.byte_to_char(mat.end());
        }

        if start < sel_end {
            result.push(Range::new(start, sel_end));
        }
    }

    // TODO: figure out a new primary index
    Selection::new(result, 0)
}

#[cfg(test)]
mod test {
    use super::*;
    use crate::Rope;

    #[test]
    #[should_panic]
    fn test_new_empty() {
        let _ = Selection::new(smallvec![], 0);
    }

    #[test]
    fn test_create_normalizes_and_merges() {
        let sel = Selection::new(
            smallvec![
                Range::new(10, 12),
                Range::new(6, 7),
                Range::new(4, 5),
                Range::new(3, 4),
                Range::new(0, 6),
                Range::new(7, 8),
                Range::new(9, 13),
                Range::new(13, 14),
            ],
            0,
        );

        let res = sel
            .ranges
            .into_iter()
            .map(|range| format!("{}/{}", range.anchor, range.head))
            .collect::<Vec<String>>()
            .join(",");

        assert_eq!(res, "0/6,6/7,7/8,9/13,13/14");

        // it correctly calculates a new primary index
        let sel = Selection::new(
            smallvec![Range::new(0, 2), Range::new(1, 5), Range::new(4, 7)],
            2,
        );

        let res = sel
            .ranges
            .into_iter()
            .map(|range| format!("{}/{}", range.anchor, range.head))
            .collect::<Vec<String>>()
            .join(",");

        assert_eq!(res, "0/7");
        assert_eq!(sel.primary_index, 0);
    }

    #[test]
    fn test_create_merges_adjacent_points() {
        let sel = Selection::new(
            smallvec![
                Range::new(10, 12),
                Range::new(12, 12),
                Range::new(12, 12),
                Range::new(10, 10),
                Range::new(8, 10),
            ],
            0,
        );

        let res = sel
            .ranges
            .into_iter()
            .map(|range| format!("{}/{}", range.anchor, range.head))
            .collect::<Vec<String>>()
            .join(",");

        assert_eq!(res, "8/10,10/12,12/12");
    }

    #[test]
    fn test_contains() {
        let range = Range::new(10, 12);

        assert!(!range.contains(9));
        assert!(range.contains(10));
        assert!(range.contains(11));
        assert!(!range.contains(12));
        assert!(!range.contains(13));

        let range = Range::new(9, 6);
        assert!(!range.contains(9));
        assert!(range.contains(7));
        assert!(range.contains(6));
    }

    #[test]
    fn test_overlaps() {
        fn overlaps(a: (usize, usize), b: (usize, usize)) -> bool {
            Range::new(a.0, a.1).overlaps(&Range::new(b.0, b.1))
        }

        // Two non-zero-width ranges, no overlap.
        assert!(!overlaps((0, 3), (3, 6)));
        assert!(!overlaps((0, 3), (6, 3)));
        assert!(!overlaps((3, 0), (3, 6)));
        assert!(!overlaps((3, 0), (6, 3)));
        assert!(!overlaps((3, 6), (0, 3)));
        assert!(!overlaps((3, 6), (3, 0)));
        assert!(!overlaps((6, 3), (0, 3)));
        assert!(!overlaps((6, 3), (3, 0)));

        // Two non-zero-width ranges, overlap.
        assert!(overlaps((0, 4), (3, 6)));
        assert!(overlaps((0, 4), (6, 3)));
        assert!(overlaps((4, 0), (3, 6)));
        assert!(overlaps((4, 0), (6, 3)));
        assert!(overlaps((3, 6), (0, 4)));
        assert!(overlaps((3, 6), (4, 0)));
        assert!(overlaps((6, 3), (0, 4)));
        assert!(overlaps((6, 3), (4, 0)));

        // Zero-width and non-zero-width range, no overlap.
        assert!(!overlaps((0, 3), (3, 3)));
        assert!(!overlaps((3, 0), (3, 3)));
        assert!(!overlaps((3, 3), (0, 3)));
        assert!(!overlaps((3, 3), (3, 0)));

        // Zero-width and non-zero-width range, overlap.
        assert!(overlaps((1, 4), (1, 1)));
        assert!(overlaps((4, 1), (1, 1)));
        assert!(overlaps((1, 1), (1, 4)));
        assert!(overlaps((1, 1), (4, 1)));

        assert!(overlaps((1, 4), (3, 3)));
        assert!(overlaps((4, 1), (3, 3)));
        assert!(overlaps((3, 3), (1, 4)));
        assert!(overlaps((3, 3), (4, 1)));

        // Two zero-width ranges, no overlap.
        assert!(!overlaps((0, 0), (1, 1)));
        assert!(!overlaps((1, 1), (0, 0)));

        // Two zero-width ranges, overlap.
        assert!(overlaps((1, 1), (1, 1)));
    }

    #[test]
    fn test_grapheme_aligned() {
        let r = Rope::from_str("\r\nHi\r\n");
        let s = r.slice(..);

        // Zero-width.
        assert_eq!(Range::new(0, 0).grapheme_aligned(s), Range::new(0, 0));
        assert_eq!(Range::new(1, 1).grapheme_aligned(s), Range::new(0, 0));
        assert_eq!(Range::new(2, 2).grapheme_aligned(s), Range::new(2, 2));
        assert_eq!(Range::new(3, 3).grapheme_aligned(s), Range::new(3, 3));
        assert_eq!(Range::new(4, 4).grapheme_aligned(s), Range::new(4, 4));
        assert_eq!(Range::new(5, 5).grapheme_aligned(s), Range::new(4, 4));
        assert_eq!(Range::new(6, 6).grapheme_aligned(s), Range::new(6, 6));

        // Forward.
        assert_eq!(Range::new(0, 1).grapheme_aligned(s), Range::new(0, 2));
        assert_eq!(Range::new(1, 2).grapheme_aligned(s), Range::new(0, 2));
        assert_eq!(Range::new(2, 3).grapheme_aligned(s), Range::new(2, 3));
        assert_eq!(Range::new(3, 4).grapheme_aligned(s), Range::new(3, 4));
        assert_eq!(Range::new(4, 5).grapheme_aligned(s), Range::new(4, 6));
        assert_eq!(Range::new(5, 6).grapheme_aligned(s), Range::new(4, 6));

        assert_eq!(Range::new(0, 2).grapheme_aligned(s), Range::new(0, 2));
        assert_eq!(Range::new(1, 3).grapheme_aligned(s), Range::new(0, 3));
        assert_eq!(Range::new(2, 4).grapheme_aligned(s), Range::new(2, 4));
        assert_eq!(Range::new(3, 5).grapheme_aligned(s), Range::new(3, 6));
        assert_eq!(Range::new(4, 6).grapheme_aligned(s), Range::new(4, 6));

        // Reverse.
        assert_eq!(Range::new(1, 0).grapheme_aligned(s), Range::new(2, 0));
        assert_eq!(Range::new(2, 1).grapheme_aligned(s), Range::new(2, 0));
        assert_eq!(Range::new(3, 2).grapheme_aligned(s), Range::new(3, 2));
        assert_eq!(Range::new(4, 3).grapheme_aligned(s), Range::new(4, 3));
        assert_eq!(Range::new(5, 4).grapheme_aligned(s), Range::new(6, 4));
        assert_eq!(Range::new(6, 5).grapheme_aligned(s), Range::new(6, 4));

        assert_eq!(Range::new(2, 0).grapheme_aligned(s), Range::new(2, 0));
        assert_eq!(Range::new(3, 1).grapheme_aligned(s), Range::new(3, 0));
        assert_eq!(Range::new(4, 2).grapheme_aligned(s), Range::new(4, 2));
        assert_eq!(Range::new(5, 3).grapheme_aligned(s), Range::new(6, 3));
        assert_eq!(Range::new(6, 4).grapheme_aligned(s), Range::new(6, 4));
    }

    #[test]
    fn test_min_width_1() {
        let r = Rope::from_str("\r\nHi\r\n");
        let s = r.slice(..);

        // Zero-width.
        assert_eq!(Range::new(0, 0).min_width_1(s), Range::new(0, 2));
        assert_eq!(Range::new(1, 1).min_width_1(s), Range::new(1, 2));
        assert_eq!(Range::new(2, 2).min_width_1(s), Range::new(2, 3));
        assert_eq!(Range::new(3, 3).min_width_1(s), Range::new(3, 4));
        assert_eq!(Range::new(4, 4).min_width_1(s), Range::new(4, 6));
        assert_eq!(Range::new(5, 5).min_width_1(s), Range::new(5, 6));
        assert_eq!(Range::new(6, 6).min_width_1(s), Range::new(6, 6));

        // Forward.
        assert_eq!(Range::new(0, 1).min_width_1(s), Range::new(0, 1));
        assert_eq!(Range::new(1, 2).min_width_1(s), Range::new(1, 2));
        assert_eq!(Range::new(2, 3).min_width_1(s), Range::new(2, 3));
        assert_eq!(Range::new(3, 4).min_width_1(s), Range::new(3, 4));
        assert_eq!(Range::new(4, 5).min_width_1(s), Range::new(4, 5));
        assert_eq!(Range::new(5, 6).min_width_1(s), Range::new(5, 6));

        // Reverse.
        assert_eq!(Range::new(1, 0).min_width_1(s), Range::new(1, 0));
        assert_eq!(Range::new(2, 1).min_width_1(s), Range::new(2, 1));
        assert_eq!(Range::new(3, 2).min_width_1(s), Range::new(3, 2));
        assert_eq!(Range::new(4, 3).min_width_1(s), Range::new(4, 3));
        assert_eq!(Range::new(5, 4).min_width_1(s), Range::new(5, 4));
        assert_eq!(Range::new(6, 5).min_width_1(s), Range::new(6, 5));
    }

    #[test]
    fn test_select_on_matches() {
        let r = Rope::from_str("Nobody expects the Spanish inquisition");
        let s = r.slice(..);

        let selection = Selection::single(0, r.len_chars());
        assert_eq!(
            select_on_matches(s, &selection, &rope::Regex::new(r"[A-Z][a-z]*").unwrap()),
            Some(Selection::new(
                smallvec![Range::new(0, 6), Range::new(19, 26)],
                0
            ))
        );

        let r = Rope::from_str("This\nString\n\ncontains multiple\nlines");
        let s = r.slice(..);

        let start_of_line = rope::RegexBuilder::new()
            .syntax(rope::Config::new().multi_line(true))
            .build(r"^")
            .unwrap();
        let end_of_line = rope::RegexBuilder::new()
            .syntax(rope::Config::new().multi_line(true))
            .build(r"$")
            .unwrap();

        // line without ending
        assert_eq!(
            select_on_matches(s, &Selection::single(0, 4), &start_of_line),
            Some(Selection::single(0, 0))
        );
        assert_eq!(
            select_on_matches(s, &Selection::single(0, 4), &end_of_line),
            None
        );
        // line with ending
        assert_eq!(
            select_on_matches(s, &Selection::single(0, 5), &start_of_line),
            Some(Selection::single(0, 0))
        );
        assert_eq!(
            select_on_matches(s, &Selection::single(0, 5), &end_of_line),
            Some(Selection::single(4, 4))
        );
        // line with start of next line
        assert_eq!(
            select_on_matches(s, &Selection::single(0, 6), &start_of_line),
            Some(Selection::new(
                smallvec![Range::point(0), Range::point(5)],
                0
            ))
        );
        assert_eq!(
            select_on_matches(s, &Selection::single(0, 6), &end_of_line),
            Some(Selection::single(4, 4))
        );

        // multiple lines
        assert_eq!(
            select_on_matches(
                s,
                &Selection::single(0, s.len_chars()),
                &rope::RegexBuilder::new()
                    .syntax(rope::Config::new().multi_line(true))
                    .build(r"^[a-z ]*$")
                    .unwrap()
            ),
            Some(Selection::new(
                smallvec![Range::point(12), Range::new(13, 30), Range::new(31, 36)],
                0
            ))
        );
    }

    #[test]
    fn test_line_range() {
        let r = Rope::from_str("\r\nHi\r\nthere!");
        let s = r.slice(..);

        // Zero-width ranges.
        assert_eq!(Range::new(0, 0).line_range(s), (0, 0));
        assert_eq!(Range::new(1, 1).line_range(s), (0, 0));
        assert_eq!(Range::new(2, 2).line_range(s), (1, 1));
        assert_eq!(Range::new(3, 3).line_range(s), (1, 1));

        // Forward ranges.
        assert_eq!(Range::new(0, 1).line_range(s), (0, 0));
        assert_eq!(Range::new(0, 2).line_range(s), (0, 0));
        assert_eq!(Range::new(0, 3).line_range(s), (0, 1));
        assert_eq!(Range::new(1, 2).line_range(s), (0, 0));
        assert_eq!(Range::new(2, 3).line_range(s), (1, 1));
        assert_eq!(Range::new(3, 8).line_range(s), (1, 2));
        assert_eq!(Range::new(0, 12).line_range(s), (0, 2));

        // Reverse ranges.
        assert_eq!(Range::new(1, 0).line_range(s), (0, 0));
        assert_eq!(Range::new(2, 0).line_range(s), (0, 0));
        assert_eq!(Range::new(3, 0).line_range(s), (0, 1));
        assert_eq!(Range::new(2, 1).line_range(s), (0, 0));
        assert_eq!(Range::new(3, 2).line_range(s), (1, 1));
        assert_eq!(Range::new(8, 3).line_range(s), (1, 2));
        assert_eq!(Range::new(12, 0).line_range(s), (0, 2));
    }

    #[test]
    fn selection_line_ranges() {
        let (text, selection) = crate::test::print(
            r#"                                           L0
            #[|these]# line #(|ranges)# are #(|merged)#   L1
                                                          L2
            single one-line #(|range)#                    L3
                                                          L4
            single #(|multiline                           L5
            range)#                                       L6
                                                          L7
            these #(|multiline                            L8
            ranges)# are #(|also                          L9
            merged)#                                      L10
                                                          L11
            adjacent #(|ranges)#                          L12
            are merged #(|the same way)#                  L13
            "#,
        );
        let rope = Rope::from_str(&text);
        assert_eq!(
            vec![(1, 1), (3, 3), (5, 6), (8, 10), (12, 13)],
            selection.line_ranges(rope.slice(..)).collect::<Vec<_>>(),
        );
    }

    #[test]
    fn test_cursor() {
        let r = Rope::from_str("\r\nHi\r\nthere!");
        let s = r.slice(..);

        // Zero-width ranges.
        assert_eq!(Range::new(0, 0).cursor(s), 0);
        assert_eq!(Range::new(2, 2).cursor(s), 2);
        assert_eq!(Range::new(3, 3).cursor(s), 3);

        // Forward ranges.
        assert_eq!(Range::new(0, 2).cursor(s), 0);
        assert_eq!(Range::new(0, 3).cursor(s), 2);
        assert_eq!(Range::new(3, 6).cursor(s), 4);

        // Reverse ranges.
        assert_eq!(Range::new(2, 0).cursor(s), 0);
        assert_eq!(Range::new(6, 2).cursor(s), 2);
        assert_eq!(Range::new(6, 3).cursor(s), 3);
    }

    #[test]
    fn test_put_cursor() {
        let r = Rope::from_str("\r\nHi\r\nthere!");
        let s = r.slice(..);

        // Zero-width ranges.
        assert_eq!(Range::new(0, 0).put_cursor(s, 0, true), Range::new(0, 2));
        assert_eq!(Range::new(0, 0).put_cursor(s, 2, true), Range::new(0, 3));
        assert_eq!(Range::new(2, 3).put_cursor(s, 4, true), Range::new(2, 6));
        assert_eq!(Range::new(2, 8).put_cursor(s, 4, true), Range::new(2, 6));
        assert_eq!(Range::new(8, 8).put_cursor(s, 4, true), Range::new(9, 4));

        // Forward ranges.
        assert_eq!(Range::new(3, 6).put_cursor(s, 0, true), Range::new(4, 0));
        assert_eq!(Range::new(3, 6).put_cursor(s, 2, true), Range::new(4, 2));
        assert_eq!(Range::new(3, 6).put_cursor(s, 3, true), Range::new(3, 4));
        assert_eq!(Range::new(3, 6).put_cursor(s, 4, true), Range::new(3, 6));
        assert_eq!(Range::new(3, 6).put_cursor(s, 6, true), Range::new(3, 7));
        assert_eq!(Range::new(3, 6).put_cursor(s, 8, true), Range::new(3, 9));

        // Reverse ranges.
        assert_eq!(Range::new(6, 3).put_cursor(s, 0, true), Range::new(6, 0));
        assert_eq!(Range::new(6, 3).put_cursor(s, 2, true), Range::new(6, 2));
        assert_eq!(Range::new(6, 3).put_cursor(s, 3, true), Range::new(6, 3));
        assert_eq!(Range::new(6, 3).put_cursor(s, 4, true), Range::new(6, 4));
        assert_eq!(Range::new(6, 3).put_cursor(s, 6, true), Range::new(4, 7));
        assert_eq!(Range::new(6, 3).put_cursor(s, 8, true), Range::new(4, 9));
    }

    #[test]
    fn test_split_on_matches() {
        let text = Rope::from(" abcd efg wrs   xyz 123 456");

        let selection = Selection::new(smallvec![Range::new(0, 9), Range::new(11, 20),], 0);

        let result = split_on_matches(
            text.slice(..),
            &selection,
            &rope::Regex::new(r"\s+").unwrap(),
        );

        assert_eq!(
            result.ranges(),
            &[
                // TODO: rather than this behavior, maybe we want it
                // to be based on which side is the anchor?
                //
                // We get a leading zero-width range when there's
                // a leading match because ranges are inclusive on
                // the left.  Imagine, for example, if the entire
                // selection range were matched: you'd still want
                // at least one range to remain after the split.
                Range::new(0, 0),
                Range::new(1, 5),
                Range::new(6, 9),
                Range::new(11, 13),
                Range::new(16, 19),
                // In contrast to the comment above, there is no
                // _trailing_ zero-width range despite the trailing
                // match, because ranges are exclusive on the right.
            ]
        );

        assert_eq!(
            result.fragments(text.slice(..)).collect::<Vec<_>>(),
            &["", "abcd", "efg", "rs", "xyz"]
        );
    }

    #[test]
    fn test_merge_consecutive_ranges() {
        let selection = Selection::new(
            smallvec![
                Range::new(0, 1),
                Range::new(1, 10),
                Range::new(15, 20),
                Range::new(25, 26),
                Range::new(26, 30)
            ],
            4,
        );

        let result = selection.merge_consecutive_ranges();

        assert_eq!(
            result.ranges(),
            &[Range::new(0, 10), Range::new(15, 20), Range::new(25, 30)]
        );
        assert_eq!(result.primary_index, 2);

        let selection = Selection::new(smallvec![Range::new(0, 1)], 0);
        let result = selection.merge_consecutive_ranges();

        assert_eq!(result.ranges(), &[Range::new(0, 1)]);
        assert_eq!(result.primary_index, 0);

        let selection = Selection::new(
            smallvec![
                Range::new(0, 1),
                Range::new(1, 5),
                Range::new(5, 8),
                Range::new(8, 10),
                Range::new(10, 15),
                Range::new(18, 25)
            ],
            3,
        );

        let result = selection.merge_consecutive_ranges();

        assert_eq!(result.ranges(), &[Range::new(0, 15), Range::new(18, 25)]);
        assert_eq!(result.primary_index, 0);
    }

    #[test]
    fn test_selection_contains() {
        fn contains(a: Vec<(usize, usize)>, b: Vec<(usize, usize)>) -> bool {
            let sela = Selection::new(a.iter().map(|a| Range::new(a.0, a.1)).collect(), 0);
            let selb = Selection::new(b.iter().map(|b| Range::new(b.0, b.1)).collect(), 0);
            sela.contains(&selb)
        }

        // exact match
        assert!(contains(vec!((1, 1)), vec!((1, 1))));

        // larger set contains smaller
        assert!(contains(vec!((1, 1), (2, 2), (3, 3)), vec!((2, 2))));

        // multiple matches
        assert!(contains(vec!((1, 1), (2, 2)), vec!((1, 1), (2, 2))));

        // smaller set can't contain bigger
        assert!(!contains(vec!((1, 1)), vec!((1, 1), (2, 2))));

        assert!(contains(
            vec!((1, 1), (2, 4), (5, 6), (7, 9), (10, 13)),
            vec!((3, 4), (7, 9))
        ));
        assert!(!contains(vec!((1, 1), (5, 6)), vec!((1, 6))));

        // multiple ranges of other are all contained in some ranges of self,
        assert!(contains(
            vec!((1, 4), (7, 10)),
            vec!((1, 2), (3, 4), (7, 9))
        ));
    }
}