Merge pull request #2086 from ghostty-org/split

font/shaper: split text runs on common bad ligature pairs

src/font/shaper/coretext.zig

@@ -698,6 +698,27 @@ test "run iterator" {
         while (try it.next(alloc)) |_| count += 1;
         try testing.expectEqual(@as(usize, 3), count);
     }
+
+    // Bad ligatures
+    for (&[_][]const u8{ "fl", "fi", "st" }) |bad| {
+        // Make a screen with some data
+        var screen = try terminal.Screen.init(alloc, 5, 3, 0);
+        defer screen.deinit();
+        try screen.testWriteString(bad);
+
+        // Get our run iterator
+        var shaper = &testdata.shaper;
+        var it = shaper.runIterator(
+            testdata.grid,
+            &screen,
+            screen.pages.pin(.{ .screen = .{ .y = 0 } }).?,
+            null,
+            null,
+        );
+        var count: usize = 0;
+        while (try it.next(alloc)) |_| count += 1;
+        try testing.expectEqual(@as(usize, 2), count);
+    }
 }
 
 test "run iterator: empty cells with background set" {

src/font/shaper/run.zig

@@ -104,6 +104,30 @@ pub const RunIterator = struct {
             if (j > self.i) style: {
                 const prev_cell = cells[j - 1];
 
+                // If the prev cell and this cell are both plain
+                // codepoints then we check if they are commonly "bad"
+                // ligatures and spit the run if they are.
+                if (prev_cell.content_tag == .codepoint and
+                    cell.content_tag == .codepoint)
+                {
+                    const prev_cp = prev_cell.codepoint();
+                    switch (prev_cp) {
+                        // fl, fi
+                        'f' => {
+                            const cp = cell.codepoint();
+                            if (cp == 'l' or cp == 'i') break;
+                        },
+
+                        // st
+                        's' => {
+                            const cp = cell.codepoint();
+                            if (cp == 't') break;
+                        },
+
+                        else => {},
+                    }
+                }
+
                 // If the style is exactly the change then fast path out.
                 if (prev_cell.style_id == cell.style_id) break :style;
 

src/quirks.zig

@@ -12,21 +12,18 @@ const font = @import("font/main.zig");
 
 /// If true, the default font features should be disabled for the given face.
 pub fn disableDefaultFontFeatures(face: *const font.Face) bool {
-    var buf: [64]u8 = undefined;
+    _ = face;
-    const name = face.name(&buf) catch |err| switch (err) {
-        // If the name doesn't fit in buf we know this will be false
-        // because we have no quirks fonts that are longer than buf!
-        error.OutOfMemory => return false,
-    };
 
-    // CodeNewRoman, Menlo and Monaco both have a default ligature of "fi" that
+    // This function used to do something, but we integrated the logic
-    // looks really bad in terminal grids, so we want to disable ligatures
+    // we checked for directly into our shaping algorithm. It's likely
-    // by default for these faces.
+    // there are other broken fonts for other reasons so I'm keeping this
-    //
+    // around so its easy to add more checks in the future.
-    // JuliaMono has a default ligature of "st" that looks bad.
+    return false;
-    return std.mem.eql(u8, name, "CodeNewRoman") or
+
-        std.mem.eql(u8, name, "CodeNewRoman Nerd Font") or
+    // var buf: [64]u8 = undefined;
-        std.mem.eql(u8, name, "JuliaMono") or
+    // const name = face.name(&buf) catch |err| switch (err) {
-        std.mem.eql(u8, name, "Menlo") or
+    //     // If the name doesn't fit in buf we know this will be false
-        std.mem.eql(u8, name, "Monaco");
+    //     // because we have no quirks fonts that are longer than buf!
+    //     error.OutOfMemory => return false,
+    // };
 }