Merge pull request #1158 from mitchellh/grapheme-config

config: grapheme-width-method, default to "unicode"

src/config/Config.zig

@@ -162,6 +162,32 @@ const c = @cImport({
 @"adjust-strikethrough-position": ?MetricModifier = null,
 @"adjust-strikethrough-thickness": ?MetricModifier = null,
 
+/// The method to use for calculating the cell width of a grapheme cluster.
+/// The default value is "unicode" which uses the Unicode standard to
+/// determine grapheme width. This results in correct grapheme width but
+/// may result in cursor-desync issues with some programs (such as shells)
+/// that may use a legacy method such as "wcswidth".
+///
+/// Valid values are:
+///
+///   - "wcswidth" - Use the wcswidth function to determine grapheme width.
+///     This maximizes compatibility with legacy programs but may result
+///     in incorrect grapheme width for certain graphemes such as skin-tone
+///     emoji, non-English characters, etc.
+///
+///     Note that this "wcswidth" functionality is based on the libc wcswidth,
+///     not any other libraries with that name.
+///
+///   - "unicode" - Use the Unicode standard to determine grapheme width.
+///
+/// If a running program explicitly enables terminal mode 2027, then
+/// "unicode" width will be forced regardless of this configuration. When
+/// mode 2027 is reset, this configuration will be used again.
+///
+/// This configuration can be changed at runtime but will not affect existing
+/// printed cells. Only new cells will use the new configuration.
+@"grapheme-width-method": GraphemeWidthMethod = .unicode,
+
 /// A named theme to use. The available themes are currently hardcoded to
 /// the themes that ship with Ghostty. On macOS, this list is in the
 /// `Ghostty.app/Contents/Resources/themes` directory. On Linux, this
@@ -2781,3 +2807,9 @@ pub const WindowSaveState = enum {
     never,
     always,
 };
+
+/// See grapheme-width-method
+pub const GraphemeWidthMethod = enum {
+    wcswidth,
+    unicode,
+};

src/terminal/Terminal.zig

@@ -120,6 +120,9 @@ flags: packed struct {
     /// then we want to capture the shift key for the mouse protocol
     /// if the configuration allows it.
     mouse_shift_capture: enum { null, false, true } = .null,
+
+    /// If true, we perform grapheme clustering even if mode 2027 is disabled.
+    default_grapheme_cluster: bool = false,
 } = .{},
 
 /// The event types that can be reported for mouse-related activities.
@@ -724,6 +727,8 @@ pub fn print(self: *Terminal, c: u21) !void {
     const tracy = trace(@src());
     defer tracy.end();
 
+    // log.debug("print={x} y={} x={}", .{ c, self.screen.cursor.y, self.screen.cursor.x });
+
     // If we're not on the main display, do nothing for now
     if (self.status_display != .main) return;
 
@@ -738,7 +743,7 @@ pub fn print(self: *Terminal, c: u21) !void {
     // purposely ordered in least-likely to most-likely so we can drop out
     // as quickly as possible.
     if (c > 255 and
-        self.modes.get(.grapheme_cluster) and
+        (self.modes.get(.grapheme_cluster) or self.flags.default_grapheme_cluster) and
         self.screen.cursor.x > 0)
     grapheme: {
         const row = self.screen.getRow(.{ .active = self.screen.cursor.y });
@@ -775,6 +780,7 @@ pub fn print(self: *Terminal, c: u21) !void {
             if (prev.cell.attrs.grapheme) {
                 var it = row.codepointIterator(prev.x);
                 while (it.next()) |cp2| {
+                    // log.debug("cp1={x} cp2={x}", .{ cp1, cp2 });
                     assert(!ziglyph.graphemeBreak(
                         cp1,
                         cp2,
@@ -785,6 +791,7 @@ pub fn print(self: *Terminal, c: u21) !void {
                 }
             }
 
+            // log.debug("cp1={x} cp2={x} end", .{ cp1, c });
             break :brk ziglyph.graphemeBreak(cp1, c, &state);
         };
 
@@ -868,7 +875,9 @@ pub fn print(self: *Terminal, c: u21) !void {
         // If we have grapheme clustering enabled, we don't blindly attach
         // any zero width character to our cells and we instead just ignore
         // it.
-        if (self.modes.get(.grapheme_cluster)) return;
+        if (self.modes.get(.grapheme_cluster) or
+            self.flags.default_grapheme_cluster)
+            return;
 
         // If we're at cell zero, then this is malformed data and we don't
         // print anything or even store this. Zero-width characters are ALWAYS

src/termio/Exec.zig

@@ -105,6 +105,7 @@ pub const DerivedConfig = struct {
     background: configpkg.Config.Color,
     osc_color_report_format: configpkg.Config.OSCColorReportFormat,
     term: []const u8,
+    grapheme_width_method: configpkg.Config.GraphemeWidthMethod,
 
     pub fn init(
         alloc_gpa: Allocator,
@@ -122,6 +123,7 @@ pub const DerivedConfig = struct {
             .background = config.background,
             .osc_color_report_format = config.@"osc-color-report-format",
             .term = config.term,
+            .grapheme_width_method = config.@"grapheme-width-method",
         };
     }
 
@@ -146,6 +148,7 @@ pub fn init(alloc: Allocator, opts: termio.Options) !Exec {
     errdefer term.deinit(alloc);
     term.default_palette = opts.config.palette;
     term.color_palette.colors = opts.config.palette;
+    term.flags.default_grapheme_cluster = opts.config.grapheme_width_method == .unicode;
 
     // Set the image size limits
     try term.screen.kitty_images.setLimit(alloc, opts.config.image_storage_limit);
@@ -350,6 +353,8 @@ pub fn changeConfig(self: *Exec, config: *DerivedConfig) !void {
     //   - command, working-directory: we never restart the underlying
     //   process so we don't care or need to know about these.
 
+    self.terminal.flags.default_grapheme_cluster = config.grapheme_width_method == .unicode;
+
     // Update the default palette. Note this will only apply to new colors drawn
     // since we decode all palette colors to RGB on usage.
     self.terminal.default_palette = config.palette;