font: move shaper into comptime interface

This commit is contained in:
Mitchell Hashimoto
2022-12-06 11:20:37 -08:00
parent b043748794
commit e3c18f3f51
3 changed files with 224 additions and 209 deletions

View File

@ -10,7 +10,8 @@ pub const Face = face.Face;
pub const Group = @import("Group.zig"); pub const Group = @import("Group.zig");
pub const GroupCache = @import("GroupCache.zig"); pub const GroupCache = @import("GroupCache.zig");
pub const Glyph = @import("Glyph.zig"); pub const Glyph = @import("Glyph.zig");
pub const Shaper = @import("Shaper.zig"); pub const shape = @import("shape.zig");
pub const Shaper = shape.Shaper;
pub const sprite = @import("sprite.zig"); pub const sprite = @import("sprite.zig");
pub const Sprite = sprite.Sprite; pub const Sprite = sprite.Sprite;
pub const Descriptor = discovery.Descriptor; pub const Descriptor = discovery.Descriptor;

14
src/font/shape.zig Normal file
View File

@ -0,0 +1,14 @@
const builtin = @import("builtin");
const options = @import("main.zig").options;
const harfbuzz = @import("shaper/harfbuzz.zig");
/// Shaper implementation for our compile options.
pub const Shaper = switch (options.backend) {
.freetype,
.fontconfig_freetype,
.coretext_freetype,
.coretext,
=> harfbuzz.Shaper,
.web_canvas => harfbuzz.Shaper,
};

View File

@ -1,230 +1,230 @@
//! This struct handles text shaping.
const Shaper = @This();
const std = @import("std"); const std = @import("std");
const assert = std.debug.assert; const assert = std.debug.assert;
const Allocator = std.mem.Allocator; const Allocator = std.mem.Allocator;
const harfbuzz = @import("harfbuzz"); const harfbuzz = @import("harfbuzz");
const trace = @import("tracy").trace; const trace = @import("tracy").trace;
const font = @import("main.zig"); const font = @import("../main.zig");
const Face = @import("main.zig").Face; const Face = font.Face;
const DeferredFace = @import("main.zig").DeferredFace; const DeferredFace = font.DeferredFace;
const Group = @import("main.zig").Group; const Group = font.Group;
const GroupCache = @import("main.zig").GroupCache; const GroupCache = font.GroupCache;
const Library = @import("main.zig").Library; const Library = font.Library;
const Style = @import("main.zig").Style; const Style = font.Style;
const Presentation = @import("main.zig").Presentation; const Presentation = font.Presentation;
const terminal = @import("../terminal/main.zig"); const terminal = @import("../../terminal/main.zig");
const log = std.log.scoped(.font_shaper); const log = std.log.scoped(.font_shaper);
/// The buffer used for text shaping. We reuse it across multiple shaping /// Shaper that uses Harfbuzz.
/// calls to prevent allocations. pub const Shaper = struct {
hb_buf: harfbuzz.Buffer, /// The buffer used for text shaping. We reuse it across multiple shaping
/// calls to prevent allocations.
hb_buf: harfbuzz.Buffer,
/// The shared memory used for shaping results. /// The shared memory used for shaping results.
cell_buf: []Cell, cell_buf: []Cell,
/// The cell_buf argument is the buffer to use for storing shaped results. /// The cell_buf argument is the buffer to use for storing shaped results.
/// This should be at least the number of columns in the terminal. /// This should be at least the number of columns in the terminal.
pub fn init(cell_buf: []Cell) !Shaper { pub fn init(cell_buf: []Cell) !Shaper {
return Shaper{ return Shaper{
.hb_buf = try harfbuzz.Buffer.create(), .hb_buf = try harfbuzz.Buffer.create(),
.cell_buf = cell_buf, .cell_buf = cell_buf,
};
}
pub fn deinit(self: *Shaper) void {
self.hb_buf.destroy();
}
/// Returns an iterator that returns one text run at a time for the
/// given terminal row. Note that text runs are are only valid one at a time
/// for a Shaper struct since they share state.
pub fn runIterator(self: *Shaper, group: *GroupCache, row: terminal.Screen.Row) RunIterator {
return .{ .shaper = self, .group = group, .row = row };
}
/// Shape the given text run. The text run must be the immediately previous
/// text run that was iterated since the text run does share state with the
/// Shaper struct.
///
/// The return value is only valid until the next shape call is called.
///
/// If there is not enough space in the cell buffer, an error is returned.
pub fn shape(self: *Shaper, run: TextRun) ![]Cell {
const tracy = trace(@src());
defer tracy.end();
// We only do shaping if the font is not a special-case. For special-case
// fonts, the codepoint == glyph_index so we don't need to run any shaping.
if (run.font_index.special() == null) {
// TODO: we do not want to hardcode these
const hb_feats = &[_]harfbuzz.Feature{
harfbuzz.Feature.fromString("dlig").?,
harfbuzz.Feature.fromString("liga").?,
}; };
const face = try run.group.group.faceFromIndex(run.font_index);
harfbuzz.shape(face.hb_font, self.hb_buf, hb_feats);
} }
// If our buffer is empty, we short-circuit the rest of the work pub fn deinit(self: *Shaper) void {
// return nothing. self.hb_buf.destroy();
if (self.hb_buf.getLength() == 0) return self.cell_buf[0..0];
const info = self.hb_buf.getGlyphInfos();
const pos = self.hb_buf.getGlyphPositions() orelse return error.HarfbuzzFailed;
// This is perhaps not true somewhere, but we currently assume it is true.
// If it isn't true, I'd like to catch it and learn more.
assert(info.len == pos.len);
// Convert all our info/pos to cells and set it.
if (info.len > self.cell_buf.len) return error.OutOfMemory;
//log.warn("info={} pos={} run={}", .{ info.len, pos.len, run });
for (info) |v, i| {
self.cell_buf[i] = .{
.x = @intCast(u16, v.cluster),
.glyph_index = v.codepoint,
};
//log.warn("i={} info={} pos={} cell={}", .{ i, v, pos[i], self.cell_buf[i] });
} }
return self.cell_buf[0..info.len]; /// Returns an iterator that returns one text run at a time for the
} /// given terminal row. Note that text runs are are only valid one at a time
/// for a Shaper struct since they share state.
pub fn runIterator(self: *Shaper, group: *GroupCache, row: terminal.Screen.Row) RunIterator {
return .{ .shaper = self, .group = group, .row = row };
}
pub const Cell = struct { /// Shape the given text run. The text run must be the immediately previous
/// The column that this cell occupies. Since a set of shaper cells is /// text run that was iterated since the text run does share state with the
/// always on the same line, only the X is stored. It is expected the /// Shaper struct.
/// caller has access to the original screen cell. ///
x: u16, /// The return value is only valid until the next shape call is called.
///
/// The glyph index for this cell. The font index to use alongside /// If there is not enough space in the cell buffer, an error is returned.
/// this cell is available in the text run. pub fn shape(self: *Shaper, run: TextRun) ![]Cell {
glyph_index: u32,
};
/// A single text run. A text run is only valid for one Shaper and
/// until the next run is created.
pub const TextRun = struct {
/// The offset in the row where this run started
offset: u16,
/// The total number of cells produced by this run.
cells: u16,
/// The font group that built this run.
group: *GroupCache,
/// The font index to use for the glyphs of this run.
font_index: Group.FontIndex,
};
pub const RunIterator = struct {
shaper: *Shaper,
group: *GroupCache,
row: terminal.Screen.Row,
i: usize = 0,
pub fn next(self: *RunIterator, alloc: Allocator) !?TextRun {
const tracy = trace(@src()); const tracy = trace(@src());
defer tracy.end(); defer tracy.end();
// Trim the right side of a row that might be empty // We only do shaping if the font is not a special-case. For special-case
const max: usize = max: { // fonts, the codepoint == glyph_index so we don't need to run any shaping.
var j: usize = self.row.lenCells(); if (run.font_index.special() == null) {
while (j > 0) : (j -= 1) if (!self.row.getCell(j - 1).empty()) break; // TODO: we do not want to hardcode these
break :max j; const hb_feats = &[_]harfbuzz.Feature{
}; harfbuzz.Feature.fromString("dlig").?,
harfbuzz.Feature.fromString("liga").?,
};
// We're over at the max const face = try run.group.group.faceFromIndex(run.font_index);
if (self.i >= max) return null; harfbuzz.shape(face.hb_font, self.hb_buf, hb_feats);
// Track the font for our curent run
var current_font: Group.FontIndex = .{};
// Reset the buffer for our current run
self.shaper.hb_buf.reset();
self.shaper.hb_buf.setContentType(.unicode);
// Go through cell by cell and accumulate while we build our run.
var j: usize = self.i;
while (j < max) : (j += 1) {
const cluster = j;
const cell = self.row.getCell(j);
// If we're a spacer, then we ignore it
if (cell.attrs.wide_spacer_tail) continue;
const style: Style = if (cell.attrs.bold)
.bold
else
.regular;
// Determine the presentation format for this glyph.
const presentation: ?Presentation = if (cell.attrs.grapheme) p: {
// We only check the FIRST codepoint because I believe the
// presentation format must be directly adjacent to the codepoint.
var it = self.row.codepointIterator(j);
if (it.next()) |cp| {
if (cp == 0xFE0E) break :p Presentation.text;
if (cp == 0xFE0F) break :p Presentation.emoji;
}
break :p null;
} else null;
// Determine the font for this cell. We'll use fallbacks
// manually here to try replacement chars and then a space
// for unknown glyphs.
const font_idx_opt = (try self.group.indexForCodepoint(
alloc,
if (cell.empty() or cell.char == 0) ' ' else cell.char,
style,
presentation,
)) orelse (try self.group.indexForCodepoint(
alloc,
0xFFFD,
style,
.text,
)) orelse
try self.group.indexForCodepoint(alloc, ' ', style, .text);
const font_idx = font_idx_opt.?;
//log.warn("char={x} idx={}", .{ cell.char, font_idx });
if (j == self.i) current_font = font_idx;
// If our fonts are not equal, then we're done with our run.
if (font_idx.int() != current_font.int()) break;
// Continue with our run
self.shaper.hb_buf.add(cell.char, @intCast(u32, cluster));
// If this cell is part of a grapheme cluster, add all the grapheme
// data points.
if (cell.attrs.grapheme) {
var it = self.row.codepointIterator(j);
while (it.next()) |cp| {
if (cp == 0xFE0E or cp == 0xFE0F) continue;
self.shaper.hb_buf.add(cp, @intCast(u32, cluster));
}
}
} }
// Finalize our buffer // If our buffer is empty, we short-circuit the rest of the work
self.shaper.hb_buf.guessSegmentProperties(); // return nothing.
if (self.hb_buf.getLength() == 0) return self.cell_buf[0..0];
const info = self.hb_buf.getGlyphInfos();
const pos = self.hb_buf.getGlyphPositions() orelse return error.HarfbuzzFailed;
// Move our cursor. Must defer since we use self.i below. // This is perhaps not true somewhere, but we currently assume it is true.
defer self.i = j; // If it isn't true, I'd like to catch it and learn more.
assert(info.len == pos.len);
return TextRun{ // Convert all our info/pos to cells and set it.
.offset = @intCast(u16, self.i), if (info.len > self.cell_buf.len) return error.OutOfMemory;
.cells = @intCast(u16, j - self.i), //log.warn("info={} pos={} run={}", .{ info.len, pos.len, run });
.group = self.group,
.font_index = current_font, for (info) |v, i| {
}; self.cell_buf[i] = .{
.x = @intCast(u16, v.cluster),
.glyph_index = v.codepoint,
};
//log.warn("i={} info={} pos={} cell={}", .{ i, v, pos[i], self.cell_buf[i] });
}
return self.cell_buf[0..info.len];
} }
pub const Cell = struct {
/// The column that this cell occupies. Since a set of shaper cells is
/// always on the same line, only the X is stored. It is expected the
/// caller has access to the original screen cell.
x: u16,
/// The glyph index for this cell. The font index to use alongside
/// this cell is available in the text run.
glyph_index: u32,
};
/// A single text run. A text run is only valid for one Shaper and
/// until the next run is created.
pub const TextRun = struct {
/// The offset in the row where this run started
offset: u16,
/// The total number of cells produced by this run.
cells: u16,
/// The font group that built this run.
group: *GroupCache,
/// The font index to use for the glyphs of this run.
font_index: Group.FontIndex,
};
pub const RunIterator = struct {
shaper: *Shaper,
group: *GroupCache,
row: terminal.Screen.Row,
i: usize = 0,
pub fn next(self: *RunIterator, alloc: Allocator) !?TextRun {
const tracy = trace(@src());
defer tracy.end();
// Trim the right side of a row that might be empty
const max: usize = max: {
var j: usize = self.row.lenCells();
while (j > 0) : (j -= 1) if (!self.row.getCell(j - 1).empty()) break;
break :max j;
};
// We're over at the max
if (self.i >= max) return null;
// Track the font for our curent run
var current_font: Group.FontIndex = .{};
// Reset the buffer for our current run
self.shaper.hb_buf.reset();
self.shaper.hb_buf.setContentType(.unicode);
// Go through cell by cell and accumulate while we build our run.
var j: usize = self.i;
while (j < max) : (j += 1) {
const cluster = j;
const cell = self.row.getCell(j);
// If we're a spacer, then we ignore it
if (cell.attrs.wide_spacer_tail) continue;
const style: Style = if (cell.attrs.bold)
.bold
else
.regular;
// Determine the presentation format for this glyph.
const presentation: ?Presentation = if (cell.attrs.grapheme) p: {
// We only check the FIRST codepoint because I believe the
// presentation format must be directly adjacent to the codepoint.
var it = self.row.codepointIterator(j);
if (it.next()) |cp| {
if (cp == 0xFE0E) break :p Presentation.text;
if (cp == 0xFE0F) break :p Presentation.emoji;
}
break :p null;
} else null;
// Determine the font for this cell. We'll use fallbacks
// manually here to try replacement chars and then a space
// for unknown glyphs.
const font_idx_opt = (try self.group.indexForCodepoint(
alloc,
if (cell.empty() or cell.char == 0) ' ' else cell.char,
style,
presentation,
)) orelse (try self.group.indexForCodepoint(
alloc,
0xFFFD,
style,
.text,
)) orelse
try self.group.indexForCodepoint(alloc, ' ', style, .text);
const font_idx = font_idx_opt.?;
//log.warn("char={x} idx={}", .{ cell.char, font_idx });
if (j == self.i) current_font = font_idx;
// If our fonts are not equal, then we're done with our run.
if (font_idx.int() != current_font.int()) break;
// Continue with our run
self.shaper.hb_buf.add(cell.char, @intCast(u32, cluster));
// If this cell is part of a grapheme cluster, add all the grapheme
// data points.
if (cell.attrs.grapheme) {
var it = self.row.codepointIterator(j);
while (it.next()) |cp| {
if (cp == 0xFE0E or cp == 0xFE0F) continue;
self.shaper.hb_buf.add(cp, @intCast(u32, cluster));
}
}
}
// Finalize our buffer
self.shaper.hb_buf.guessSegmentProperties();
// Move our cursor. Must defer since we use self.i below.
defer self.i = j;
return TextRun{
.offset = @intCast(u16, self.i),
.cells = @intCast(u16, j - self.i),
.group = self.group,
.font_index = current_font,
};
}
};
}; };
test "run iterator" { test "run iterator" {
@ -619,7 +619,7 @@ const TestShaper = struct {
shaper: Shaper, shaper: Shaper,
cache: *GroupCache, cache: *GroupCache,
lib: Library, lib: Library,
cell_buf: []Cell, cell_buf: []Shaper.Cell,
pub fn deinit(self: *TestShaper) void { pub fn deinit(self: *TestShaper) void {
self.shaper.deinit(); self.shaper.deinit();
@ -632,9 +632,9 @@ const TestShaper = struct {
/// Helper to return a fully initialized shaper. /// Helper to return a fully initialized shaper.
fn testShaper(alloc: Allocator) !TestShaper { fn testShaper(alloc: Allocator) !TestShaper {
const testFont = @import("test.zig").fontRegular; const testFont = @import("../test.zig").fontRegular;
const testEmoji = @import("test.zig").fontEmoji; const testEmoji = @import("../test.zig").fontEmoji;
const testEmojiText = @import("test.zig").fontEmojiText; const testEmojiText = @import("../test.zig").fontEmojiText;
var lib = try Library.init(); var lib = try Library.init();
errdefer lib.deinit(); errdefer lib.deinit();
@ -653,10 +653,10 @@ fn testShaper(alloc: Allocator) !TestShaper {
try cache_ptr.group.addFace(alloc, .regular, DeferredFace.initLoaded(try Face.init(lib, testEmoji, .{ .points = 12 }))); try cache_ptr.group.addFace(alloc, .regular, DeferredFace.initLoaded(try Face.init(lib, testEmoji, .{ .points = 12 })));
try cache_ptr.group.addFace(alloc, .regular, DeferredFace.initLoaded(try Face.init(lib, testEmojiText, .{ .points = 12 }))); try cache_ptr.group.addFace(alloc, .regular, DeferredFace.initLoaded(try Face.init(lib, testEmojiText, .{ .points = 12 })));
var cell_buf = try alloc.alloc(Cell, 80); var cell_buf = try alloc.alloc(Shaper.Cell, 80);
errdefer alloc.free(cell_buf); errdefer alloc.free(cell_buf);
var shaper = try init(cell_buf); var shaper = try Shaper.init(cell_buf);
errdefer shaper.deinit(); errdefer shaper.deinit();
return TestShaper{ return TestShaper{