repose_text/

lib.rs

use cosmic_text::{
    Attrs, Buffer, CacheKey, Family, FontSystem, Metrics, Shaping, SwashCache, SwashContent,
};
use once_cell::sync::OnceCell;
use rapidhash::{HashMapExt, RapidHashMap, fast::RapidHasher};
use std::sync::atomic::{AtomicU64, Ordering};
use std::{
    collections::{HashMap, VecDeque},
    hash::{Hash, Hasher},
    sync::Mutex,
};
use unicode_segmentation::UnicodeSegmentation;

/// Frame counter for cache invalidation strategies.
static FRAME_COUNTER: AtomicU64 = AtomicU64::new(0);

/// Call this at the start of each frame to enable frame-aware caching.
pub fn begin_frame() {
    FRAME_COUNTER.fetch_add(1, Ordering::Relaxed);
}

pub fn current_frame() -> u64 {
    FRAME_COUNTER.load(Ordering::Relaxed)
}

const WRAP_CACHE_CAP: usize = 1024;
const ELLIP_CACHE_CAP: usize = 2048;

static METRICS_LRU: OnceCell<Mutex<Lru<(u64, u32, u64), TextMetrics>>> = OnceCell::new();
fn metrics_cache() -> &'static Mutex<Lru<(u64, u32, u64), TextMetrics>> {
    METRICS_LRU.get_or_init(|| Mutex::new(Lru::new(4096)))
}

struct Lru<K, V> {
    map: RapidHashMap<K, V>,
    order: VecDeque<K>,
    cap: usize,
}
impl<K: std::hash::Hash + Eq + Clone, V> Lru<K, V> {
    fn new(cap: usize) -> Self {
        Self {
            map: RapidHashMap::new(),
            order: VecDeque::new(),
            cap,
        }
    }
    fn get(&mut self, k: &K) -> Option<&V> {
        if self.map.contains_key(k) {
            // move to back
            if let Some(pos) = self.order.iter().position(|x| x == k) {
                let key = self.order.remove(pos).unwrap();
                self.order.push_back(key);
            }
        }
        self.map.get(k)
    }
    fn put(&mut self, k: K, v: V) {
        if self.map.contains_key(&k) {
            self.map.insert(k.clone(), v);
            if let Some(pos) = self.order.iter().position(|x| x == &k) {
                let key = self.order.remove(pos).unwrap();
                self.order.push_back(key);
            }
            return;
        }
        if self.map.len() >= self.cap
            && let Some(old) = self.order.pop_front()
        {
            self.map.remove(&old);
        }
        self.order.push_back(k.clone());
        self.map.insert(k, v);
    }
}

static WRAP_LRU: OnceCell<Mutex<Lru<(u64, u32, u32, u16, bool), (Vec<String>, bool)>>> =
    OnceCell::new();

static WRAP_RANGES_LRU: OnceCell<
    Mutex<Lru<(u64, u32, u32, u16, bool), (Vec<(usize, usize)>, bool)>>,
> = OnceCell::new();

static ELLIP_LRU: OnceCell<Mutex<Lru<(u64, u32, u32), String>>> = OnceCell::new();

fn wrap_cache() -> &'static Mutex<Lru<(u64, u32, u32, u16, bool), (Vec<String>, bool)>> {
    WRAP_LRU.get_or_init(|| Mutex::new(Lru::new(WRAP_CACHE_CAP)))
}

fn wrap_ranges_cache()
-> &'static Mutex<Lru<(u64, u32, u32, u16, bool), (Vec<(usize, usize)>, bool)>> {
    WRAP_RANGES_LRU.get_or_init(|| Mutex::new(Lru::new(WRAP_CACHE_CAP)))
}

fn ellip_cache() -> &'static Mutex<Lru<(u64, u32, u32), String>> {
    ELLIP_LRU.get_or_init(|| Mutex::new(Lru::new(ELLIP_CACHE_CAP)))
}

fn fast_hash(s: &str) -> u64 {
    use std::hash::{Hash, Hasher};
    let mut h = RapidHasher::default();
    s.hash(&mut h);
    h.finish()
}

#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
pub struct GlyphKey(pub u64);

pub struct ShapedGlyph {
    pub key: GlyphKey,
    pub x: f32,
    pub y: f32,
    pub w: f32,
    pub h: f32,
    pub bearing_x: f32,
    pub bearing_y: f32,
    pub advance: f32,
}

pub struct GlyphBitmap {
    pub key: GlyphKey,
    pub w: u32,
    pub h: u32,
    pub content: SwashContent,
    pub data: Vec<u8>, // Mask: A8; Color/Subpixel: RGBA8
}

struct Engine {
    fs: FontSystem,
    cache: SwashCache,
    // Map our compact atlas key -> full cosmic_text CacheKey
    key_map: HashMap<GlyphKey, CacheKey>,
}

impl Engine {
    fn get_image(&mut self, key: CacheKey) -> Option<cosmic_text::SwashImage> {
        // inside this method we may freely borrow both fields
        self.cache.get_image(&mut self.fs, key).clone()
    }
}

static ENGINE: OnceCell<Mutex<Engine>> = OnceCell::new();

fn engine() -> &'static Mutex<Engine> {
    ENGINE.get_or_init(|| {
        #[allow(unused_mut)]
        let mut fs = FontSystem::new();

        let cache = SwashCache::new();

        // #[cfg(any(target_os = "android", target_arch = "wasm32"))]
        // // Until cosmic-text has android/web font loading support, would save around 15mb?
        {
            static FALLBACK_TTF: &[u8] = include_bytes!("assets/OpenSans-Regular.ttf"); // GFonts, OFL licensed
            static FALLBACK_EMOJI_TTF: &[u8] = include_bytes!("assets/NotoColorEmoji-Regular.ttf"); // GFonts, OFL licensed
            static FALLBACK_SYMBOLS_TTF: &[u8] =
                include_bytes!("assets/NotoSansSymbols2-Regular.ttf"); // GFonts, OFL licensed
            static MATERIAL_SYMBOLS_TTF: &[u8] =
                include_bytes!("assets/MaterialSymbolsOutlined.ttf"); // Google Fonts, Apache 2.0 licensed
            {
                // Register fallback font data into font DB
                let db = fs.db_mut();
                db.load_font_data(FALLBACK_TTF.to_vec());
                db.set_sans_serif_family("Open Sans".to_string());

                db.load_font_data(FALLBACK_SYMBOLS_TTF.to_vec());
                db.load_font_data(FALLBACK_EMOJI_TTF.to_vec());
                db.load_font_data(MATERIAL_SYMBOLS_TTF.to_vec());
            }
        }
        Mutex::new(Engine {
            fs,
            cache,
            key_map: HashMap::new(),
        })
    })
}

/// Register a font blob into the global FontSystem.
pub fn register_font_data(bytes: &'static [u8]) {
    let mut eng = engine().lock().unwrap();
    eng.fs.db_mut().load_font_data(bytes.to_vec());
}

// Utility: stable u64 key from a CacheKey using its Hash impl
fn key_from_cachekey(k: &CacheKey) -> GlyphKey {
    let mut h = RapidHasher::default();
    k.hash(&mut h);
    GlyphKey(h.finish())
}

// Shape a single-line string (no wrapping). Returns positioned glyphs relative to baseline y=0.
// `font_family` optionally overrides the default font (e.g. "Material Symbols Outlined").
pub fn shape_line(text: &str, px: f32, font_family: Option<&str>) -> Vec<ShapedGlyph> {
    let mut eng = engine().lock().unwrap();

    // Construct a temporary buffer each call; FontSystem and caches are retained globally
    let mut buf = Buffer::new(&mut eng.fs, Metrics::new(px, px * 1.3));
    {
        // Borrow with FS for ergonomic setters (no FS arg)
        let mut b = buf.borrow_with(&mut eng.fs);
        b.set_size(None, None);
        let attrs = match font_family {
            Some(family) => Attrs::new().family(Family::Name(family)),
            None => Attrs::new(),
        };
        b.set_text(text, &attrs, Shaping::Advanced, None);
        b.shape_until_scroll(true);
    }

    let mut out = Vec::new();
    for run in buf.layout_runs() {
        for g in run.glyphs {
            // Compute physical glyph: gives cache_key and integer pixel position
            let phys = g.physical((0.0, run.line_y), 1.0);
            let key = key_from_cachekey(&phys.cache_key);
            eng.key_map.insert(key, phys.cache_key);

            // Query raster cache to get placement for metrics
            let img_opt = eng.get_image(phys.cache_key);
            let (w, h, left, top) = if let Some(img) = img_opt.as_ref() {
                (
                    img.placement.width as f32,
                    img.placement.height as f32,
                    img.placement.left as f32,
                    img.placement.top as f32,
                )
            } else {
                (0.0, 0.0, 0.0, 0.0)
            };

            out.push(ShapedGlyph {
                key,
                x: g.x + g.x_offset, // visual x
                y: run.line_y,       // baseline y
                w,
                h,
                bearing_x: left,
                bearing_y: top,
                advance: g.w,
            });
        }
    }
    out
}

// Rasterize a glyph mask (A8) or color/subpixel (RGBA8) for a given shaped key.
// Returns owned pixels to avoid borrowing from the cache.
pub fn rasterize(key: GlyphKey, _px: f32) -> Option<GlyphBitmap> {
    let mut eng = engine().lock().unwrap();
    let &ck = eng.key_map.get(&key)?;

    let img = eng.get_image(ck).as_ref()?.clone();
    Some(GlyphBitmap {
        key,
        w: img.placement.width,
        h: img.placement.height,
        content: img.content,
        data: img.data, // already a Vec<u8>
    })
}

/// Look up the full CacheKey for a compact GlyphKey.
pub fn lookup_cache_key(key: GlyphKey) -> Option<cosmic_text::CacheKey> {
    let eng = engine().lock().unwrap();
    eng.key_map.get(&key).copied()
}

/// Extract vector outline commands for a glyph (uncached — caller should cache).
/// Returns quadratic + cubic bezier commands in font-unit scaled coordinates.
pub fn extract_outline_commands(
    cache_key: cosmic_text::CacheKey,
) -> Option<Box<[cosmic_text::Command]>> {
    let mut eng = engine().lock().unwrap();
    let Engine {
        ref mut cache,
        ref mut fs,
        ..
    } = *eng;
    cache.get_outline_commands_uncached(fs, cache_key)
}

/// Look up the CacheKey for a GlyphKey and extract outline commands in one engine lock.
/// Returns `None` if the GlyphKey isn't in the key map or extraction fails.
pub fn lookup_and_extract_outline(
    key: GlyphKey,
) -> Option<(cosmic_text::CacheKey, Box<[cosmic_text::Command]>)> {
    let mut eng = engine().lock().unwrap();
    let ck = eng.key_map.get(&key).copied()?;
    let Engine {
        ref mut cache,
        ref mut fs,
        ..
    } = *eng;
    let cmds = cache.get_outline_commands_uncached(fs, ck)?;
    Some((ck, cmds))
}

// Text metrics for TextField: positions per grapheme boundary and byte offsets.
#[derive(Clone)]
pub struct TextMetrics {
    pub positions: Vec<f32>,      // cumulative advance per boundary (len == n+1)
    pub byte_offsets: Vec<usize>, // byte index per boundary (len == n+1)
}

/// Computes caret mapping using shaping (no wrapping).
/// `font_family` optionally overrides the default font (e.g. "Material Symbols Outlined").
pub fn metrics_for_textfield(text: &str, px: f32, font_family: Option<&str>) -> TextMetrics {
    let family_hash = font_family.map(fast_hash).unwrap_or(0);
    let key = (fast_hash(text), (px * 100.0) as u32, family_hash);
    if let Some(m) = metrics_cache().lock().unwrap().get(&key).cloned() {
        return m;
    }
    let mut eng = engine().lock().unwrap();
    let mut buf = Buffer::new(&mut eng.fs, Metrics::new(px, px * 1.3));
    {
        let mut b = buf.borrow_with(&mut eng.fs);
        b.set_size(None, None);
        let attrs = match font_family {
            Some(family) => Attrs::new().family(Family::Name(family)),
            None => Attrs::new(),
        };
        b.set_text(text, &attrs, Shaping::Advanced, None);
        b.shape_until_scroll(true);
    }
    let mut edges: Vec<(usize, f32)> = Vec::new();
    let mut last_x = 0.0f32;
    for run in buf.layout_runs() {
        for g in run.glyphs {
            let right = g.x + g.w;
            last_x = right.max(last_x);
            edges.push((g.end, right));
        }
    }
    if edges.last().map(|e| e.0) != Some(text.len()) {
        edges.push((text.len(), last_x));
    }
    let mut positions = Vec::with_capacity(text.graphemes(true).count() + 1);
    let mut byte_offsets = Vec::with_capacity(positions.capacity());
    positions.push(0.0);
    byte_offsets.push(0);
    let mut last_byte = 0usize;
    for (b, _) in text.grapheme_indices(true) {
        positions
            .push(positions.last().copied().unwrap_or(0.0) + width_between(&edges, last_byte, b));
        byte_offsets.push(b);
        last_byte = b;
    }
    if *byte_offsets.last().unwrap_or(&0) != text.len() {
        positions.push(
            positions.last().copied().unwrap_or(0.0) + width_between(&edges, last_byte, text.len()),
        );
        byte_offsets.push(text.len());
    }
    let m = TextMetrics {
        positions,
        byte_offsets,
    };
    metrics_cache().lock().unwrap().put(key, m.clone());
    m
}

fn width_between(edges: &[(usize, f32)], start_b: usize, end_b: usize) -> f32 {
    let x0 = lookup_right(edges, start_b);
    let x1 = lookup_right(edges, end_b);
    (x1 - x0).max(0.0)
}
fn lookup_right(edges: &[(usize, f32)], b: usize) -> f32 {
    match edges.binary_search_by_key(&b, |e| e.0) {
        Ok(i) => edges[i].1,
        Err(i) => {
            if i == 0 {
                0.0
            } else {
                edges[i - 1].1
            }
        }
    }
}

/// Greedy wrap into lines that fit max_width. Prefers breaking at whitespace,
/// falls back to grapheme boundaries. If max_lines is Some and we truncate,
/// caller can choose to ellipsize the last visible line.
pub fn wrap_lines(
    text: &str,
    px: f32,
    max_width: f32,
    max_lines: Option<usize>,
    soft_wrap: bool,
) -> (Vec<String>, bool) {
    if text.is_empty() || max_width <= 0.0 {
        return (vec![String::new()], false);
    }
    if !soft_wrap {
        return (vec![text.to_string()], false);
    }

    let max_lines_key: u16 = match max_lines {
        None => 0,
        Some(n) => {
            let n = n.min(u16::MAX as usize - 1) as u16;
            n.saturating_add(1)
        }
    };
    let key = (
        fast_hash(text),
        (px * 100.0) as u32,
        (max_width * 100.0) as u32,
        max_lines_key,
        soft_wrap,
    );
    if let Some(h) = wrap_cache().lock().unwrap().get(&key).cloned() {
        return h;
    }

    // Shape once and reuse positions/byte mapping.
    let m = metrics_for_textfield(text, px, None);
    // Fast path: fits
    if let Some(&last) = m.positions.last()
        && last <= max_width + 0.5
    {
        return (vec![text.to_string()], false);
    }

    // Helper: width of substring [start..end] in bytes
    let width_of = |start_b: usize, end_b: usize| -> f32 {
        let i0 = match m.byte_offsets.binary_search(&start_b) {
            Ok(i) | Err(i) => i,
        };
        let i1 = match m.byte_offsets.binary_search(&end_b) {
            Ok(i) | Err(i) => i,
        };
        (m.positions.get(i1).copied().unwrap_or(0.0) - m.positions.get(i0).copied().unwrap_or(0.0))
            .max(0.0)
    };

    let mut out: Vec<String> = Vec::new();
    let mut truncated = false;

    let mut line_start = 0usize; // byte index
    let mut best_break = line_start;

    // Iterate word boundaries (keep whitespace tokens so they factor widths)
    for tok in text.split_word_bounds() {
        let tok_start = best_break;
        let tok_end = tok_start + tok.len();
        let w = width_of(line_start, tok_end);

        if w <= max_width + 0.5 {
            best_break = tok_end;
            continue;
        }

        // Need to break the line before tok_end.
        if best_break > line_start {
            // Break at last good boundary
            out.push(text[line_start..best_break].trim_end().to_string());
            line_start = best_break;
        } else {
            // Token itself too wide: force break inside token at grapheme boundaries
            let mut cut = tok_start;
            for g in tok.grapheme_indices(true) {
                let next = tok_start + g.0 + g.1.len();
                if width_of(line_start, next) <= max_width + 0.5 {
                    cut = next;
                } else {
                    break;
                }
            }
            if cut == line_start {
                // nothing fits; fall back to single grapheme
                if let Some((ofs, grapheme)) = tok.grapheme_indices(true).next() {
                    cut = tok_start + ofs + grapheme.len();
                }
            }
            out.push(text[line_start..cut].to_string());
            line_start = cut;
        }

        // Check max_lines
        if let Some(ml) = max_lines
            && out.len() >= ml
        {
            truncated = true;
            // Stop; caller may ellipsize the last line
            line_start = line_start.min(text.len());
            break;
        }

        // Reset best_break for new line
        best_break = line_start;

        // Re-consider current token if not fully consumed
        if line_start < tok_end {
            // recompute width with the remaining token portion
            if width_of(line_start, tok_end) <= max_width + 0.5 {
                best_break = tok_end;
            } else {
                // will be handled in next iterations (or forced again)
            }
        }
    }

    // Push tail if allowed
    if line_start < text.len() && max_lines.is_none_or(|ml| out.len() < ml) {
        out.push(text[line_start..].trim_end().to_string());
    }

    let res = (out, truncated);

    wrap_cache().lock().unwrap().put(key, res.clone());
    res
}

/// Like `wrap_lines`, but returns byte ranges into the original `text`
/// for each visual line. This is required for multi-line editing so
/// caret/selection mapping stays correct.
///
/// Ranges are half-open `[start, end)`, and never include the '\n' char
/// (hard line breaks end a range at the '\n' byte index).
pub fn wrap_line_ranges(
    text: &str,
    px: f32,
    max_width: f32,
    max_lines: Option<usize>,
    soft_wrap: bool,
) -> (Vec<(usize, usize)>, bool) {
    if text.is_empty() || max_width <= 0.0 {
        return (vec![(0, 0)], false);
    }
    if !soft_wrap {
        // Hard lines only (split on '\n' but no width wrapping)
        let mut out = Vec::new();
        let mut start = 0usize;
        for (i, ch) in text.char_indices() {
            if ch == '\n' {
                out.push((start, i));
                start = i + 1;
            }
        }
        out.push((start, text.len()));
        return (out, false);
    }

    let max_lines_key: u16 = match max_lines {
        None => 0,
        Some(n) => {
            let n = n.min(u16::MAX as usize - 1) as u16;
            n.saturating_add(1)
        }
    };
    let key = (
        fast_hash(text),
        (px * 100.0) as u32,
        (max_width * 100.0) as u32,
        max_lines_key,
        soft_wrap,
    );
    if let Some(v) = wrap_ranges_cache().lock().unwrap().get(&key).cloned() {
        return v;
    }

    // Shape once for width queries (whole string)
    let m = metrics_for_textfield(text, px, None);

    // Helper: width of substring [start..end] in bytes using m
    let width_of = |start_b: usize, end_b: usize| -> f32 {
        let i0 = match m.byte_offsets.binary_search(&start_b) {
            Ok(i) | Err(i) => i,
        };
        let i1 = match m.byte_offsets.binary_search(&end_b) {
            Ok(i) | Err(i) => i,
        };
        (m.positions.get(i1).copied().unwrap_or(0.0) - m.positions.get(i0).copied().unwrap_or(0.0))
            .max(0.0)
    };

    let mut out: Vec<(usize, usize)> = Vec::new();
    let mut truncated = false;

    // Process hard lines split by '\n' while preserving original indices.
    let mut line0_start = 0usize;
    for (i, ch) in text.char_indices() {
        if ch == '\n' {
            let (mut ranges, tr) = wrap_one_hard_line_ranges(
                text,
                line0_start,
                i,
                max_width,
                max_lines.map(|ml| ml.saturating_sub(out.len())),
                &width_of,
            );
            out.append(&mut ranges);
            if tr {
                truncated = true;
                break;
            }
            line0_start = i + 1;

            if let Some(ml) = max_lines
                && out.len() >= ml
            {
                truncated = true;
                break;
            }
        }
    }
    if !truncated {
        let (mut ranges, tr) = wrap_one_hard_line_ranges(
            text,
            line0_start,
            text.len(),
            max_width,
            max_lines.map(|ml| ml.saturating_sub(out.len())),
            &width_of,
        );
        out.append(&mut ranges);
        truncated = tr;
    }

    if out.is_empty() {
        out.push((0, 0));
    }

    let res = (out, truncated);
    wrap_ranges_cache().lock().unwrap().put(key, res.clone());
    res
}

fn wrap_one_hard_line_ranges(
    text: &str,
    start: usize,
    end: usize,
    max_width: f32,
    max_lines: Option<usize>,
    width_of: &dyn Fn(usize, usize) -> f32,
) -> (Vec<(usize, usize)>, bool) {
    let mut out = Vec::new();
    let mut t = false;

    if start >= end {
        out.push((start, start));
        return (out, false);
    }

    // Fast path: whole line fits
    if width_of(start, end) <= max_width + 0.5 {
        out.push((start, end));
        return (out, false);
    }

    let mut line_start = start;
    let mut best_break = line_start;
    let mut unconsumed_start = start;

    for tok in text[line_start..end].split_word_bounds() {
        let tok_abs_start = unconsumed_start;
        let tok_abs_end = tok_abs_start + tok.len();
        unconsumed_start = tok_abs_end;

        let w = width_of(line_start, tok_abs_end);
        if w <= max_width + 0.5 {
            best_break = tok_abs_end;
            continue;
        }

        // Need break before tok_abs_end.
        if best_break > line_start {
            out.push((line_start, best_break));
            line_start = best_break;
        } else {
            // Token too wide: force break at grapheme boundaries
            let mut cut = tok_abs_start;
            for (ofs, g) in tok.grapheme_indices(true) {
                let next = tok_abs_start + ofs + g.len();
                if width_of(line_start, next) <= max_width + 0.5 {
                    cut = next;
                } else {
                    break;
                }
            }
            if cut == line_start
                && let Some((ofs, gr)) = tok.grapheme_indices(true).next()
            {
                cut = tok_abs_start + ofs + gr.len();
            }
            out.push((line_start, cut));
            line_start = cut;
        }

        // Max lines check
        if let Some(ml) = max_lines
            && out.len() >= ml
        {
            t = true;
            break;
        }

        best_break = line_start;
    }

    // Tail
    if !t && line_start < end && max_lines.is_none_or(|ml| out.len() < ml) {
        out.push((line_start, end));
    }

    (out, t)
}

/// Return a string truncated to fit max_width at the given px size, appending '…' if truncated.
pub fn ellipsize_line(text: &str, px: f32, max_width: f32) -> String {
    if text.is_empty() || max_width <= 0.0 {
        return String::new();
    }
    let key = (
        fast_hash(text),
        (px * 100.0) as u32,
        (max_width * 100.0) as u32,
    );
    if let Some(s) = ellip_cache().lock().unwrap().get(&key).cloned() {
        return s;
    }
    let m = metrics_for_textfield(text, px, None);
    if let Some(&last) = m.positions.last()
        && last <= max_width + 0.5
    {
        return text.to_string();
    }
    let _el = "…";
    let e_w = ellipsis_width(px);
    if e_w >= max_width {
        return String::new();
    }
    // Find last grapheme index whose width + ellipsis fits
    let mut cut_i = 0usize;
    for i in 0..m.positions.len() {
        if m.positions[i] + e_w <= max_width {
            cut_i = i;
        } else {
            break;
        }
    }
    let byte = m
        .byte_offsets
        .get(cut_i)
        .copied()
        .unwrap_or(0)
        .min(text.len());
    let mut out = String::with_capacity(byte + 3);
    out.push_str(&text[..byte]);
    out.push('…');

    let s = out;
    ellip_cache().lock().unwrap().put(key, s.clone());

    s
}

fn ellipsis_width(px: f32) -> f32 {
    static ELLIP_W_LRU: OnceCell<Mutex<Lru<u32, f32>>> = OnceCell::new();
    let cache = ELLIP_W_LRU.get_or_init(|| Mutex::new(Lru::new(64)));
    let key = (px * 100.0) as u32;
    if let Some(w) = cache.lock().unwrap().get(&key).copied() {
        return w;
    }
    let w = if let Some(g) = crate::shape_line("…", px, None).last() {
        g.x + g.advance
    } else {
        0.0
    };
    cache.lock().unwrap().put(key, w);
    w
}