Changes when case folded check; 20ms faster

3 files changed, 89 insertions, 9 deletions

diff --git a/codegen/fold.zig b/codegen/fold.zig
index 7977e61..b3192e7 100644
--- a/codegen/fold.zig
+++ b/codegen/fold.zig
@@ -8,7 +8,51 @@ pub fn main() !void {
     defer arena.deinit();
     const allocator = arena.allocator();
 
-    // Process DerivedEastAsianWidth.txt
+    // Process DerivedCoreProperties.txt
+    var cp_file = try std.fs.cwd().openFile("data/unicode/DerivedCoreProperties.txt", .{});
+    defer cp_file.close();
+    var cp_buf = std.io.bufferedReader(cp_file.reader());
+    const cp_reader = cp_buf.reader();
+
+    var cp_map = std.AutoHashMap(u21, void).init(allocator);
+    defer cp_map.deinit();
+
+    var line_buf: [4096]u8 = undefined;
+
+    cp_lines: while (try cp_reader.readUntilDelimiterOrEof(&line_buf, '\n')) |line| {
+        if (line.len == 0 or line[0] == '#') continue;
+
+        const no_comment = if (std.mem.indexOfScalar(u8, line, '#')) |octo| line[0..octo] else line;
+
+        var field_iter = std.mem.tokenizeAny(u8, no_comment, "; ");
+        var current_code: [2]u21 = undefined;
+
+        var i: usize = 0;
+        while (field_iter.next()) |field| : (i += 1) {
+            switch (i) {
+                0 => {
+                    // Code point(s)
+                    if (std.mem.indexOf(u8, field, "..")) |dots| {
+                        current_code = .{
+                            try std.fmt.parseInt(u21, field[0..dots], 16),
+                            try std.fmt.parseInt(u21, field[dots + 2 ..], 16),
+                        };
+                    } else {
+                        const code = try std.fmt.parseInt(u21, field, 16);
+                        current_code = .{ code, code };
+                    }
+                },
+                1 => {
+                    // Core property
+                    if (!mem.eql(u8, field, "Changes_When_Casefolded")) continue :cp_lines;
+                    for (current_code[0]..current_code[1] + 1) |cp| try cp_map.put(@intCast(cp), {});
+                },
+                else => {},
+            }
+        }
+    }
+
+    // Process CaseFolding.txt
     var in_file = try std.fs.cwd().openFile("data/unicode/CaseFolding.txt", .{});
     defer in_file.close();
     var in_buf = std.io.bufferedReader(in_file.reader());
@@ -27,7 +71,6 @@ pub fn main() !void {
     const writer = out_comp.writer();
 
     const endian = builtin.cpu.arch.endian();
-    var line_buf: [4096]u8 = undefined;
 
     lines: while (try in_reader.readUntilDelimiterOrEof(&line_buf, '\n')) |line| {
         if (line.len == 0 or line[0] == '#') continue;
@@ -41,7 +84,12 @@ pub fn main() !void {
         var i: usize = 0;
         while (field_iter.next()) |field| : (i += 1) {
             switch (i) {
-                0 => cps[0] = try fmt.parseInt(u24, field, 16),
+                0 => {
+                    var cp = try fmt.parseInt(u21, field, 16);
+                    cp <<= 1;
+                    if (cp_map.contains(cp)) cp |= 1;
+                    cps[0] = cp;
+                },
 
                 1 => {
                     if (!mem.eql(u8, field, "C") and !mem.eql(u8, field, "F")) continue :lines;
diff --git a/src/FoldData.zig b/src/FoldData.zig
index 139c677..2a9a1f5 100644
--- a/src/FoldData.zig
+++ b/src/FoldData.zig
@@ -5,6 +5,7 @@ const mem = std.mem;
 
 allocator: mem.Allocator,
 fold: [][]u21 = undefined,
+cwcf: []bool = undefined,
 
 const Self = @This();
 
@@ -20,18 +21,21 @@ pub fn init(allocator: mem.Allocator) !Self {
     var self = Self{
         .allocator = allocator,
         .fold = try allocator.alloc([]u21, 0x110000),
+        .cwcf = try allocator.alloc(bool, 0x110000),
     };
 
     @memset(self.fold, &.{});
+    @memset(self.cwcf, false);
 
     while (true) {
         const len: u8 = try reader.readInt(u8, endian);
         if (len == 0) break;
         const cp = try reader.readInt(u24, endian);
-        self.fold[cp] = try allocator.alloc(u21, len - 1);
+        self.fold[cp >> 1] = try allocator.alloc(u21, len - 1);
         for (0..len - 1) |i| {
-            self.fold[cp][i] = @intCast(try reader.readInt(u24, endian));
+            self.fold[cp >> 1][i] = @intCast(try reader.readInt(u24, endian));
         }
+        self.cwcf[cp >> 1] = cp & 1 == 1;
     }
 
     return self;
@@ -40,9 +44,15 @@ pub fn init(allocator: mem.Allocator) !Self {
 pub fn deinit(self: *Self) void {
     for (self.fold) |slice| self.allocator.free(slice);
     self.allocator.free(self.fold);
+    self.allocator.free(self.cwcf);
 }
 
 /// Returns the case fold for `cp`.
 pub inline fn caseFold(self: Self, cp: u21) []const u21 {
     return self.fold[cp];
 }
+
+/// Returns true when caseFold(NFD(`cp`)) != NFD(`cp`).
+pub inline fn changesWhenCaseFolded(self: Self, cp: u21) bool {
+    return self.cwcf[cp];
+}
diff --git a/src/Normalizer.zig b/src/Normalizer.zig
index c68b2ec..5a26dfa 100644
--- a/src/Normalizer.zig
+++ b/src/Normalizer.zig
@@ -389,6 +389,12 @@ fn nfkdCodePoints(
     return try dcp_list.toOwnedSlice();
 }
 
+fn changesWhenCaseFolded(self: Self, cps: []const u21) bool {
+    return for (cps) |cp| {
+        if (self.norm_data.fold_data.changesWhenCaseFolded(cp)) break true;
+    } else false;
+}
+
 pub fn eqlIgnoreCase(
     self: Self,
     allocator: mem.Allocator,
@@ -397,10 +403,18 @@ pub fn eqlIgnoreCase(
 ) !bool {
     if (ascii.isAsciiOnly(a) and ascii.isAsciiOnly(b)) return std.ascii.eqlIgnoreCase(a, b);
 
+    // Process a
     const nfd_a = try self.nfxdCodePoints(allocator, a, .nfd);
     defer allocator.free(nfd_a);
-    const cf_nfd_a = try self.caseFold(allocator, nfd_a);
-    defer allocator.free(cf_nfd_a);
+
+    var need_frr_cf_nfd_a = false;
+    var cf_nfd_a: []const u21 = nfd_a;
+    if (self.changesWhenCaseFolded(nfd_a)) {
+        cf_nfd_a = try self.caseFold(allocator, nfd_a);
+        need_frr_cf_nfd_a = true;
+    }
+    defer if (need_frr_cf_nfd_a) allocator.free(cf_nfd_a);
+
     const nfkd_cf_nfd_a = try self.nfkdCodePoints(allocator, cf_nfd_a);
     defer allocator.free(nfkd_cf_nfd_a);
     const cf_nfkd_cf_nfd_a = try self.caseFold(allocator, nfkd_cf_nfd_a);
@@ -408,10 +422,18 @@ pub fn eqlIgnoreCase(
     const nfkd_cf_nfkd_cf_nfd_a = try self.nfkdCodePoints(allocator, cf_nfkd_cf_nfd_a);
     defer allocator.free(nfkd_cf_nfkd_cf_nfd_a);
 
+    // Process b
     const nfd_b = try self.nfxdCodePoints(allocator, b, .nfd);
     defer allocator.free(nfd_b);
-    const cf_nfd_b = try self.caseFold(allocator, nfd_b);
-    defer allocator.free(cf_nfd_b);
+
+    var need_frr_cf_nfd_b = false;
+    var cf_nfd_b: []const u21 = nfd_b;
+    if (self.changesWhenCaseFolded(nfd_b)) {
+        cf_nfd_b = try self.caseFold(allocator, nfd_b);
+        need_frr_cf_nfd_b = true;
+    }
+    defer if (need_frr_cf_nfd_b) allocator.free(cf_nfd_b);
+
     const nfkd_cf_nfd_b = try self.nfkdCodePoints(allocator, cf_nfd_b);
     defer allocator.free(nfkd_cf_nfd_b);
     const cf_nfkd_cf_nfd_b = try self.caseFold(allocator, nfkd_cf_nfd_b);