1 files changed, 159 insertions, 23 deletions

diff --git a/codegen/canon.zig b/codegen/canon.zig
index d95a905..e92be5d 100644
--- a/codegen/canon.zig
+++ b/codegen/canon.zig
@@ -1,12 +1,38 @@
 const std = @import("std");
 const builtin = @import("builtin");
 
+const block_size = 256;
+const Block = [block_size]Canonicalization;
+
+const Canonicalization = struct {
+    len: u3 = 0,
+    cps: [2]u21 = [_]u21{0} ** 2,
+};
+
+const BlockMap = std.HashMap(
+    Block,
+    u16,
+    struct {
+        pub fn hash(_: @This(), k: Block) u64 {
+            var hasher = std.hash.Wyhash.init(0);
+            std.hash.autoHashStrat(&hasher, k, .DeepRecursive);
+            return hasher.final();
+        }
+
+        pub fn eql(_: @This(), aBlock: Block, bBlock: Block) bool {
+            return for (aBlock, bBlock) |a, b| {
+                if (a.len != b.len or a.cps[0] != b.cps[0] or a.cps[1] != b.cps[1]) return false;
+            } else true;
+        }
+    },
+    std.hash_map.default_max_load_percentage,
+);
+
 pub fn main() anyerror!void {
     var arena = std.heap.ArenaAllocator.init(std.heap.page_allocator);
     defer arena.deinit();
     const allocator = arena.allocator();
 
-    var write_buf: [4096]u8 = undefined;
     // Process UnicodeData.txt
     var in_reader = std.io.Reader.fixed(@embedFile("UnicodeData.txt"));
     var args_iter = try std.process.argsWithAllocator(allocator);
@@ -14,53 +40,163 @@ pub fn main() anyerror!void {
     _ = args_iter.skip();
     const output_path = args_iter.next() orelse @panic("No output file arg!");
 
-    var out_file = try std.fs.cwd().createFile(output_path, .{});
-    defer out_file.close();
-    var file_writer = out_file.writer(&write_buf);
-    var writer = &file_writer.interface;
-    const endian = builtin.cpu.arch.endian();
+    var canon_map = std.AutoHashMap(u21, Canonicalization).init(allocator);
+    defer canon_map.deinit();
+
+    var composite_set = std.AutoArrayHashMap(u21, [2]u21).init(allocator);
 
-    lines: while (in_reader.takeDelimiterInclusive('\n')) |took| {
-        const line = std.mem.trimRight(u8, took, "\n");
+    while (in_reader.takeDelimiterInclusive('\n')) |line| {
         if (line.len == 0) continue;
 
         var field_iter = std.mem.splitScalar(u8, line, ';');
-        var cps: [3]u24 = undefined;
-        var len: u8 = 2;
+        var cp: u21 = undefined;
 
         var i: usize = 0;
         while (field_iter.next()) |field| : (i += 1) {
+            if (field.len == 0) continue;
+
             switch (i) {
-                0 => cps[0] = try std.fmt.parseInt(u24, field, 16),
+                0 => cp = try std.fmt.parseInt(u21, field, 16),
 
                 5 => {
                     // Not canonical.
-                    if (field.len == 0 or field[0] == '<') continue :lines;
+                    if (field[0] == '<') continue;
+
                     if (std.mem.indexOfScalar(u8, field, ' ')) |space| {
                         // Canonical
-                        len = 3;
-                        cps[1] = try std.fmt.parseInt(u24, field[0..space], 16);
-                        cps[2] = try std.fmt.parseInt(u24, field[space + 1 ..], 16);
+                        const c0, const c1 = .{
+                            try std.fmt.parseInt(u21, field[0..space], 16),
+                            try std.fmt.parseInt(u21, field[space + 1 ..], 16),
+                        };
+                        try canon_map.put(cp, Canonicalization{
+                            .len = 2,
+                            .cps = [_]u21{ c0, c1 },
+                        });
+                        try composite_set.put(cp, [_]u21{ c0, c1 });
                     } else {
                         // Singleton
-                        cps[1] = try std.fmt.parseInt(u24, field, 16);
+                        try canon_map.put(cp, Canonicalization{
+                            .len = 1,
+                            .cps = [_]u21{
+                                try std.fmt.parseInt(u21, field, 16),
+                                0,
+                            },
+                        });
                     }
                 },
 
-                2 => if (line[0] == '<') continue :lines,
-
                 else => {},
             }
         }
-
-        try writer.writeInt(u8, @intCast(len), endian);
-        for (cps[0..len]) |cp| try writer.writeInt(u24, cp, endian);
     } else |err| switch (err) {
         error.EndOfStream => {},
         else => {
             return err;
         },
     }
-    try writer.writeInt(u16, 0, endian);
-    try writer.flush();
+
+    // Build multi-tiered lookup tables for decompositions
+    var blocks_map = BlockMap.init(allocator);
+    defer blocks_map.deinit();
+
+    var stage1 = std.array_list.Managed(u16).init(allocator);
+    defer stage1.deinit();
+
+    var stage2 = std.array_list.Managed(Canonicalization).init(allocator);
+    defer stage2.deinit();
+
+    var block: Block = [_]Canonicalization{.{}} ** block_size;
+    var block_len: u16 = 0;
+
+    for (0..0x110000) |i| {
+        const cp: u21 = @intCast(i);
+
+        const canon: Canonicalization = canon_map.get(cp) orelse .{};
+
+        block[block_len] = canon;
+        block_len += 1;
+
+        if (block_len < block_size and cp != 0x10ffff) continue;
+
+        const gop = try blocks_map.getOrPut(block);
+        if (!gop.found_existing) {
+            gop.value_ptr.* = @intCast(stage2.items.len);
+            try stage2.appendSlice(&block);
+        }
+
+        try stage1.append(gop.value_ptr.*);
+        block_len = 0;
+    }
+
+    var write_buf: [4096]u8 = undefined;
+    var out_file = try std.fs.cwd().createFile(output_path, .{});
+    defer out_file.close();
+    var writer = out_file.writer(&write_buf);
+
+    try writer.interface.print(
+        \\//! This file is auto-generated. Do not edit.
+        \\
+        \\pub const Canonicalization = struct {{
+        \\    len: u3,
+        \\    cps: [2]u21,
+        \\}};
+        \\
+        \\pub const s1: [{}]u16 = .{{
+    , .{stage1.items.len});
+    for (stage1.items) |entry| try writer.interface.print("{}, ", .{entry});
+
+    try writer.interface.print(
+        \\
+        \\}};
+        \\
+        \\pub const s2: [{}]Canonicalization = .{{
+    , .{stage2.items.len});
+    for (stage2.items) |entry| {
+        try writer.interface.print(".{{ .len = {}, .cps = .{{ {}, {} }} }}, ", .{
+            entry.len,
+            entry.cps[0],
+            entry.cps[1],
+        });
+    }
+
+    const composite = composite_set.keys();
+    // TODO: cut
+    try writer.interface.print(
+        \\
+        \\}};
+        \\
+        \\pub const composite: [{}]u21 = .{{
+    , .{composite.len});
+    for (composite) |entry| try writer.interface.print("{}, ", .{entry});
+
+    try writer.interface.writeAll(
+        \\};
+    );
+
+    try writer.interface.print(
+        \\
+        \\ pub const c_map: [{}]struct {{ [2]u21, u21 }} = .{{
+    , .{composite.len});
+    for (composite) |comp| {
+        const canon = canon_map.get(comp).?;
+        std.debug.assert(canon.len == 2);
+        try writer.interface.print(
+            \\ .{{ .{{{}, {}}}, {}}},
+        ,
+            .{ canon.cps[0], canon.cps[1], comp },
+        );
+    }
+    // var c_entries = composite_set.iterator();
+    // while (c_entries.next()) |entry| {
+    //     try writer.interface.print(
+    //         \\ .{{ .{{{}, {}}}, {}}},
+    //     ,
+    //         .{ entry.value_ptr[0], entry.value_ptr[1], entry.key_ptr.* },
+    //     );
+    // }
+    try writer.interface.writeAll(
+        \\};
+    );
+
+    try writer.interface.flush();
 }