Cargoify servo

components/gfx/text/glyph.rs

@@ -0,0 +1,752 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+use servo_util::vec::*;
+use servo_util::range;
+use servo_util::range::{Range, RangeIndex, IntRangeIndex, EachIndex};
+use servo_util::geometry::Au;
+
+use std::cmp::{PartialOrd, PartialEq};
+use std::num::{NumCast, Zero};
+use std::mem;
+use std::u16;
+use std::vec::Vec;
+use geom::point::Point2D;
+
+/// GlyphEntry is a port of Gecko's CompressedGlyph scheme for storing glyph data compactly.
+///
+/// In the common case (reasonable glyph advances, no offsets from the font em-box, and one glyph
+/// per character), we pack glyph advance, glyph id, and some flags into a single u32.
+///
+/// In the uncommon case (multiple glyphs per unicode character, large glyph index/advance, or
+/// glyph offsets), we pack the glyph count into GlyphEntry, and store the other glyph information
+/// in DetailedGlyphStore.
+#[deriving(Clone)]
+struct GlyphEntry {
+    value: u32,
+}
+
+impl GlyphEntry {
+    fn new(value: u32) -> GlyphEntry {
+        GlyphEntry {
+            value: value,
+        }
+    }
+
+    fn initial() -> GlyphEntry {
+        GlyphEntry::new(0)
+    }
+
+    // Creates a GlyphEntry for the common case
+    fn simple(id: GlyphId, advance: Au) -> GlyphEntry {
+        assert!(is_simple_glyph_id(id));
+        assert!(is_simple_advance(advance));
+
+        let id_mask = id as u32;
+        let Au(advance) = advance;
+        let advance_mask = (advance as u32) << GLYPH_ADVANCE_SHIFT as uint;
+
+        GlyphEntry::new(id_mask | advance_mask | FLAG_IS_SIMPLE_GLYPH)
+    }
+
+    // Create a GlyphEntry for uncommon case; should be accompanied by
+    // initialization of the actual DetailedGlyph data in DetailedGlyphStore
+    fn complex(starts_cluster: bool, starts_ligature: bool, glyph_count: int) -> GlyphEntry {
+        assert!(glyph_count <= u16::MAX as int);
+
+        debug!("creating complex glyph entry: starts_cluster={}, starts_ligature={}, \
+                glyph_count={}",
+               starts_cluster,
+               starts_ligature,
+               glyph_count);
+
+        let mut val = FLAG_NOT_MISSING;
+
+        if !starts_cluster {
+            val |= FLAG_NOT_CLUSTER_START;
+        }
+        if !starts_ligature {
+            val |= FLAG_NOT_LIGATURE_GROUP_START;
+        }
+        val |= (glyph_count as u32) << GLYPH_COUNT_SHIFT as uint;
+
+        GlyphEntry::new(val)
+    }
+
+    /// Create a GlyphEntry for the case where glyphs couldn't be found for the specified
+    /// character.
+    fn missing(glyph_count: int) -> GlyphEntry {
+        assert!(glyph_count <= u16::MAX as int);
+
+        GlyphEntry::new((glyph_count as u32) << GLYPH_COUNT_SHIFT as uint)
+    }
+}
+
+/// The id of a particular glyph within a font
+pub type GlyphId = u32;
+
+// TODO: unify with bit flags?
+#[deriving(PartialEq)]
+pub enum BreakType {
+    BreakTypeNone,
+    BreakTypeNormal,
+    BreakTypeHyphen,
+}
+
+static BREAK_TYPE_NONE:   u8 = 0x0;
+static BREAK_TYPE_NORMAL: u8 = 0x1;
+static BREAK_TYPE_HYPHEN: u8 = 0x2;
+
+fn break_flag_to_enum(flag: u8) -> BreakType {
+    if (flag & BREAK_TYPE_NORMAL) != 0 {
+        BreakTypeNormal
+    } else if (flag & BREAK_TYPE_HYPHEN) != 0 {
+        BreakTypeHyphen
+    } else {
+        BreakTypeNone
+    }
+}
+
+fn break_enum_to_flag(e: BreakType) -> u8 {
+    match e {
+        BreakTypeNone   => BREAK_TYPE_NONE,
+        BreakTypeNormal => BREAK_TYPE_NORMAL,
+        BreakTypeHyphen => BREAK_TYPE_HYPHEN,
+    }
+}
+
+// TODO: make this more type-safe.
+
+static FLAG_CHAR_IS_SPACE: u32      = 0x10000000;
+// These two bits store some BREAK_TYPE_* flags
+static FLAG_CAN_BREAK_MASK: u32     = 0x60000000;
+static FLAG_CAN_BREAK_SHIFT: u32    = 29;
+static FLAG_IS_SIMPLE_GLYPH: u32    = 0x80000000;
+
+// glyph advance; in Au's.
+static GLYPH_ADVANCE_MASK: u32      = 0x0FFF0000;
+static GLYPH_ADVANCE_SHIFT: u32     = 16;
+static GLYPH_ID_MASK: u32           = 0x0000FFFF;
+
+// Non-simple glyphs (more than one glyph per char; missing glyph,
+// newline, tab, large advance, or nonzero x/y offsets) may have one
+// or more detailed glyphs associated with them. They are stored in a
+// side array so that there is a 1:1 mapping of GlyphEntry to
+// unicode char.
+
+// The number of detailed glyphs for this char. If the char couldn't
+// be mapped to a glyph (!FLAG_NOT_MISSING), then this actually holds
+// the UTF8 code point instead.
+static GLYPH_COUNT_MASK:              u32 = 0x00FFFF00;
+static GLYPH_COUNT_SHIFT:             u32 = 8;
+// N.B. following Gecko, these are all inverted so that a lot of
+// missing chars can be memset with zeros in one fell swoop.
+static FLAG_NOT_MISSING:              u32 = 0x00000001;
+static FLAG_NOT_CLUSTER_START:        u32 = 0x00000002;
+static FLAG_NOT_LIGATURE_GROUP_START: u32 = 0x00000004;
+
+static FLAG_CHAR_IS_TAB:              u32 = 0x00000008;
+static FLAG_CHAR_IS_NEWLINE:          u32 = 0x00000010;
+//static FLAG_CHAR_IS_LOW_SURROGATE:    u32 = 0x00000020;
+//static CHAR_IDENTITY_FLAGS_MASK:      u32 = 0x00000038;
+
+fn is_simple_glyph_id(id: GlyphId) -> bool {
+    ((id as u32) & GLYPH_ID_MASK) == id
+}
+
+fn is_simple_advance(advance: Au) -> bool {
+    let unsignedAu = advance.to_u32().unwrap();
+    (unsignedAu & (GLYPH_ADVANCE_MASK >> GLYPH_ADVANCE_SHIFT as uint)) == unsignedAu
+}
+
+type DetailedGlyphCount = u16;
+
+// Getters and setters for GlyphEntry. Setter methods are functional,
+// because GlyphEntry is immutable and only a u32 in size.
+impl GlyphEntry {
+    // getter methods
+    #[inline(always)]
+    fn advance(&self) -> Au {
+        NumCast::from((self.value & GLYPH_ADVANCE_MASK) >> GLYPH_ADVANCE_SHIFT as uint).unwrap()
+    }
+
+    fn id(&self) -> GlyphId {
+        self.value & GLYPH_ID_MASK
+    }
+
+    fn is_ligature_start(&self) -> bool {
+        self.has_flag(!FLAG_NOT_LIGATURE_GROUP_START)
+    }
+
+    fn is_cluster_start(&self) -> bool {
+        self.has_flag(!FLAG_NOT_CLUSTER_START)
+    }
+
+    // True if original char was normal (U+0020) space. Other chars may
+    // map to space glyph, but this does not account for them.
+    fn char_is_space(&self) -> bool {
+        self.has_flag(FLAG_CHAR_IS_SPACE)
+    }
+
+    fn char_is_tab(&self) -> bool {
+        !self.is_simple() && self.has_flag(FLAG_CHAR_IS_TAB)
+    }
+
+    fn char_is_newline(&self) -> bool {
+        !self.is_simple() && self.has_flag(FLAG_CHAR_IS_NEWLINE)
+    }
+
+    fn can_break_before(&self) -> BreakType {
+        let flag = ((self.value & FLAG_CAN_BREAK_MASK) >> FLAG_CAN_BREAK_SHIFT as uint) as u8;
+        break_flag_to_enum(flag)
+    }
+
+    // setter methods
+    #[inline(always)]
+    fn set_char_is_space(&self) -> GlyphEntry {
+        GlyphEntry::new(self.value | FLAG_CHAR_IS_SPACE)
+    }
+
+    #[inline(always)]
+    fn set_char_is_tab(&self) -> GlyphEntry {
+        assert!(!self.is_simple());
+        GlyphEntry::new(self.value | FLAG_CHAR_IS_TAB)
+    }
+
+    #[inline(always)]
+    fn set_char_is_newline(&self) -> GlyphEntry {
+        assert!(!self.is_simple());
+        GlyphEntry::new(self.value | FLAG_CHAR_IS_NEWLINE)
+    }
+
+    #[inline(always)]
+    fn set_can_break_before(&self, e: BreakType) -> GlyphEntry {
+        let flag = (break_enum_to_flag(e) as u32) << FLAG_CAN_BREAK_SHIFT as uint;
+        GlyphEntry::new(self.value | flag)
+    }
+
+    // helper methods
+
+    fn glyph_count(&self) -> u16 {
+        assert!(!self.is_simple());
+        ((self.value & GLYPH_COUNT_MASK) >> GLYPH_COUNT_SHIFT as uint) as u16
+    }
+
+    #[inline(always)]
+    fn is_simple(&self) -> bool {
+        self.has_flag(FLAG_IS_SIMPLE_GLYPH)
+    }
+
+    #[inline(always)]
+    fn has_flag(&self, flag: u32) -> bool {
+        (self.value & flag) != 0
+    }
+
+    #[inline(always)]
+    fn adapt_character_flags_of_entry(&self, other: GlyphEntry) -> GlyphEntry {
+        GlyphEntry { value: self.value | other.value }
+    }
+}
+
+// Stores data for a detailed glyph, in the case that several glyphs
+// correspond to one character, or the glyph's data couldn't be packed.
+#[deriving(Clone)]
+struct DetailedGlyph {
+    id: GlyphId,
+    // glyph's advance, in the text's direction (RTL or RTL)
+    advance: Au,
+    // glyph's offset from the font's em-box (from top-left)
+    offset: Point2D<Au>,
+}
+
+impl DetailedGlyph {
+    fn new(id: GlyphId, advance: Au, offset: Point2D<Au>) -> DetailedGlyph {
+        DetailedGlyph {
+            id: id,
+            advance: advance,
+            offset: offset,
+        }
+    }
+}
+
+#[deriving(PartialEq, Clone, Eq)]
+struct DetailedGlyphRecord {
+    // source string offset/GlyphEntry offset in the TextRun
+    entry_offset: CharIndex,
+    // offset into the detailed glyphs buffer
+    detail_offset: int,
+}
+
+impl PartialOrd for DetailedGlyphRecord {
+    fn partial_cmp(&self, other: &DetailedGlyphRecord) -> Option<Ordering> {
+        self.entry_offset.partial_cmp(&other.entry_offset)
+    }
+}
+
+impl Ord for DetailedGlyphRecord {
+    fn cmp(&self, other: &DetailedGlyphRecord) -> Ordering {
+        self.entry_offset.cmp(&other.entry_offset)
+    }
+}
+
+// Manages the lookup table for detailed glyphs. Sorting is deferred
+// until a lookup is actually performed; this matches the expected
+// usage pattern of setting/appending all the detailed glyphs, and
+// then querying without setting.
+struct DetailedGlyphStore {
+    // TODO(pcwalton): Allocation of this buffer is expensive. Consider a small-vector
+    // optimization.
+    detail_buffer: Vec<DetailedGlyph>,
+    // TODO(pcwalton): Allocation of this buffer is expensive. Consider a small-vector
+    // optimization.
+    detail_lookup: Vec<DetailedGlyphRecord>,
+    lookup_is_sorted: bool,
+}
+
+impl<'a> DetailedGlyphStore {
+    fn new() -> DetailedGlyphStore {
+        DetailedGlyphStore {
+            detail_buffer: vec!(), // TODO: default size?
+            detail_lookup: vec!(),
+            lookup_is_sorted: false,
+        }
+    }
+
+    fn add_detailed_glyphs_for_entry(&mut self, entry_offset: CharIndex, glyphs: &[DetailedGlyph]) {
+        let entry = DetailedGlyphRecord {
+            entry_offset: entry_offset,
+            detail_offset: self.detail_buffer.len() as int,
+        };
+
+        debug!("Adding entry[off={}] for detailed glyphs: {:?}", entry_offset, glyphs);
+
+        /* TODO: don't actually assert this until asserts are compiled
+        in/out based on severity, debug/release, etc. This assertion
+        would wreck the complexity of the lookup.
+
+        See Rust Issue #3647, #2228, #3627 for related information.
+
+        do self.detail_lookup.borrow |arr| {
+            assert !arr.contains(entry)
+        }
+        */
+
+        self.detail_lookup.push(entry);
+        self.detail_buffer.push_all(glyphs);
+        self.lookup_is_sorted = false;
+    }
+
+    fn get_detailed_glyphs_for_entry(&'a self, entry_offset: CharIndex, count: u16)
+                                  -> &'a [DetailedGlyph] {
+        debug!("Requesting detailed glyphs[n={}] for entry[off={}]", count, entry_offset);
+
+        // FIXME: Is this right? --pcwalton
+        // TODO: should fix this somewhere else
+        if count == 0 {
+            return self.detail_buffer.slice(0, 0);
+        }
+
+        assert!((count as uint) <= self.detail_buffer.len());
+        assert!(self.lookup_is_sorted);
+
+        let key = DetailedGlyphRecord {
+            entry_offset: entry_offset,
+            detail_offset: 0, // unused
+        };
+
+        let i = self.detail_lookup.as_slice().binary_search_index(&key)
+            .expect("Invalid index not found in detailed glyph lookup table!");
+
+        assert!(i + (count as uint) <= self.detail_buffer.len());
+        // return a slice into the buffer
+        self.detail_buffer.slice(i, i + count as uint)
+    }
+
+    fn get_detailed_glyph_with_index(&'a self,
+                                     entry_offset: CharIndex,
+                                     detail_offset: u16)
+            -> &'a DetailedGlyph {
+        assert!((detail_offset as uint) <= self.detail_buffer.len());
+        assert!(self.lookup_is_sorted);
+
+        let key = DetailedGlyphRecord {
+            entry_offset: entry_offset,
+            detail_offset: 0, // unused
+        };
+
+        let i = self.detail_lookup.as_slice().binary_search_index(&key)
+            .expect("Invalid index not found in detailed glyph lookup table!");
+
+        assert!(i + (detail_offset as uint) < self.detail_buffer.len());
+        &self.detail_buffer[i + (detail_offset as uint)]
+    }
+
+    fn ensure_sorted(&mut self) {
+        if self.lookup_is_sorted {
+            return;
+        }
+
+        // Sorting a unique vector is surprisingly hard. The follwing
+        // code is a good argument for using DVecs, but they require
+        // immutable locations thus don't play well with freezing.
+
+        // Thar be dragons here. You have been warned. (Tips accepted.)
+        let mut unsorted_records: Vec<DetailedGlyphRecord> = vec!();
+        mem::swap(&mut self.detail_lookup, &mut unsorted_records);
+        let mut mut_records : Vec<DetailedGlyphRecord> = unsorted_records;
+        mut_records.sort_by(|a, b| {
+            if a < b {
+                Less
+            } else {
+                Greater
+            }
+        });
+        let mut sorted_records = mut_records;
+        mem::swap(&mut self.detail_lookup, &mut sorted_records);
+
+        self.lookup_is_sorted = true;
+    }
+}
+
+// This struct is used by GlyphStore clients to provide new glyph data.
+// It should be allocated on the stack and passed by reference to GlyphStore.
+pub struct GlyphData {
+    id: GlyphId,
+    advance: Au,
+    offset: Point2D<Au>,
+    is_missing: bool,
+    cluster_start: bool,
+    ligature_start: bool,
+}
+
+impl GlyphData {
+    pub fn new(id: GlyphId,
+               advance: Au,
+               offset: Option<Point2D<Au>>,
+               is_missing: bool,
+               cluster_start: bool,
+               ligature_start: bool)
+            -> GlyphData {
+        GlyphData {
+            id: id,
+            advance: advance,
+            offset: offset.unwrap_or(Zero::zero()),
+            is_missing: is_missing,
+            cluster_start: cluster_start,
+            ligature_start: ligature_start,
+        }
+    }
+}
+
+// This enum is a proxy that's provided to GlyphStore clients when iterating
+// through glyphs (either for a particular TextRun offset, or all glyphs).
+// Rather than eagerly assembling and copying glyph data, it only retrieves
+// values as they are needed from the GlyphStore, using provided offsets.
+pub enum GlyphInfo<'a> {
+    SimpleGlyphInfo(&'a GlyphStore, CharIndex),
+    DetailGlyphInfo(&'a GlyphStore, CharIndex, u16),
+}
+
+impl<'a> GlyphInfo<'a> {
+    pub fn id(self) -> GlyphId {
+        match self {
+            SimpleGlyphInfo(store, entry_i) => store.entry_buffer[entry_i.to_uint()].id(),
+            DetailGlyphInfo(store, entry_i, detail_j) => {
+                store.detail_store.get_detailed_glyph_with_index(entry_i, detail_j).id
+            }
+        }
+    }
+
+    #[inline(always)]
+    // FIXME: Resolution conflicts with IteratorUtil trait so adding trailing _
+    pub fn advance(self) -> Au {
+        match self {
+            SimpleGlyphInfo(store, entry_i) => store.entry_buffer[entry_i.to_uint()].advance(),
+            DetailGlyphInfo(store, entry_i, detail_j) => {
+                store.detail_store.get_detailed_glyph_with_index(entry_i, detail_j).advance
+            }
+        }
+    }
+
+    pub fn offset(self) -> Option<Point2D<Au>> {
+        match self {
+            SimpleGlyphInfo(_, _) => None,
+            DetailGlyphInfo(store, entry_i, detail_j) => {
+                Some(store.detail_store.get_detailed_glyph_with_index(entry_i, detail_j).offset)
+            }
+        }
+    }
+}
+
+/// Stores the glyph data belonging to a text run.
+///
+/// Simple glyphs are stored inline in the `entry_buffer`, detailed glyphs are
+/// stored as pointers into the `detail_store`.
+///
+/// ~~~
+/// +- GlyphStore --------------------------------+
+/// |               +---+---+---+---+---+---+---+ |
+/// | entry_buffer: |   | s |   | s |   | s | s | |  d = detailed
+/// |               +-|-+---+-|-+---+-|-+---+---+ |  s = simple
+/// |                 |       |       |           |
+/// |                 |   +---+-------+           |
+/// |                 |   |                       |
+/// |               +-V-+-V-+                     |
+/// | detail_store: | d | d |                     |
+/// |               +---+---+                     |
+/// +---------------------------------------------+
+/// ~~~
+pub struct GlyphStore {
+    // TODO(pcwalton): Allocation of this buffer is expensive. Consider a small-vector
+    // optimization.
+    /// A buffer of glyphs within the text run, in the order in which they
+    /// appear in the input text
+    entry_buffer: Vec<GlyphEntry>,
+    /// A store of the detailed glyph data. Detailed glyphs contained in the
+    /// `entry_buffer` point to locations in this data structure.
+    detail_store: DetailedGlyphStore,
+
+    is_whitespace: bool,
+}
+
+int_range_index! {
+    #[deriving(Encodable)]
+    #[doc = "An index that refers to a character in a text run. This could \
+             point to the middle of a glyph."]
+    struct CharIndex(int)
+}
+
+impl<'a> GlyphStore {
+    // Initializes the glyph store, but doesn't actually shape anything.
+    // Use the set_glyph, set_glyphs() methods to store glyph data.
+    pub fn new(length: int, is_whitespace: bool) -> GlyphStore {
+        assert!(length > 0);
+
+        GlyphStore {
+            entry_buffer: Vec::from_elem(length as uint, GlyphEntry::initial()),
+            detail_store: DetailedGlyphStore::new(),
+            is_whitespace: is_whitespace,
+        }
+    }
+
+    pub fn char_len(&self) -> CharIndex {
+        CharIndex(self.entry_buffer.len() as int)
+    }
+
+    pub fn is_whitespace(&self) -> bool {
+        self.is_whitespace
+    }
+
+    pub fn finalize_changes(&mut self) {
+        self.detail_store.ensure_sorted();
+    }
+
+    pub fn add_glyph_for_char_index(&mut self, i: CharIndex, data: &GlyphData) {
+        fn glyph_is_compressible(data: &GlyphData) -> bool {
+            is_simple_glyph_id(data.id)
+                && is_simple_advance(data.advance)
+                && data.offset.is_zero()
+                && data.cluster_start  // others are stored in detail buffer
+        }
+
+        assert!(data.ligature_start); // can't compress ligature continuation glyphs.
+        assert!(i < self.char_len());
+
+        let entry = match (data.is_missing, glyph_is_compressible(data)) {
+            (true, _) => GlyphEntry::missing(1),
+            (false, true) => GlyphEntry::simple(data.id, data.advance),
+            (false, false) => {
+                let glyph = [DetailedGlyph::new(data.id, data.advance, data.offset)];
+                self.detail_store.add_detailed_glyphs_for_entry(i, glyph);
+                GlyphEntry::complex(data.cluster_start, data.ligature_start, 1)
+            }
+        }.adapt_character_flags_of_entry(self.entry_buffer[i.to_uint()]);
+
+        *self.entry_buffer.get_mut(i.to_uint()) = entry;
+    }
+
+    pub fn add_glyphs_for_char_index(&mut self, i: CharIndex, data_for_glyphs: &[GlyphData]) {
+        assert!(i < self.char_len());
+        assert!(data_for_glyphs.len() > 0);
+
+        let glyph_count = data_for_glyphs.len() as int;
+
+        let first_glyph_data = data_for_glyphs[0];
+        let entry = match first_glyph_data.is_missing {
+            true  => GlyphEntry::missing(glyph_count),
+            false => {
+                let glyphs_vec = Vec::from_fn(glyph_count as uint, |i| {
+                    DetailedGlyph::new(data_for_glyphs[i].id,
+                                       data_for_glyphs[i].advance,
+                                       data_for_glyphs[i].offset)
+                });
+
+                self.detail_store.add_detailed_glyphs_for_entry(i, glyphs_vec.as_slice());
+                GlyphEntry::complex(first_glyph_data.cluster_start,
+                                    first_glyph_data.ligature_start,
+                                    glyph_count)
+            }
+        }.adapt_character_flags_of_entry(self.entry_buffer[i.to_uint()]);
+
+        debug!("Adding multiple glyphs[idx={}, count={}]: {:?}", i, glyph_count, entry);
+
+        *self.entry_buffer.get_mut(i.to_uint()) = entry;
+    }
+
+    // used when a character index has no associated glyph---for example, a ligature continuation.
+    pub fn add_nonglyph_for_char_index(&mut self, i: CharIndex, cluster_start: bool, ligature_start: bool) {
+        assert!(i < self.char_len());
+
+        let entry = GlyphEntry::complex(cluster_start, ligature_start, 0);
+        debug!("adding spacer for chracter without associated glyph[idx={}]", i);
+
+        *self.entry_buffer.get_mut(i.to_uint()) = entry;
+    }
+
+    pub fn iter_glyphs_for_char_index(&'a self, i: CharIndex) -> GlyphIterator<'a> {
+        self.iter_glyphs_for_char_range(&Range::new(i, CharIndex(1)))
+    }
+
+    #[inline]
+    pub fn iter_glyphs_for_char_range(&'a self, rang: &Range<CharIndex>) -> GlyphIterator<'a> {
+        if rang.begin() >= self.char_len() {
+            fail!("iter_glyphs_for_range: range.begin beyond length!");
+        }
+        if rang.end() > self.char_len() {
+            fail!("iter_glyphs_for_range: range.end beyond length!");
+        }
+
+        GlyphIterator {
+            store:       self,
+            char_index:  rang.begin(),
+            char_range:  rang.each_index(),
+            glyph_range: None,
+        }
+    }
+
+    #[inline]
+    pub fn advance_for_char_range(&self, rang: &Range<CharIndex>) -> Au {
+        self.iter_glyphs_for_char_range(rang)
+            .fold(Au(0), |advance, (_, glyph)| advance + glyph.advance())
+    }
+
+    // getter methods
+    pub fn char_is_space(&self, i: CharIndex) -> bool {
+        assert!(i < self.char_len());
+        self.entry_buffer[i.to_uint()].char_is_space()
+    }
+
+    pub fn char_is_tab(&self, i: CharIndex) -> bool {
+        assert!(i < self.char_len());
+        self.entry_buffer[i.to_uint()].char_is_tab()
+    }
+
+    pub fn char_is_newline(&self, i: CharIndex) -> bool {
+        assert!(i < self.char_len());
+        self.entry_buffer[i.to_uint()].char_is_newline()
+    }
+
+    pub fn is_ligature_start(&self, i: CharIndex) -> bool {
+        assert!(i < self.char_len());
+        self.entry_buffer[i.to_uint()].is_ligature_start()
+    }
+
+    pub fn is_cluster_start(&self, i: CharIndex) -> bool {
+        assert!(i < self.char_len());
+        self.entry_buffer[i.to_uint()].is_cluster_start()
+    }
+
+    pub fn can_break_before(&self, i: CharIndex) -> BreakType {
+        assert!(i < self.char_len());
+        self.entry_buffer[i.to_uint()].can_break_before()
+    }
+
+    // setter methods
+    pub fn set_char_is_space(&mut self, i: CharIndex) {
+        assert!(i < self.char_len());
+        let entry = self.entry_buffer[i.to_uint()];
+        *self.entry_buffer.get_mut(i.to_uint()) = entry.set_char_is_space();
+    }
+
+    pub fn set_char_is_tab(&mut self, i: CharIndex) {
+        assert!(i < self.char_len());
+        let entry = self.entry_buffer[i.to_uint()];
+        *self.entry_buffer.get_mut(i.to_uint()) = entry.set_char_is_tab();
+    }
+
+    pub fn set_char_is_newline(&mut self, i: CharIndex) {
+        assert!(i < self.char_len());
+        let entry = self.entry_buffer[i.to_uint()];
+        *self.entry_buffer.get_mut(i.to_uint()) = entry.set_char_is_newline();
+    }
+
+    pub fn set_can_break_before(&mut self, i: CharIndex, t: BreakType) {
+        assert!(i < self.char_len());
+        let entry = self.entry_buffer[i.to_uint()];
+        *self.entry_buffer.get_mut(i.to_uint()) = entry.set_can_break_before(t);
+    }
+}
+
+/// An iterator over the glyphs in a character range in a `GlyphStore`.
+pub struct GlyphIterator<'a> {
+    store:       &'a GlyphStore,
+    char_index:  CharIndex,
+    char_range:  EachIndex<int, CharIndex>,
+    glyph_range: Option<EachIndex<int, CharIndex>>,
+}
+
+impl<'a> GlyphIterator<'a> {
+    // Slow path when there is a glyph range.
+    #[inline(never)]
+    fn next_glyph_range(&mut self) -> Option<(CharIndex, GlyphInfo<'a>)> {
+        match self.glyph_range.get_mut_ref().next() {
+            Some(j) => Some((self.char_index,
+                DetailGlyphInfo(self.store, self.char_index, j.get() as u16 /* ??? */))),
+            None => {
+                // No more glyphs for current character.  Try to get another.
+                self.glyph_range = None;
+                self.next()
+            }
+        }
+    }
+
+    // Slow path when there is a complex glyph.
+    #[inline(never)]
+    fn next_complex_glyph(&mut self, entry: &GlyphEntry, i: CharIndex)
+                          -> Option<(CharIndex, GlyphInfo<'a>)> {
+        let glyphs = self.store.detail_store.get_detailed_glyphs_for_entry(i, entry.glyph_count());
+        self.glyph_range = Some(range::each_index(CharIndex(0), CharIndex(glyphs.len() as int)));
+        self.next()
+    }
+}
+
+impl<'a> Iterator<(CharIndex, GlyphInfo<'a>)> for GlyphIterator<'a> {
+    // I tried to start with something simpler and apply FlatMap, but the
+    // inability to store free variables in the FlatMap struct was problematic.
+    //
+    // This function consists of the fast path and is designed to be inlined into its caller. The
+    // slow paths, which should not be inlined, are `next_glyph_range()` and
+    // `next_complex_glyph()`.
+    #[inline(always)]
+    fn next(&mut self) -> Option<(CharIndex, GlyphInfo<'a>)> {
+        // Would use 'match' here but it borrows contents in a way that
+        // interferes with mutation.
+        if self.glyph_range.is_some() {
+            self.next_glyph_range()
+        } else {
+            // No glyph range. Look at next character.
+            self.char_range.next().and_then(|i| {
+                self.char_index = i;
+                assert!(i < self.store.char_len());
+                let entry = self.store.entry_buffer[i.to_uint()];
+                if entry.is_simple() {
+                    Some((self.char_index, SimpleGlyphInfo(self.store, i)))
+                } else {
+                    // Fall back to the slow path.
+                    self.next_complex_glyph(&entry, i)
+                }
+            })
+        }
+    }
+}

components/gfx/text/mod.rs

@@ -0,0 +1,18 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+/* This file exists just to make it easier to import things inside of
+ ./text/ without specifying the file they came out of imports.
+
+Note that you still must define each of the files as a module in
+servo.rc. This is not ideal and may be changed in the future. */
+
+pub use text::shaping::Shaper;
+pub use text::text_run::TextRun;
+
+pub mod glyph;
+#[path="shaping/mod.rs"] pub mod shaping;
+pub mod text_run;
+pub mod util;
+

components/gfx/text/shaping/harfbuzz.rs

@@ -0,0 +1,541 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+extern crate harfbuzz;
+
+use font::{Font, FontHandleMethods, FontTableMethods, FontTableTag};
+use platform::font::FontTable;
+use text::glyph::{CharIndex, GlyphStore, GlyphId, GlyphData};
+use text::shaping::ShaperMethods;
+use text::util::{float_to_fixed, fixed_to_float};
+
+use geom::Point2D;
+use harfbuzz::{HB_MEMORY_MODE_READONLY, HB_DIRECTION_LTR};
+use harfbuzz::{hb_blob_create, hb_face_create_for_tables};
+use harfbuzz::{hb_blob_t};
+use harfbuzz::{hb_bool_t};
+use harfbuzz::{hb_buffer_add_utf8};
+use harfbuzz::{hb_buffer_destroy};
+use harfbuzz::{hb_buffer_get_glyph_positions};
+use harfbuzz::{hb_buffer_set_direction};
+use harfbuzz::{hb_face_destroy};
+use harfbuzz::{hb_face_t, hb_font_t};
+use harfbuzz::{hb_font_create};
+use harfbuzz::{hb_font_destroy, hb_buffer_create};
+use harfbuzz::{hb_font_funcs_create};
+use harfbuzz::{hb_font_funcs_destroy};
+use harfbuzz::{hb_font_funcs_set_glyph_func};
+use harfbuzz::{hb_font_funcs_set_glyph_h_advance_func};
+use harfbuzz::{hb_font_funcs_set_glyph_h_kerning_func};
+use harfbuzz::{hb_font_funcs_t, hb_buffer_t, hb_codepoint_t};
+use harfbuzz::{hb_font_set_funcs};
+use harfbuzz::{hb_font_set_ppem};
+use harfbuzz::{hb_font_set_scale};
+use harfbuzz::{hb_glyph_info_t};
+use harfbuzz::{hb_glyph_position_t};
+use harfbuzz::{hb_position_t, hb_tag_t};
+use harfbuzz::{hb_shape, hb_buffer_get_glyph_infos};
+use libc::{c_uint, c_int, c_void, c_char};
+use servo_util::geometry::Au;
+use servo_util::range::Range;
+use std::mem;
+use std::char;
+use std::cmp;
+use std::ptr;
+
+static NO_GLYPH: i32 = -1;
+static CONTINUATION_BYTE: i32 = -2;
+
+pub struct ShapedGlyphData {
+    count: int,
+    glyph_infos: *mut hb_glyph_info_t,
+    pos_infos: *mut hb_glyph_position_t,
+}
+
+pub struct ShapedGlyphEntry {
+    codepoint: GlyphId,
+    advance: Au,
+    offset: Option<Point2D<Au>>,
+}
+
+impl ShapedGlyphData {
+    pub fn new(buffer: *mut hb_buffer_t) -> ShapedGlyphData {
+        unsafe {
+            let mut glyph_count = 0;
+            let glyph_infos = hb_buffer_get_glyph_infos(buffer, &mut glyph_count);
+            let glyph_count = glyph_count as int;
+            assert!(glyph_infos.is_not_null());
+            let mut pos_count = 0;
+            let pos_infos = hb_buffer_get_glyph_positions(buffer, &mut pos_count);
+            let pos_count = pos_count as int;
+            assert!(pos_infos.is_not_null());
+            assert!(glyph_count == pos_count);
+
+            ShapedGlyphData {
+                count: glyph_count,
+                glyph_infos: glyph_infos,
+                pos_infos: pos_infos,
+            }
+        }
+    }
+
+    #[inline(always)]
+    fn byte_offset_of_glyph(&self, i: int) -> int {
+        assert!(i < self.count);
+
+        unsafe {
+            let glyph_info_i = self.glyph_infos.offset(i);
+            (*glyph_info_i).cluster as int
+        }
+    }
+
+    pub fn len(&self) -> int {
+        self.count
+    }
+
+    /// Returns shaped glyph data for one glyph, and updates the y-position of the pen.
+    pub fn get_entry_for_glyph(&self, i: int, y_pos: &mut Au) -> ShapedGlyphEntry {
+        assert!(i < self.count);
+
+        unsafe {
+            let glyph_info_i = self.glyph_infos.offset(i);
+            let pos_info_i = self.pos_infos.offset(i);
+            let x_offset = Shaper::fixed_to_float((*pos_info_i).x_offset);
+            let y_offset = Shaper::fixed_to_float((*pos_info_i).y_offset);
+            let x_advance = Shaper::fixed_to_float((*pos_info_i).x_advance);
+            let y_advance = Shaper::fixed_to_float((*pos_info_i).y_advance);
+
+            let x_offset = Au::from_frac_px(x_offset);
+            let y_offset = Au::from_frac_px(y_offset);
+            let x_advance = Au::from_frac_px(x_advance);
+            let y_advance = Au::from_frac_px(y_advance);
+
+            let offset = if x_offset == Au(0) && y_offset == Au(0) && y_advance == Au(0) {
+                None
+            } else {
+                // adjust the pen..
+                if y_advance > Au(0) {
+                    *y_pos = *y_pos - y_advance;
+                }
+
+                Some(Point2D(x_offset, *y_pos - y_offset))
+            };
+
+            ShapedGlyphEntry {
+                codepoint: (*glyph_info_i).codepoint as GlyphId,
+                advance: x_advance,
+                offset: offset,
+            }
+        }
+    }
+}
+
+pub struct Shaper {
+    hb_face: *mut hb_face_t,
+    hb_font: *mut hb_font_t,
+    hb_funcs: *mut hb_font_funcs_t,
+}
+
+#[unsafe_destructor]
+impl Drop for Shaper {
+    fn drop(&mut self) {
+        unsafe {
+            assert!(self.hb_face.is_not_null());
+            hb_face_destroy(self.hb_face);
+
+            assert!(self.hb_font.is_not_null());
+            hb_font_destroy(self.hb_font);
+
+            assert!(self.hb_funcs.is_not_null());
+            hb_font_funcs_destroy(self.hb_funcs);
+        }
+    }
+}
+
+impl Shaper {
+    pub fn new(font: &mut Font) -> Shaper {
+        unsafe {
+            // Indirection for Rust Issue #6248, dynamic freeze scope artifically extended
+            let font_ptr = font as *mut Font;
+            let hb_face: *mut hb_face_t = hb_face_create_for_tables(get_font_table_func,
+                                                                    font_ptr as *mut c_void,
+                                                                    None);
+            let hb_font: *mut hb_font_t = hb_font_create(hb_face);
+
+            // Set points-per-em. if zero, performs no hinting in that direction.
+            let pt_size = font.pt_size;
+            hb_font_set_ppem(hb_font, pt_size as c_uint, pt_size as c_uint);
+
+            // Set scaling. Note that this takes 16.16 fixed point.
+            hb_font_set_scale(hb_font,
+                              Shaper::float_to_fixed(pt_size) as c_int,
+                              Shaper::float_to_fixed(pt_size) as c_int);
+
+            // configure static function callbacks.
+            // NB. This funcs structure could be reused globally, as it never changes.
+            let hb_funcs: *mut hb_font_funcs_t = hb_font_funcs_create();
+            hb_font_funcs_set_glyph_func(hb_funcs, glyph_func, ptr::mut_null(), None);
+            hb_font_funcs_set_glyph_h_advance_func(hb_funcs, glyph_h_advance_func, ptr::mut_null(), None);
+            hb_font_funcs_set_glyph_h_kerning_func(hb_funcs, glyph_h_kerning_func, ptr::mut_null(), ptr::mut_null());
+            hb_font_set_funcs(hb_font, hb_funcs, font_ptr as *mut c_void, None);
+
+            Shaper {
+                hb_face: hb_face,
+                hb_font: hb_font,
+                hb_funcs: hb_funcs,
+            }
+        }
+    }
+
+    fn float_to_fixed(f: f64) -> i32 {
+        float_to_fixed(16, f)
+    }
+
+    fn fixed_to_float(i: hb_position_t) -> f64 {
+        fixed_to_float(16, i)
+    }
+}
+
+impl ShaperMethods for Shaper {
+    /// Calculate the layout metrics associated with the given text when rendered in a specific
+    /// font.
+    fn shape_text(&self, text: &str, glyphs: &mut GlyphStore) {
+        unsafe {
+            let hb_buffer: *mut hb_buffer_t = hb_buffer_create();
+            hb_buffer_set_direction(hb_buffer, HB_DIRECTION_LTR);
+
+            hb_buffer_add_utf8(hb_buffer,
+                               text.as_ptr() as *const c_char,
+                               text.len() as c_int,
+                               0,
+                               text.len() as c_int);
+
+            hb_shape(self.hb_font, hb_buffer, ptr::mut_null(), 0);
+            self.save_glyph_results(text, glyphs, hb_buffer);
+            hb_buffer_destroy(hb_buffer);
+        }
+    }
+}
+
+impl Shaper {
+    fn save_glyph_results(&self, text: &str, glyphs: &mut GlyphStore, buffer: *mut hb_buffer_t) {
+        let glyph_data = ShapedGlyphData::new(buffer);
+        let glyph_count = glyph_data.len();
+        let byte_max = text.len() as int;
+        let char_max = text.char_len() as int;
+
+        // GlyphStore records are indexed by character, not byte offset.
+        // so, we must be careful to increment this when saving glyph entries.
+        let mut char_idx = CharIndex(0);
+
+        assert!(glyph_count <= char_max);
+
+        debug!("Shaped text[char count={}], got back {} glyph info records.",
+               char_max,
+               glyph_count);
+
+        if char_max != glyph_count {
+            debug!("NOTE: Since these are not equal, we probably have been given some complex \
+                    glyphs.");
+        }
+
+        // make map of what chars have glyphs
+        let mut byteToGlyph: Vec<i32>;
+
+        // fast path: all chars are single-byte.
+        if byte_max == char_max {
+            byteToGlyph = Vec::from_elem(byte_max as uint, NO_GLYPH);
+        } else {
+            byteToGlyph = Vec::from_elem(byte_max as uint, CONTINUATION_BYTE);
+            for (i, _) in text.char_indices() {
+                *byteToGlyph.get_mut(i) = NO_GLYPH;
+            }
+        }
+
+        debug!("(glyph idx) -> (text byte offset)");
+        for i in range(0, glyph_data.len()) {
+            // loc refers to a *byte* offset within the utf8 string.
+            let loc = glyph_data.byte_offset_of_glyph(i);
+            if loc < byte_max {
+                assert!(*byteToGlyph.get(loc as uint) != CONTINUATION_BYTE);
+                *byteToGlyph.get_mut(loc as uint) = i as i32;
+            } else {
+                debug!("ERROR: tried to set out of range byteToGlyph: idx={}, glyph idx={}",
+                       loc,
+                       i);
+            }
+            debug!("{} -> {}", i, loc);
+        }
+
+        debug!("text: {:s}", text);
+        debug!("(char idx): char->(glyph index):");
+        for (i, ch) in text.char_indices() {
+            debug!("{}: {} --> {:d}", i, ch, *byteToGlyph.get(i) as int);
+        }
+
+        // some helpers
+        let mut glyph_span: Range<int> = Range::empty();
+        // this span contains first byte of first char, to last byte of last char in range.
+        // so, end() points to first byte of last+1 char, if it's less than byte_max.
+        let mut char_byte_span: Range<int> = Range::empty();
+        let mut y_pos = Au(0);
+
+        // main loop over each glyph. each iteration usually processes 1 glyph and 1+ chars.
+        // in cases with complex glyph-character assocations, 2+ glyphs and 1+ chars can be
+        // processed.
+        while glyph_span.begin() < glyph_count {
+            // start by looking at just one glyph.
+            glyph_span.extend_by(1);
+            debug!("Processing glyph at idx={}", glyph_span.begin());
+
+            let char_byte_start = glyph_data.byte_offset_of_glyph(glyph_span.begin());
+            char_byte_span.reset(char_byte_start, 0);
+
+            // find a range of chars corresponding to this glyph, plus
+            // any trailing chars that do not have associated glyphs.
+            while char_byte_span.end() < byte_max {
+                let range = text.char_range_at(char_byte_span.end() as uint);
+                drop(range.ch);
+                char_byte_span.extend_to(range.next as int);
+
+                debug!("Processing char byte span: off={}, len={} for glyph idx={}",
+                       char_byte_span.begin(), char_byte_span.length(), glyph_span.begin());
+
+                while char_byte_span.end() != byte_max &&
+                        byteToGlyph[char_byte_span.end() as uint] == NO_GLYPH {
+                    debug!("Extending char byte span to include byte offset={} with no associated \
+                            glyph", char_byte_span.end());
+                    let range = text.char_range_at(char_byte_span.end() as uint);
+                    drop(range.ch);
+                    char_byte_span.extend_to(range.next as int);
+                }
+
+                // extend glyph range to max glyph index covered by char_span,
+                // in cases where one char made several glyphs and left some unassociated chars.
+                let mut max_glyph_idx = glyph_span.end();
+                for i in char_byte_span.each_index() {
+                    if byteToGlyph[i as uint] > NO_GLYPH {
+                        max_glyph_idx = cmp::max(byteToGlyph[i as uint] as int + 1, max_glyph_idx);
+                    }
+                }
+
+                if max_glyph_idx > glyph_span.end() {
+                    glyph_span.extend_to(max_glyph_idx);
+                    debug!("Extended glyph span (off={}, len={}) to cover char byte span's max \
+                            glyph index",
+                           glyph_span.begin(), glyph_span.length());
+                }
+
+
+                // if there's just one glyph, then we don't need further checks.
+                if glyph_span.length() == 1 { break; }
+
+                // if no glyphs were found yet, extend the char byte range more.
+                if glyph_span.length() == 0 { continue; }
+
+                debug!("Complex (multi-glyph to multi-char) association found. This case \
+                        probably doesn't work.");
+
+                let mut all_glyphs_are_within_cluster: bool = true;
+                for j in glyph_span.each_index() {
+                    let loc = glyph_data.byte_offset_of_glyph(j);
+                    if !char_byte_span.contains(loc) {
+                        all_glyphs_are_within_cluster = false;
+                        break
+                    }
+                }
+
+                debug!("All glyphs within char_byte_span cluster?: {}",
+                       all_glyphs_are_within_cluster);
+
+                // found a valid range; stop extending char_span.
+                if all_glyphs_are_within_cluster {
+                    break
+                }
+            }
+
+            // character/glyph clump must contain characters.
+            assert!(char_byte_span.length() > 0);
+            // character/glyph clump must contain glyphs.
+            assert!(glyph_span.length() > 0);
+
+            // now char_span is a ligature clump, formed by the glyphs in glyph_span.
+            // we need to find the chars that correspond to actual glyphs (char_extended_span),
+            //and set glyph info for those and empty infos for the chars that are continuations.
+
+            // a simple example:
+            // chars:  'f'     't'   't'
+            // glyphs: 'ftt'   ''    ''
+            // cgmap:  t        f     f
+            // gspan:  [-]
+            // cspan:  [-]
+            // covsp:  [---------------]
+
+            let mut covered_byte_span = char_byte_span.clone();
+            // extend, clipping at end of text range.
+            while covered_byte_span.end() < byte_max
+                    && byteToGlyph[covered_byte_span.end() as uint] == NO_GLYPH {
+                let range = text.char_range_at(covered_byte_span.end() as uint);
+                drop(range.ch);
+                covered_byte_span.extend_to(range.next as int);
+            }
+
+            if covered_byte_span.begin() >= byte_max {
+                // oops, out of range. clip and forget this clump.
+                let end = glyph_span.end(); // FIXME: borrow checker workaround
+                glyph_span.reset(end, 0);
+                let end = char_byte_span.end(); // FIXME: borrow checker workaround
+                char_byte_span.reset(end, 0);
+            }
+
+            // clamp to end of text. (I don't think this will be necessary, but..)
+            let end = covered_byte_span.end(); // FIXME: borrow checker workaround
+            covered_byte_span.extend_to(cmp::min(end, byte_max));
+
+            // fast path: 1-to-1 mapping of single char and single glyph.
+            if glyph_span.length() == 1 {
+                // TODO(Issue #214): cluster ranges need to be computed before
+                // shaping, and then consulted here.
+                // for now, just pretend that every character is a cluster start.
+                // (i.e., pretend there are no combining character sequences).
+                // 1-to-1 mapping of character to glyph also treated as ligature start.
+                let shape = glyph_data.get_entry_for_glyph(glyph_span.begin(), &mut y_pos);
+                let data = GlyphData::new(shape.codepoint,
+                                          shape.advance,
+                                          shape.offset,
+                                          false,
+                                          true,
+                                          true);
+                glyphs.add_glyph_for_char_index(char_idx, &data);
+            } else {
+                // collect all glyphs to be assigned to the first character.
+                let mut datas = vec!();
+
+                for glyph_i in glyph_span.each_index() {
+                    let shape = glyph_data.get_entry_for_glyph(glyph_i, &mut y_pos);
+                    datas.push(GlyphData::new(shape.codepoint,
+                                              shape.advance,
+                                              shape.offset,
+                                              false, // not missing
+                                              true,  // treat as cluster start
+                                              glyph_i > glyph_span.begin()));
+                                              // all but first are ligature continuations
+                }
+
+                // now add the detailed glyph entry.
+                glyphs.add_glyphs_for_char_index(char_idx, datas.as_slice());
+
+                // set the other chars, who have no glyphs
+                let mut i = covered_byte_span.begin();
+                loop {
+                    let range = text.char_range_at(i as uint);
+                    drop(range.ch);
+                    i = range.next as int;
+                    if i >= covered_byte_span.end() { break; }
+                    char_idx = char_idx + CharIndex(1);
+                    glyphs.add_nonglyph_for_char_index(char_idx, false, false);
+                }
+            }
+
+            // shift up our working spans past things we just handled.
+            let end = glyph_span.end(); // FIXME: borrow checker workaround
+            glyph_span.reset(end, 0);
+            let end = char_byte_span.end();; // FIXME: borrow checker workaround
+            char_byte_span.reset(end, 0);
+            char_idx = char_idx + CharIndex(1);
+        }
+
+        // this must be called after adding all glyph data; it sorts the
+        // lookup table for finding detailed glyphs by associated char index.
+        glyphs.finalize_changes();
+    }
+}
+
+/// Callbacks from Harfbuzz when font map and glyph advance lookup needed.
+extern fn glyph_func(_: *mut hb_font_t,
+                     font_data: *mut c_void,
+                     unicode: hb_codepoint_t,
+                     _: hb_codepoint_t,
+                     glyph: *mut hb_codepoint_t,
+                     _: *mut c_void)
+                  -> hb_bool_t {
+    let font: *const Font = font_data as *const Font;
+    assert!(font.is_not_null());
+
+    unsafe {
+        match (*font).glyph_index(char::from_u32(unicode).unwrap()) {
+            Some(g) => {
+                *glyph = g as hb_codepoint_t;
+                true as hb_bool_t
+            }
+            None => false as hb_bool_t
+        }
+    }
+}
+
+extern fn glyph_h_advance_func(_: *mut hb_font_t,
+                               font_data: *mut c_void,
+                               glyph: hb_codepoint_t,
+                               _: *mut c_void)
+                            -> hb_position_t {
+    let font: *mut Font = font_data as *mut Font;
+    assert!(font.is_not_null());
+
+    unsafe {
+        let advance = (*font).glyph_h_advance(glyph as GlyphId);
+        Shaper::float_to_fixed(advance)
+    }
+}
+
+extern fn glyph_h_kerning_func(_: *mut hb_font_t,
+                               font_data: *mut c_void,
+                               first_glyph: hb_codepoint_t,
+                               second_glyph: hb_codepoint_t,
+                               _: *mut c_void)
+                            -> hb_position_t {
+    let font: *mut Font = font_data as *mut Font;
+    assert!(font.is_not_null());
+
+    unsafe {
+        let advance = (*font).glyph_h_kerning(first_glyph as GlyphId, second_glyph as GlyphId);
+        Shaper::float_to_fixed(advance)
+    }
+}
+
+// Callback to get a font table out of a font.
+extern fn get_font_table_func(_: *mut hb_face_t, tag: hb_tag_t, user_data: *mut c_void) -> *mut hb_blob_t {
+    unsafe {
+        let font: *const Font = user_data as *const Font;
+        assert!(font.is_not_null());
+
+        // TODO(Issue #197): reuse font table data, which will change the unsound trickery here.
+        match (*font).get_table_for_tag(tag as FontTableTag) {
+            None => ptr::mut_null(),
+            Some(ref font_table) => {
+                let skinny_font_table_ptr: *const FontTable = font_table;   // private context
+
+                let mut blob: *mut hb_blob_t = ptr::mut_null();
+                (*skinny_font_table_ptr).with_buffer(|buf: *const u8, len: uint| {
+                    // HarfBuzz calls `destroy_blob_func` when the buffer is no longer needed.
+                    blob = hb_blob_create(buf as *const c_char,
+                                          len as c_uint,
+                                          HB_MEMORY_MODE_READONLY,
+                                          mem::transmute(skinny_font_table_ptr),
+                                          destroy_blob_func);
+                });
+
+                assert!(blob.is_not_null());
+                blob
+            }
+        }
+    }
+}
+
+// TODO(Issue #197): reuse font table data, which will change the unsound trickery here.
+// In particular, we'll need to cast to a boxed, rather than owned, FontTable.
+
+// even better, should cache the harfbuzz blobs directly instead of recreating a lot.
+extern fn destroy_blob_func(_: *mut c_void) {
+    // TODO: Previous code here was broken. Rewrite.
+}

components/gfx/text/shaping/mod.rs

@@ -0,0 +1,19 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+//! Shaper encapsulates a specific shaper, such as Harfbuzz,
+//! Uniscribe, Pango, or Coretext.
+//!
+//! Currently, only harfbuzz bindings are implemented.
+
+use text::glyph::GlyphStore;
+
+pub use Shaper = text::shaping::harfbuzz::Shaper;
+
+pub mod harfbuzz;
+
+pub trait ShaperMethods {
+    fn shape_text(&self, text: &str, glyphs: &mut GlyphStore);
+}
+

components/gfx/text/text_run.rs

@@ -0,0 +1,271 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+use font::{Font, RunMetrics, FontMetrics};
+use servo_util::geometry::Au;
+use servo_util::range::Range;
+use servo_util::vec::{Comparator, FullBinarySearchMethods};
+use std::slice::Items;
+use sync::Arc;
+use text::glyph::{CharIndex, GlyphStore};
+use font::FontHandleMethods;
+use platform::font_template::FontTemplateData;
+
+/// A single "paragraph" of text in one font size and style.
+#[deriving(Clone)]
+pub struct TextRun {
+    pub text: Arc<String>,
+    pub font_template: Arc<FontTemplateData>,
+    pub pt_size: f64,
+    pub font_metrics: FontMetrics,
+    /// The glyph runs that make up this text run.
+    pub glyphs: Arc<Vec<GlyphRun>>,
+}
+
+/// A single series of glyphs within a text run.
+#[deriving(Clone)]
+pub struct GlyphRun {
+    /// The glyphs.
+    glyph_store: Arc<GlyphStore>,
+    /// The range of characters in the containing run.
+    range: Range<CharIndex>,
+}
+
+pub struct SliceIterator<'a> {
+    glyph_iter: Items<'a, GlyphRun>,
+    range:      Range<CharIndex>,
+}
+
+struct CharIndexComparator;
+
+impl Comparator<CharIndex,GlyphRun> for CharIndexComparator {
+    fn compare(&self, key: &CharIndex, value: &GlyphRun) -> Ordering {
+        if *key < value.range.begin() {
+            Less
+        } else if *key >= value.range.end() {
+            Greater
+        } else {
+            Equal
+        }
+    }
+}
+
+impl<'a> Iterator<(&'a GlyphStore, CharIndex, Range<CharIndex>)> for SliceIterator<'a> {
+    // inline(always) due to the inefficient rt failures messing up inline heuristics, I think.
+    #[inline(always)]
+    fn next(&mut self) -> Option<(&'a GlyphStore, CharIndex, Range<CharIndex>)> {
+        let slice_glyphs = self.glyph_iter.next();
+        if slice_glyphs.is_none() {
+            return None;
+        }
+        let slice_glyphs = slice_glyphs.unwrap();
+
+        let mut char_range = self.range.intersect(&slice_glyphs.range);
+        let slice_range_begin = slice_glyphs.range.begin();
+        char_range.shift_by(-slice_range_begin);
+        if !char_range.is_empty() {
+            return Some((&*slice_glyphs.glyph_store, slice_range_begin, char_range))
+        }
+
+        return None;
+    }
+}
+
+pub struct LineIterator<'a> {
+    range:  Range<CharIndex>,
+    clump:  Option<Range<CharIndex>>,
+    slices: SliceIterator<'a>,
+}
+
+impl<'a> Iterator<Range<CharIndex>> for LineIterator<'a> {
+    fn next(&mut self) -> Option<Range<CharIndex>> {
+        // Loop until we hit whitespace and are in a clump.
+        loop {
+            match self.slices.next() {
+                Some((glyphs, offset, slice_range)) => {
+                    match (glyphs.is_whitespace(), self.clump) {
+                        (false, Some(ref mut c))  => {
+                            c.extend_by(slice_range.length());
+                        }
+                        (false, None) => {
+                            let mut c = slice_range;
+                            c.shift_by(offset);
+                            self.clump = Some(c);
+                        }
+                        (true, None)    => { /* chomp whitespace */ }
+                        (true, Some(c)) => {
+                            self.clump = None;
+                            // The final whitespace clump is not included.
+                            return Some(c);
+                        }
+                    }
+                },
+                None => {
+                    // flush any remaining chars as a line
+                    if self.clump.is_some() {
+                        let mut c = self.clump.take_unwrap();
+                        c.extend_to(self.range.end());
+                        return Some(c);
+                    } else {
+                        return None;
+                    }
+                }
+            }
+        }
+    }
+}
+
+impl<'a> TextRun {
+    pub fn new(font: &mut Font, text: String) -> TextRun {
+        let glyphs = TextRun::break_and_shape(font, text.as_slice());
+        let run = TextRun {
+            text: Arc::new(text),
+            font_metrics: font.metrics.clone(),
+            font_template: font.handle.get_template(),
+            pt_size: font.pt_size,
+            glyphs: Arc::new(glyphs),
+        };
+        return run;
+    }
+
+    pub fn break_and_shape(font: &mut Font, text: &str) -> Vec<GlyphRun> {
+        // TODO(Issue #230): do a better job. See Gecko's LineBreaker.
+        let mut glyphs = vec!();
+        let (mut byte_i, mut char_i) = (0u, CharIndex(0));
+        let mut cur_slice_is_whitespace = false;
+        let (mut byte_last_boundary, mut char_last_boundary) = (0, CharIndex(0));
+        while byte_i < text.len() {
+            let range = text.char_range_at(byte_i);
+            let ch = range.ch;
+            let next = range.next;
+
+            // Slices alternate between whitespace and non-whitespace,
+            // representing line break opportunities.
+            let can_break_before = if cur_slice_is_whitespace {
+                match ch {
+                    ' ' | '\t' | '\n' => false,
+                    _ => {
+                        cur_slice_is_whitespace = false;
+                        true
+                    }
+                }
+            } else {
+                match ch {
+                    ' ' | '\t' | '\n' => {
+                        cur_slice_is_whitespace = true;
+                        true
+                    },
+                    _ => false
+                }
+            };
+
+            // Create a glyph store for this slice if it's nonempty.
+            if can_break_before && byte_i > byte_last_boundary {
+                let slice = text.slice(byte_last_boundary, byte_i).to_string();
+                debug!("creating glyph store for slice {} (ws? {}), {} - {} in run {}",
+                        slice, !cur_slice_is_whitespace, byte_last_boundary, byte_i, text);
+                glyphs.push(GlyphRun {
+                    glyph_store: font.shape_text(slice, !cur_slice_is_whitespace),
+                    range: Range::new(char_last_boundary, char_i - char_last_boundary),
+                });
+                byte_last_boundary = byte_i;
+                char_last_boundary = char_i;
+            }
+
+            byte_i = next;
+            char_i = char_i + CharIndex(1);
+        }
+
+        // Create a glyph store for the final slice if it's nonempty.
+        if byte_i > byte_last_boundary {
+            let slice = text.slice_from(byte_last_boundary).to_string();
+            debug!("creating glyph store for final slice {} (ws? {}), {} - {} in run {}",
+                slice, cur_slice_is_whitespace, byte_last_boundary, text.len(), text);
+            glyphs.push(GlyphRun {
+                glyph_store: font.shape_text(slice, cur_slice_is_whitespace),
+                range: Range::new(char_last_boundary, char_i - char_last_boundary),
+            });
+        }
+
+        glyphs
+    }
+
+    pub fn char_len(&self) -> CharIndex {
+        match self.glyphs.last() {
+            None => CharIndex(0),
+            Some(ref glyph_run) => glyph_run.range.end(),
+        }
+    }
+
+    pub fn glyphs(&'a self) -> &'a Vec<GlyphRun> {
+        &*self.glyphs
+    }
+
+    pub fn range_is_trimmable_whitespace(&self, range: &Range<CharIndex>) -> bool {
+        self.iter_slices_for_range(range).all(|(slice_glyphs, _, _)| {
+            slice_glyphs.is_whitespace()
+        })
+    }
+
+    pub fn ascent(&self) -> Au {
+        self.font_metrics.ascent
+    }
+
+    pub fn descent(&self) -> Au {
+        self.font_metrics.descent
+    }
+
+    pub fn advance_for_range(&self, range: &Range<CharIndex>) -> Au {
+        // TODO(Issue #199): alter advance direction for RTL
+        // TODO(Issue #98): using inter-char and inter-word spacing settings  when measuring text
+        self.iter_slices_for_range(range)
+            .fold(Au(0), |advance, (glyphs, _, slice_range)| {
+                advance + glyphs.advance_for_char_range(&slice_range)
+            })
+    }
+
+    pub fn metrics_for_range(&self, range: &Range<CharIndex>) -> RunMetrics {
+        RunMetrics::new(self.advance_for_range(range),
+                        self.font_metrics.ascent,
+                        self.font_metrics.descent)
+    }
+
+    pub fn metrics_for_slice(&self, glyphs: &GlyphStore, slice_range: &Range<CharIndex>) -> RunMetrics {
+        RunMetrics::new(glyphs.advance_for_char_range(slice_range),
+                        self.font_metrics.ascent,
+                        self.font_metrics.descent)
+    }
+
+    pub fn min_width_for_range(&self, range: &Range<CharIndex>) -> Au {
+        debug!("iterating outer range {:?}", range);
+        self.iter_slices_for_range(range).fold(Au(0), |max_piece_width, (_, offset, slice_range)| {
+            debug!("iterated on {:?}[{:?}]", offset, slice_range);
+            Au::max(max_piece_width, self.advance_for_range(&slice_range))
+        })
+    }
+
+    /// Returns the index of the first glyph run containing the given character index.
+    fn index_of_first_glyph_run_containing(&self, index: CharIndex) -> Option<uint> {
+        self.glyphs.as_slice().binary_search_index_by(&index, CharIndexComparator)
+    }
+
+    pub fn iter_slices_for_range(&'a self, range: &Range<CharIndex>) -> SliceIterator<'a> {
+        let index = match self.index_of_first_glyph_run_containing(range.begin()) {
+            None => self.glyphs.len(),
+            Some(index) => index,
+        };
+        SliceIterator {
+            glyph_iter: self.glyphs.slice_from(index).iter(),
+            range:      *range,
+        }
+    }
+
+    pub fn iter_natural_lines_for_range(&'a self, range: &Range<CharIndex>) -> LineIterator<'a> {
+        LineIterator {
+            range:  *range,
+            clump:  None,
+            slices: self.iter_slices_for_range(range),
+        }
+    }
+}

components/gfx/text/util.rs

@@ -0,0 +1,285 @@
+/* This Source Code Form is subject to the terms of the Mozilla Public
+ * License, v. 2.0. If a copy of the MPL was not distributed with this
+ * file, You can obtain one at http://mozilla.org/MPL/2.0/. */
+
+use text::glyph::CharIndex;
+
+#[deriving(PartialEq)]
+pub enum CompressionMode {
+    CompressNone,
+    CompressWhitespace,
+    CompressWhitespaceNewline,
+    DiscardNewline
+}
+
+// ported from Gecko's nsTextFrameUtils::TransformText.
+//
+// High level TODOs:
+//
+// * Issue #113: consider incoming text state (arabic, etc)
+//               and propogate outgoing text state (dual of above)
+//
+// * Issue #114: record skipped and kept chars for mapping original to new text
+//
+// * Untracked: various edge cases for bidi, CJK, etc.
+pub fn transform_text(text: &str, mode: CompressionMode,
+                      incoming_whitespace: bool,
+                      new_line_pos: &mut Vec<CharIndex>) -> (String, bool) {
+    let mut out_str = String::new();
+    let out_whitespace = match mode {
+        CompressNone | DiscardNewline => {
+            let mut new_line_index = CharIndex(0);
+            for ch in text.chars() {
+                if is_discardable_char(ch, mode) {
+                    // TODO: record skipped char
+                } else {
+                    // TODO: record kept char
+                    if ch == '\t' {
+                        // TODO: set "has tab" flag
+                    } else if ch == '\n' {
+                        // Save new-line's position for line-break
+                        // This value is relative(not absolute)
+                        new_line_pos.push(new_line_index);
+                        new_line_index = CharIndex(0);
+                    }
+
+                    if ch != '\n' {
+                        new_line_index = new_line_index + CharIndex(1);
+                    }
+                    out_str.push_char(ch);
+                }
+            }
+            text.len() > 0 && is_in_whitespace(text.char_at_reverse(0), mode)
+        },
+
+        CompressWhitespace | CompressWhitespaceNewline => {
+            let mut in_whitespace: bool = incoming_whitespace;
+            for ch in text.chars() {
+                // TODO: discard newlines between CJK chars
+                let mut next_in_whitespace: bool = is_in_whitespace(ch, mode);
+
+                if !next_in_whitespace {
+                    if is_always_discardable_char(ch) {
+                        // revert whitespace setting, since this char was discarded
+                        next_in_whitespace = in_whitespace;
+                        // TODO: record skipped char
+                    } else {
+                        // TODO: record kept char
+                        out_str.push_char(ch);
+                    }
+                } else { /* next_in_whitespace; possibly add a space char */
+                    if in_whitespace {
+                        // TODO: record skipped char
+                    } else {
+                        // TODO: record kept char
+                        out_str.push_char(' ');
+                    }
+                }
+                // save whitespace context for next char
+                in_whitespace = next_in_whitespace;
+            } /* /for str::each_char */
+            in_whitespace
+        }
+    };
+
+    return (out_str.into_string(), out_whitespace);
+
+    fn is_in_whitespace(ch: char, mode: CompressionMode) -> bool {
+        match (ch, mode) {
+            (' ', _)  => true,
+            ('\t', _) => true,
+            ('\n', CompressWhitespaceNewline) => true,
+            (_, _)    => false
+        }
+    }
+
+    fn is_discardable_char(ch: char, mode: CompressionMode) -> bool {
+        if is_always_discardable_char(ch) {
+            return true;
+        }
+        match mode {
+            DiscardNewline | CompressWhitespaceNewline => ch == '\n',
+            _ => false
+        }
+    }
+
+    fn is_always_discardable_char(_ch: char) -> bool {
+        // TODO: check for bidi control chars, soft hyphens.
+        false
+    }
+}
+
+pub fn float_to_fixed(before: int, f: f64) -> i32 {
+    (1i32 << before as uint) * (f as i32)
+}
+
+pub fn fixed_to_float(before: int, f: i32) -> f64 {
+    f as f64 * 1.0f64 / ((1i32 << before as uint) as f64)
+}
+
+pub fn fixed_to_rounded_int(before: int, f: i32) -> int {
+    let half = 1i32 << (before-1) as uint;
+    if f > 0i32 {
+        ((half + f) >> before as uint) as int
+    } else {
+       -((half - f) >> before as uint) as int
+    }
+}
+
+/* Generate a 32-bit TrueType tag from its 4 characters */
+pub fn true_type_tag(a: char, b: char, c: char, d: char) -> u32 {
+    let a = a as u32;
+    let b = b as u32;
+    let c = c as u32;
+    let d = d as u32;
+    (a << 24 | b << 16 | c << 8 | d) as u32
+}
+
+#[test]
+fn test_true_type_tag() {
+    assert_eq!(true_type_tag('c', 'm', 'a', 'p'), 0x_63_6D_61_70_u32);
+}
+
+#[test]
+fn test_transform_compress_none() {
+    let test_strs = vec!(
+        "  foo bar",
+        "foo bar  ",
+        "foo\n bar",
+        "foo \nbar",
+        "  foo  bar  \nbaz",
+        "foo bar baz",
+        "foobarbaz\n\n"
+    );
+    let mode = CompressNone;
+
+    for test in test_strs.iter() {
+        let mut new_line_pos = vec!();
+        let (trimmed_str, _out) = transform_text(*test, mode, true, &mut new_line_pos);
+        assert_eq!(trimmed_str.as_slice(), *test)
+    }
+}
+
+#[test]
+fn test_transform_discard_newline() {
+    let test_strs = vec!(
+        "  foo bar",
+        "foo bar  ",
+        "foo\n bar",
+        "foo \nbar",
+        "  foo  bar  \nbaz",
+        "foo bar baz",
+        "foobarbaz\n\n"
+    );
+
+    let oracle_strs = vec!(
+        "  foo bar",
+        "foo bar  ",
+        "foo bar",
+        "foo bar",
+        "  foo  bar  baz",
+        "foo bar baz",
+        "foobarbaz"
+    );
+
+    assert_eq!(test_strs.len(), oracle_strs.len());
+    let mode = DiscardNewline;
+
+    for (test, oracle) in test_strs.iter().zip(oracle_strs.iter()) {
+        let mut new_line_pos = vec!();
+        let (trimmed_str, _out) = transform_text(*test, mode, true, &mut new_line_pos);
+        assert_eq!(trimmed_str.as_slice(), *oracle)
+    }
+}
+
+/* FIXME: Fix and re-enable
+#[test]
+fn test_transform_compress_whitespace() {
+    let  test_strs : ~[String] = ~["  foo bar".to_string(),
+                                 "foo bar  ".to_string(),
+                                 "foo\n bar".to_string(),
+                                 "foo \nbar".to_string(),
+                                 "  foo  bar  \nbaz".to_string(),
+                                 "foo bar baz".to_string(),
+                                 "foobarbaz\n\n".to_string()];
+
+    let oracle_strs : ~[String] = ~[" foo bar".to_string(),
+                                 "foo bar ".to_string(),
+                                 "foo\n bar".to_string(),
+                                 "foo \nbar".to_string(),
+                                 " foo bar \nbaz".to_string(),
+                                 "foo bar baz".to_string(),
+                                 "foobarbaz\n\n".to_string()];
+
+    assert_eq!(test_strs.len(), oracle_strs.len());
+    let mode = CompressWhitespace;
+
+    for i in range(0, test_strs.len()) {
+        let mut new_line_pos = ~[];
+        let (trimmed_str, _out) = transform_text(test_strs[i], mode, true, &mut new_line_pos);
+        assert_eq!(&trimmed_str, &oracle_strs[i])
+    }
+}
+
+#[test]
+fn test_transform_compress_whitespace_newline() {
+    let  test_strs : ~[String] = ~["  foo bar".to_string(),
+                                 "foo bar  ".to_string(),
+                                 "foo\n bar".to_string(),
+                                 "foo \nbar".to_string(),
+                                 "  foo  bar  \nbaz".to_string(),
+                                 "foo bar baz".to_string(),
+                                 "foobarbaz\n\n".to_string()];
+
+    let oracle_strs : ~[String] = ~["foo bar".to_string(),
+                                 "foo bar ".to_string(),
+                                 "foo bar".to_string(),
+                                 "foo bar".to_string(),
+                                 " foo bar baz".to_string(),
+                                 "foo bar baz".to_string(),
+                                 "foobarbaz ".to_string()];
+
+    assert_eq!(test_strs.len(), oracle_strs.len());
+    let mode = CompressWhitespaceNewline;
+
+    for i in range(0, test_strs.len()) {
+        let mut new_line_pos = ~[];
+        let (trimmed_str, _out) = transform_text(test_strs[i], mode, true, &mut new_line_pos);
+        assert_eq!(&trimmed_str, &oracle_strs[i])
+    }
+}
+*/
+
+#[test]
+fn test_transform_compress_whitespace_newline_no_incoming() {
+    let test_strs = vec!(
+        "  foo bar",
+        "\nfoo bar",
+        "foo bar  ",
+        "foo\n bar",
+        "foo \nbar",
+        "  foo  bar  \nbaz",
+        "foo bar baz",
+        "foobarbaz\n\n"
+    );
+
+    let oracle_strs = vec!(
+        " foo bar",
+        " foo bar",
+        "foo bar ",
+        "foo bar",
+        "foo bar",
+        " foo bar baz",
+        "foo bar baz",
+        "foobarbaz "
+    );
+
+    assert_eq!(test_strs.len(), oracle_strs.len());
+    let mode = CompressWhitespaceNewline;
+
+    for (test, oracle) in test_strs.iter().zip(oracle_strs.iter()) {
+        let mut new_line_pos = vec!();
+        let (trimmed_str, _out) = transform_text(*test, mode, false, &mut new_line_pos);
+        assert_eq!(trimmed_str.as_slice(), *oracle)
+    }
+}