diff --git a/stage2/utils/arena.spl b/stage2/utils/arena.spl
new file mode 100644
index 0000000..816b7a7
--- /dev/null
+++ b/stage2/utils/arena.spl
@@ -0,0 +1,177 @@
+// Expect retcode = 0
+// 分块 bump 分配器：按块分配、块内顺序推进，只在 drop 时整体释放。
+// 适合 AST/IR 这类生命周期一致的批量小对象；不支持单独 free。
+// 失败（内存不足/参数非法）返回 null，不改变已有分配。
+
+type .{ .Vec } = @import("vec.spl");
+type Mem = @import("../sys/mem.spl");
+
+type Chunk = struct {
+    base: *u8,
+    size: usize,
+    used: usize,
+}
+
+type Arena = @self;
+
+chunk_size: usize,
+chunks: Vec, // 元素类型 Chunk
+current: usize,
+
+fn new(chunk_size: usize): Arena {
+    var cs = chunk_size;
+    if cs == 0 { cs = 4096; }
+    ret Arena {
+        .chunk_size = cs,
+        .chunks = Vec.blank(@sizeof(Chunk), 8),
+        .current = 0,
+    };
+}
+
+fn is_pow2(x: usize): bool {
+    ret x != 0 && (x & (x - 1)) == 0;
+}
+
+fn align_up(x: usize, align: usize): usize {
+    var mask = align - 1;
+    ret (x + mask) & ~mask;
+}
+
+fn chunk_at(self: *Arena, index: usize): *Chunk {
+    match Vec.get(&self.chunks, index) {
+        .Ok(ptr) => { ret ptr as *Chunk; },
+        .Err(error) => { ret null; },
+    }
+    ret null;
+}
+
+fn push_chunk(self: *Arena, size: usize): bool {
+    var base = Mem.alloc(size) as *u8;
+    if base == null { ret false; }
+    var c = Chunk { .base = base, .size = size, .used = 0 };
+    match Vec.push(&self.chunks, &c as *u8) {
+        .Ok => {},
+        .Err(error) => { Mem.free(base as *any); ret false; },
+    }
+    self.current = self.chunks.len - 1;
+    ret true;
+}
+
+// 分配 size 字节，地址按 align（非零二次幂）对齐。失败返回 null。
+fn alloc(self: *Arena, size: usize, align: usize): *u8 {
+    if size == 0 { ret null; }
+    if align == 0 { align = 1; }
+    if !is_pow2(align) { ret null; }
+
+    while true {
+        if self.current >= self.chunks.len {
+            var want = self.chunk_size;
+            var need = size + align;
+            if need > want { want = need; }
+            if !push_chunk(self, want) { ret null; }
+        }
+
+        var c = chunk_at(self, self.current);
+        if c == null { ret null; }
+        var addr = c.base as usize;
+        var aligned = align_up(addr + c.used, align);
+        var offset = aligned - addr;
+        if offset <= c.size && size <= c.size - offset {
+            c.used = offset + size;
+            ret aligned as *u8;
+        }
+        self.current = self.current + 1;
+    }
+    ret null;
+}
+
+fn chunk_count(self: Arena): usize {
+    ret self.chunks.len;
+}
+
+// 复用已分配的块，不释放内存。
+fn reset(self: *Arena) {
+    var i: usize = 0;
+    while i < self.chunks.len {
+        var c = chunk_at(self, i);
+        if c != null { c.used = 0; }
+        i = i + 1;
+    }
+    self.current = 0;
+}
+
+fn drop(self: *Arena) {
+    var i: usize = 0;
+    while i < self.chunks.len {
+        var c = chunk_at(self, i);
+        if c != null { Mem.free(c.base as *any); }
+        i = i + 1;
+    }
+    Vec.drop(&self.chunks);
+    self.current = 0;
+}
+
+// ---- 测试辅助 ----
+
+fn is_aligned(p: *u8, align: usize): bool {
+    ret ((p as usize) & (align - 1)) == 0;
+}
+
+fn test_alloc_and_alignment() {
+    var a = new(64);
+    @assert(chunk_count(a) == 0, "lazy");
+
+    var p1 = alloc(&a, 1, 1);
+    var p8 = alloc(&a, 16, 8);
+    var p16 = alloc(&a, 1, 16);
+    @assert(p1 != null && p8 != null && p16 != null, "alloc");
+    @assert(p1 != p8 && p8 != p16 && p1 != p16, "distinct");
+    @assert(is_aligned(p8, 8), "align 8");
+    @assert(is_aligned(p16, 16), "align 16");
+
+    p1[0] = 7;
+    @assert(p1[0] == 7, "writable");
+    @assert(p8[0] == p8[0], "no crash");
+
+    @assert(chunk_count(a) == 1, "one chunk");
+    reset(&a);
+    @assert(chunk_count(a) == 1, "reset keeps chunks");
+    var q = alloc(&a, 1, 1);
+    @assert(q == p1, "reset reuses first chunk");
+
+    drop(&a);
+    @assert(chunk_count(a) == 0, "drop empties");
+}
+
+fn test_growth() {
+    var a = new(64);
+    var i: i32 = 0;
+    while i < 100 {
+        var p = alloc(&a, 16, 8);
+        @assert(p != null, "grow alloc");
+        @assert(is_aligned(p, 8), "grow align");
+        p[0] = i as u8;
+        i = i + 1;
+    }
+    @assert(chunk_count(a) > 1, "grew multiple chunks");
+
+    // 超过块大小的单次分配也能成功（新块按需放大）。
+    var big = alloc(&a, 1024, 16);
+    @assert(big != null && is_aligned(big, 16), "oversized alloc");
+    drop(&a);
+}
+
+fn test_failures() {
+    var a = new(64);
+    @assert(alloc(&a, 0, 1) == null, "zero size");
+    var d = alloc(&a, 8, 0);
+    @assert(d != null && is_aligned(d, 1), "align 0 defaults to 1");
+    @assert(alloc(&a, 8, 3) == null, "non-pow2 align");
+    drop(&a);
+}
+
+fn main() {
+    test_alloc_and_alignment();
+    test_growth();
+    test_failures();
+}
diff --git a/stage2/utils/ascii.spl b/stage2/utils/ascii.spl
index d538465..a4cd82f 100644
--- a/stage2/utils/ascii.spl
+++ b/stage2/utils/ascii.spl
@@ -17,23 +17,23 @@ fn is_control(c: u8): bool {
 }
 
 fn is_space(c: u8): bool {
-    ret c == 32 || (c >= 9 && c <= 13);
+    ret c == ' ' || (c >= '\t' && c <= '\r');
 }
 
 fn is_blank(c: u8): bool {
-    ret c == 32 || c == 9;
+    ret c == ' ' || c == '\t';
 }
 
 fn is_digit(c: u8): bool {
-    ret c >= 48 && c <= 57;
+    ret c >= '0' && c <= '9';
 }
 
 fn is_lower(c: u8): bool {
-    ret c >= 97 && c <= 122;
+    ret c >= 'a' && c <= 'z';
 }
 
 fn is_upper(c: u8): bool {
-    ret c >= 65 && c <= 90;
+    ret c >= 'A' && c <= 'Z';
 }
 
 fn is_alpha(c: u8): bool {
@@ -45,15 +45,15 @@ fn is_alnum(c: u8): bool {
 }
 
 fn is_hex_digit(c: u8): bool {
-    ret is_digit(c) || (c >= 65 && c <= 70) || (c >= 97 && c <= 102);
+    ret is_digit(c) || (c >= 'A' && c <= 'F') || (c >= 'a' && c <= 'f');
 }
 
 fn is_print(c: u8): bool {
-    ret c >= 32 && c <= 126;
+    ret c >= ' ' && c <= '~';
 }
 
 fn is_graph(c: u8): bool {
-    ret c >= 33 && c <= 126;
+    ret c >= '!' && c <= '~';
 }
 
 fn is_punct(c: u8): bool {
@@ -61,12 +61,12 @@ fn is_punct(c: u8): bool {
 }
 
 fn to_lower(c: u8): u8 {
-    if is_upper(c) { ret c + 32; }
+    if is_upper(c) { ret c + ('a' - 'A'); }
     ret c;
 }
 
 fn to_upper(c: u8): u8 {
-    if is_lower(c) { ret c - 32; }
+    if is_lower(c) { ret c - ('a' - 'A'); }
     ret c;
 }
 
@@ -78,11 +78,11 @@ fn digit_value(c: u8, radix: u8): DigitResult {
 
     var value: u8 = 0;
     if is_digit(c) {
-        value = c - 48;
+        value = c - '0';
     } else if is_upper(c) {
-        value = c - 65 + 10;
+        value = c - 'A' + 10;
     } else if is_lower(c) {
-        value = c - 97 + 10;
+        value = c - 'a' + 10;
     } else {
         ret .Err(make(.InvalidDigit));
     }
@@ -94,9 +94,9 @@ fn digit_value(c: u8, radix: u8): DigitResult {
 // value 必须位于 0..35。uppercase 只影响 10..35 的字母形式。
 fn digit_char(value: u8, uppercase: bool): DigitResult {
     if value >= 36 { ret .Err(make(.OutOfRange)); }
-    if value < 10 { ret .Ok(value + 48); }
-    if uppercase { ret .Ok(value - 10 + 65); }
-    ret .Ok(value - 10 + 97);
+    if value < 10 { ret .Ok(value + '0'); }
+    if uppercase { ret .Ok(value - 10 + 'A'); }
+    ret .Ok(value - 10 + 'a');
 }
 
 fn expect_digit(result: DigitResult, expected: u8) {
@@ -141,27 +141,28 @@ fn test_all_bytes() {
         var c = i as u8;
         @assert(is_ascii(c) == (i <= 127), "ASCII boundary");
         @assert(is_control(c) == (i <= 31 || i == 127), "control boundary");
-        @assert(is_print(c) == (i >= 32 && i <= 126), "print boundary");
-        @assert(is_graph(c) == (i >= 33 && i <= 126), "graph boundary");
-        @assert(is_blank(c) == (i == 9 || i == 32), "blank set");
-        @assert(is_space(c) == (i == 9 || i == 10 || i == 11 || i == 12 || i == 13 || i == 32),
+        @assert(is_print(c) == (c >= ' ' && c <= '~'), "print boundary");
+        @assert(is_graph(c) == (c >= '!' && c <= '~'), "graph boundary");
+        @assert(is_blank(c) == (c == '\t' || c == ' '), "blank set");
+        @assert(is_space(c) == (c == '\t' || c == '\n' || c == '\x0B' || c == '\x0C' ||
+                                c == '\r' || c == ' '),
                 "space set");
-        @assert(is_digit(c) == (i >= 48 && i <= 57), "digit set");
-        @assert(is_upper(c) == (i >= 65 && i <= 90), "upper set");
-        @assert(is_lower(c) == (i >= 97 && i <= 122), "lower set");
-        @assert(is_hex_digit(c) == ((i >= 48 && i <= 57) ||
-                                    (i >= 65 && i <= 70) ||
-                                    (i >= 97 && i <= 102)),
+        @assert(is_digit(c) == (c >= '0' && c <= '9'), "digit set");
+        @assert(is_upper(c) == (c >= 'A' && c <= 'Z'), "upper set");
+        @assert(is_lower(c) == (c >= 'a' && c <= 'z'), "lower set");
+        @assert(is_hex_digit(c) == ((c >= '0' && c <= '9') ||
+                                    (c >= 'A' && c <= 'F') ||
+                                    (c >= 'a' && c <= 'f')),
                 "hex digit set");
         @assert(is_alpha(c) == (is_upper(c) || is_lower(c)), "alpha relation");
         @assert(is_alnum(c) == (is_alpha(c) || is_digit(c)), "alnum relation");
         @assert(is_punct(c) == (is_graph(c) && !is_alnum(c)), "punct relation");
 
         if is_upper(c) {
-            @assert(to_lower(c) == c + 32, "upper to lower");
+            @assert(to_lower(c) == c + ('a' - 'A'), "upper to lower");
             @assert(to_upper(to_lower(c)) == c, "case round trip");
         } else if is_lower(c) {
-            @assert(to_upper(c) == c - 32, "lower to upper");
+            @assert(to_upper(c) == c - ('a' - 'A'), "lower to upper");
             @assert(to_lower(to_upper(c)) == c, "case round trip");
         } else {
             @assert(to_lower(c) == c, "lower preserves non-letter");
@@ -192,20 +193,20 @@ fn test_digits() {
             if value < 10 {
                 @assert(lower == upper, "decimal digits ignore case");
             } else {
-                @assert(lower == upper + 32, "letter digit case distance");
+                @assert(lower == upper + ('a' - 'A'), "letter digit case distance");
             }
             value = value + 1;
         }
         radix = radix + 1;
     }
 
-    expect_error(digit_value(48, 0), true);
-    expect_error(digit_value(48, 1), true);
-    expect_error(digit_value(48, 37), true);
-    expect_error(digit_value(47, 10), false);
-    expect_error(digit_value(58, 10), false);
-    expect_error(digit_value(50, 2), false);
-    expect_error(digit_value(103, 16), false);
+    expect_error(digit_value('0', 0), true);
+    expect_error(digit_value('0', 1), true);
+    expect_error(digit_value('0', 37), true);
+    expect_error(digit_value('/', 10), false);
+    expect_error(digit_value(':', 10), false);
+    expect_error(digit_value('2', 2), false);
+    expect_error(digit_value('g', 16), false);
     expect_range_error(digit_char(36, false));
     expect_range_error(digit_char(255, true));
 }
diff --git a/stage2/utils/big_int.spl b/stage2/utils/big_int.spl
index a3a78cc..18a491c 100644
--- a/stage2/utils/big_int.spl
+++ b/stage2/utils/big_int.spl
@@ -528,8 +528,8 @@ fn parse(out: *BigInt, bytes: []u8, radix: u8): Status {
 
     var i: usize = 0;
     var negative = false;
-    if bytes.ptr[0] == 45 || bytes.ptr[0] == 43 {
-        negative = bytes.ptr[0] == 45;
+    if bytes.ptr[0] == '-' || bytes.ptr[0] == '+' {
+        negative = bytes.ptr[0] == '-';
         i = 1;
     }
     if i == bytes.len { ret .Err(Error.at(.UnexpectedEnd, i)); }
@@ -578,7 +578,7 @@ fn format(out: *ByteBuffer, source: BigInt, radix: u8, uppercase: bool): Status
     }
     magnitude.negative = false;
     if magnitude.len == 0 {
-        match ByteBuffer.append_byte(&reverse, 48) {
+        match ByteBuffer.append_byte(&reverse, '0') {
             .Ok => {}, .Err(error) => { drop(&magnitude); ret .Err(error); },
         }
     }
@@ -597,7 +597,7 @@ fn format(out: *ByteBuffer, source: BigInt, radix: u8, uppercase: bool): Status
         }
     }
     if source.negative {
-        match ByteBuffer.append_byte(&fresh, 45) {
+        match ByteBuffer.append_byte(&fresh, '-') {
             .Ok => {},
             .Err(error) => {
                 drop(&magnitude); ByteBuffer.drop(&reverse); ret .Err(error);
@@ -773,7 +773,7 @@ fn test_parse_format_and_conversion() {
     var output = ByteBuffer.new();
     expect_ok(format(&output, value, 16, false));
     var text = ByteBuffer.as_slice(output);
-    @assert(text.len == 9 && text.ptr[0] == 45 && text.ptr[1] == 100 && text.ptr[8] == 102);
+    @assert(text.len == 9 && text.ptr[0] == '-' && text.ptr[1] == 'd' && text.ptr[8] == 'f');
     expect_text_radix(value, 16, true, "-DEADBEEF");
 
     var base36 = zero();
diff --git a/stage2/utils/byte_buffer.spl b/stage2/utils/byte_buffer.spl
index c374a9d..e825876 100644
--- a/stage2/utils/byte_buffer.spl
+++ b/stage2/utils/byte_buffer.spl
@@ -211,7 +211,7 @@ fn test_append_and_resize() {
     @assert(capacity(buffer) >= 1);
 
     expect_ok(append(&buffer, "abc"));
-    expect_ok(append_byte(&buffer, 0));
+    expect_ok(append_byte(&buffer, '\0'));
     expect_ok(append(&buffer, "de"));
     @assert(buffer.storage.len == 6, "append logical length");
     var embedded: [6]u8 = .[97, 98, 99, 0, 100, 101];
diff --git a/stage2/utils/diag.spl b/stage2/utils/diag.spl
new file mode 100644
index 0000000..74c4a04
--- /dev/null
+++ b/stage2/utils/diag.spl
@@ -0,0 +1,243 @@
+// Expect retcode = 0
+// 诊断收集与渲染：severity + 文件/行/列 + 消息；统计错误/警告数。
+// 消息与文件名按拥有型 Str 保存（深拷贝），Diag 拥有它们直到 clear/drop。
+// 渲染输出到 ByteBuffer；不直接写文件，便于 LSP/测试。
+
+type .{ .Error, .Status, .make } = @import("error.spl");
+type .{ .Vec } = @import("vec.spl");
+type .{ .ByteBuffer } = @import("byte_buffer.spl");
+type .{ .Str } = @import("str.spl");
+type .{ .append_usize } = @import("fmt.spl");
+
+type Severity = enum {
+    Debug,
+    Info,
+    Warning,
+    Error,
+    Fatal,
+}
+
+type DiagItem = struct {
+    severity: Severity,
+    line: usize,
+    col: usize,
+    fname: Str,
+    message: Str,
+}
+
+type Diag = @self;
+
+items: Vec, // 元素类型 DiagItem
+errors: usize,
+warnings: usize,
+
+fn new(): Diag {
+    ret Diag {
+        .items = Vec.blank(@sizeof(DiagItem), 8),
+        .errors = 0,
+        .warnings = 0,
+    };
+}
+
+fn item_at(self: *Diag, index: usize): *DiagItem {
+    match Vec.get(&self.items, index) {
+        .Ok(ptr) => { ret ptr as *DiagItem; },
+        .Err(error) => { ret null; },
+    }
+    ret null;
+}
+
+fn count(self: Diag): usize {
+    ret self.items.len;
+}
+
+fn error_count(self: Diag): usize {
+    ret self.errors;
+}
+
+fn warning_count(self: Diag): usize {
+    ret self.warnings;
+}
+
+fn has_error(self: Diag): bool {
+    ret self.errors > 0;
+}
+
+// 添加一条诊断；失败不改变已有内容。
+fn emit(self: *Diag, fname: []u8, line: usize, col: usize, severity: Severity,
+        msg: []u8): Status {
+    var item = DiagItem {
+        .severity = severity,
+        .line = line,
+        .col = col,
+        .fname = Str.new(),
+        .message = Str.new(),
+    };
+
+    match Str.from_slice(fname) {
+        .Ok(v) => { item.fname = v; },
+        .Err(error) => { ret .Err(error); },
+    }
+    match Str.from_slice(msg) {
+        .Ok(v) => { item.message = v; },
+        .Err(error) => {
+            Str.drop(&item.fname);
+            ret .Err(error);
+        },
+    }
+    match Vec.push(&self.items, &item as *u8) {
+        .Ok => {},
+        .Err(error) => {
+            Str.drop(&item.fname);
+            Str.drop(&item.message);
+            ret .Err(error);
+        },
+    }
+
+    match severity {
+        .Error => { self.errors = self.errors + 1; },
+        .Fatal => { self.errors = self.errors + 1; },
+        .Warning => { self.warnings = self.warnings + 1; },
+        _ => {},
+    }
+    ret .Ok;
+}
+
+fn err(self: *Diag, fname: []u8, line: usize, col: usize, msg: []u8): Status {
+    ret emit(self, fname, line, col, .Error, msg);
+}
+
+fn warn(self: *Diag, fname: []u8, line: usize, col: usize, msg: []u8): Status {
+    ret emit(self, fname, line, col, .Warning, msg);
+}
+
+fn info(self: *Diag, fname: []u8, line: usize, col: usize, msg: []u8): Status {
+    ret emit(self, fname, line, col, .Info, msg);
+}
+
+fn level_name(severity: Severity): []u8 {
+    match severity {
+        .Debug => { ret "debug"; },
+        .Info => { ret "info"; },
+        .Warning => { ret "warning"; },
+        .Error => { ret "error"; },
+        .Fatal => { ret "fatal"; },
+    }
+    ret "?";
+}
+
+fn append_str(out: *ByteBuffer, s: *Str): Status {
+    ret ByteBuffer.append(out, Str.as_slice(s));
+}
+
+// 渲染为 "file:line:col: level: message\n"（file 为空时省略前缀）。
+fn render(self: *Diag, out: *ByteBuffer): Status {
+    var i: usize = 0;
+    while i < self.items.len {
+        var it = item_at(self, i);
+        if it == null { ret .Err(make(.OutOfRange)); }
+
+        if Str.size(it.fname) != 0 {
+            match append_str(out, &it.fname) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            match ByteBuffer.append_byte(out, ':') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        }
+        match append_usize(out, it.line, 10, false) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match ByteBuffer.append_byte(out, ':') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match append_usize(out, it.col, 10, false) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match ByteBuffer.append(out, ": ") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match ByteBuffer.append(out, level_name(it.severity)) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match ByteBuffer.append(out, ": ") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match append_str(out, &it.message) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match ByteBuffer.append_byte(out, '\n') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        i = i + 1;
+    }
+    ret .Ok;
+}
+
+// 丢弃所有条目内容但保留缓冲区。
+fn clear(self: *Diag) {
+    var i: usize = 0;
+    while i < self.items.len {
+        var it = item_at(self, i);
+        if it != null {
+            Str.drop(&it.fname);
+            Str.drop(&it.message);
+        }
+        i = i + 1;
+    }
+    Vec.clear(&self.items);
+    self.errors = 0;
+    self.warnings = 0;
+}
+
+fn drop(self: *Diag) {
+    var i: usize = 0;
+    while i < self.items.len {
+        var it = item_at(self, i);
+        if it != null {
+            Str.drop(&it.fname);
+            Str.drop(&it.message);
+        }
+        i = i + 1;
+    }
+    Vec.drop(&self.items);
+    self.errors = 0;
+    self.warnings = 0;
+}
+
+// ---- 测试辅助 ----
+
+fn expect_ok(result: Status) {
+    match result {
+        .Ok => {},
+        .Err(error) => { @assert(false, "unexpected diag error"); },
+    }
+}
+
+fn expect_bytes(out: *ByteBuffer, expected: []u8) {
+    var actual = ByteBuffer.as_slice(out.*);
+    @assert(actual.len == expected.len, "render length");
+    var i: usize = 0;
+    while i < expected.len {
+        @assert(actual.ptr[i] == expected.ptr[i], "render byte");
+        i = i + 1;
+    }
+}
+
+fn test_collect_and_count() {
+    var d = new();
+    @assert(count(d) == 0 && !has_error(d), "empty");
+
+    expect_ok(err(&d, "a.spl", 3, 5, "undefined identifier `x`"));
+    expect_ok(warn(&d, "a.spl", 7, 1, "unused variable"));
+    expect_ok(info(&d, "a.spl", 9, 2, "note"));
+    expect_ok(err(&d, "", 1, 1, "fatal-ish"));
+
+    @assert(count(d) == 4, "count");
+    @assert(error_count(d) == 2, "errors");
+    @assert(warning_count(d) == 1, "warnings");
+    @assert(has_error(d), "has_error");
+
+    var out = ByteBuffer.new();
+    expect_ok(render(&d, &out));
+    expect_bytes(&out, "a.spl:3:5: error: undefined identifier `x`\na.spl:7:1: warning: unused variable\na.spl:9:2: info: note\n1:1: error: fatal-ish\n");
+    ByteBuffer.drop(&out);
+    drop(&d);
+}
+
+fn test_clear_and_reuse() {
+    var d = new();
+    expect_ok(err(&d, "f", 1, 1, "one"));
+    expect_ok(warn(&d, "f", 2, 2, "two"));
+    clear(&d);
+    @assert(count(d) == 0 && !has_error(d) && warning_count(d) == 0, "cleared");
+
+    expect_ok(err(&d, "f", 3, 3, "three"));
+    @assert(count(d) == 1 && error_count(d) == 1, "reuse");
+    drop(&d);
+}
+
+fn main() {
+    test_collect_and_count();
+    test_clear_and_reuse();
+}
diff --git a/stage2/utils/fmt.spl b/stage2/utils/fmt.spl
index 79c7292..dc93fe9 100644
--- a/stage2/utils/fmt.spl
+++ b/stage2/utils/fmt.spl
@@ -19,7 +19,7 @@ fn append_u64(out: *ByteBuffer, value: u64, radix: u8, uppercase: bool): Status
     var count: usize = 0;
 
     if value == 0 {
-        digits[0] = 48;
+        digits[0] = '0';
         count = 1;
     } else {
         var v = value;
@@ -50,7 +50,7 @@ fn append_u64(out: *ByteBuffer, value: u64, radix: u8, uppercase: bool): Status
 fn append_i64(out: *ByteBuffer, value: i64, radix: u8, uppercase: bool): Status {
     if !valid_radix(radix) { ret .Err(make(.InvalidArgument)); }
     if value < 0 {
-        match ByteBuffer.append_byte(out, 45) { // '-'
+        match ByteBuffer.append_byte(out, '-') {
             .Ok => {},
             .Err(error) => { ret .Err(error); },
         }
@@ -221,14 +221,14 @@ fn test_composition_and_growth() {
     var i: i32 = 0;
     while i < 200 {
         expect_ok(append_i64(&b, i as i64, 10, false));
-        expect_ok(ByteBuffer.append_byte(&b, 44)); // ','
+        expect_ok(ByteBuffer.append_byte(&b, ',')); // ','
         i = i + 1;
     }
     expect_ok(append_i64(&b, 200, 10, false));
     @assert(ByteBuffer.size(b) > 0, "growth content");
     var s = ByteBuffer.as_slice(b);
-    @assert(s.ptr[0] == 48, "starts with 0");
-    @assert(s.ptr[1] == 44, "0 then comma");
+    @assert(s.ptr[0] == '0', "starts with 0");
+    @assert(s.ptr[1] == ',', "0 then comma");
 
     ByteBuffer.drop(&b);
 }
diff --git a/stage2/utils/frame.spl b/stage2/utils/frame.spl
new file mode 100644
index 0000000..e0cc6b5
--- /dev/null
+++ b/stage2/utils/frame.spl
@@ -0,0 +1,184 @@
+// Expect retcode = 0
+// LSP 传输帧：Content-Length 头 + 空行 + 正文。
+// 这里只做纯逻辑（写入/解析 ByteBuffer/ByteView），真正的 stdin/stdout I/O 由上层负责。
+
+type .{ .Error, .Status, .make } = @import("error.spl");
+type .{ .ByteView } = @import("byte_view.spl");
+type .{ .ByteBuffer } = @import("byte_buffer.spl");
+type .{ .append_usize } = @import("fmt.spl");
+
+type FrameResult = enum {
+    Ok: usize, // 消耗的字节数
+    Incomplete,
+    Err: Error,
+}
+
+fn ascii_lower(c: u8): u8 {
+    if c >= 'A' && c <= 'Z' { ret c + ('a' - 'A'); }
+    ret c;
+}
+
+// data[j..] 是否以 word 开头（ASCII 大小写不敏感）。
+fn ci_prefix(data: *u8, len: usize, j: usize, word: []u8): bool {
+    if j + word.len > len { ret false; }
+    var i: usize = 0;
+    while i < word.len {
+        if ascii_lower(data[j + i]) != ascii_lower(word.ptr[i]) { ret false; }
+        i = i + 1;
+    }
+    ret true;
+}
+
+fn write_frame(out: *ByteBuffer, body: []u8): Status {
+    match ByteBuffer.append(out, "Content-Length: ") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match append_usize(out, body.len, 10, false) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match ByteBuffer.append(out, "\r\n\r\n") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    ret ByteBuffer.append(out, body);
+}
+
+// 从 input 解析一个帧；正文复制到 body_out。缺数据返回 Incomplete。
+fn parse_frame(input: ByteView, body_out: *ByteBuffer): FrameResult {
+    var data = ByteView.as_slice(input);
+    var len = data.len;
+
+    // 找头结束标记 \r\n\r\n
+    var term = ~(0 as usize);
+    var i: usize = 0;
+    while i + 3 < len {
+        if data[i] == '\r' && data[i + 1] == '\n' && data[i + 2] == '\r' && data[i + 3] == '\n' {
+            term = i;
+            break;
+        }
+        i = i + 1;
+    }
+    if term == ~(0 as usize) { ret .Incomplete; }
+
+    // 在头内找 Content-Length（大小写不敏感）
+    var content_len: usize = 0;
+    var found = false;
+    var j: usize = 0;
+    while j + 15 <= term {
+        if ci_prefix(data.ptr, term, j, "content-length:") {
+            var k = j + 15;
+            while k < term && (data[k] == ' ' || data[k] == '\t') { k = k + 1; }
+            var n: usize = 0;
+            var digits = 0;
+            while k < term && data[k] >= '0' && data[k] <= '9' {
+                n = n * 10 + (data[k] - '0') as usize;
+                digits = digits + 1;
+                k = k + 1;
+            }
+            if digits == 0 { ret .Err(make(.InvalidDigit)); }
+            content_len = n;
+            found = true;
+            break;
+        }
+        j = j + 1;
+    }
+    if !found { ret .Err(make(.InvalidFormat)); }
+
+    var body_start = term + 4;
+    if len < body_start || content_len > len - body_start { ret .Incomplete; }
+
+    ByteBuffer.clear(body_out);
+    var body = data[body_start..body_start + content_len];
+    match ByteBuffer.append(body_out, body) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    ret .Ok(body_start + content_len);
+}
+
+// ---- 测试辅助 ----
+
+fn expect_ok(result: Status) {
+    match result {
+        .Ok => {},
+        .Err(error) => { @assert(false, "unexpected frame error"); },
+    }
+}
+
+fn expect_body(out: *ByteBuffer, expected: []u8) {
+    var actual = ByteBuffer.as_slice(out.*);
+    @assert(actual.len == expected.len, "body length");
+    var i: usize = 0;
+    while i < expected.len {
+        @assert(actual.ptr[i] == expected.ptr[i], "body byte");
+        i = i + 1;
+    }
+}
+
+fn test_roundtrip() {
+    var wire = ByteBuffer.new();
+    expect_ok(write_frame(&wire, "{\"jsonrpc\":\"2.0\"}"));
+    var bytes = ByteBuffer.as_slice(wire);
+    @assert(bytes.len == 16 + 2 + 4 + 17, "wire length"); // "Content-Length: 17\r\n\r\n" + body
+
+    var body = ByteBuffer.new();
+    match parse_frame(ByteView.from_slice(bytes), &body) {
+        .Ok(used) => {
+            @assert(used == bytes.len, "consumed all");
+            expect_body(&body, "{\"jsonrpc\":\"2.0\"}");
+        },
+        .Incomplete => { @assert(false, "unexpected incomplete"); },
+        .Err(error) => { @assert(false, "unexpected err"); },
+    }
+    ByteBuffer.drop(&body);
+    ByteBuffer.drop(&wire);
+}
+
+fn test_multiple_and_leftover() {
+    var wire = ByteBuffer.new();
+    expect_ok(write_frame(&wire, "abcd"));
+    expect_ok(write_frame(&wire, "ef"));
+    var bytes = ByteBuffer.as_slice(wire);
+
+    var body = ByteBuffer.new();
+    match parse_frame(ByteView.from_slice(bytes), &body) {
+        .Ok(used) => {
+            expect_body(&body, "abcd");
+            var rest = bytes[used..bytes.len];
+            match parse_frame(ByteView.from_slice(rest), &body) {
+                .Ok(used2) => { expect_body(&body, "ef"); @assert(used2 == rest.len, "second"); },
+                _ => { @assert(false, "second frame"); },
+            }
+        },
+        _ => { @assert(false, "first frame"); },
+    }
+    ByteBuffer.drop(&body);
+    ByteBuffer.drop(&wire);
+}
+
+fn test_incomplete_and_errors() {
+    var body = ByteBuffer.new();
+
+    // 头未结束
+    var h1 = "Content-Length: 3\r\n";
+    match parse_frame(ByteView.from_slice(h1), &body) {
+        .Incomplete => {},
+        _ => { @assert(false, "h1"); },
+    }
+    // 正文不足
+    var h2 = "Content-Length: 3\r\n\r\nab";
+    match parse_frame(ByteView.from_slice(h2), &body) {
+        .Incomplete => {},
+        _ => { @assert(false, "h2"); },
+    }
+    // 缺 Content-Length
+    var h3 = "X: 1\r\n\r\n";
+    match parse_frame(ByteView.from_slice(h3), &body) {
+        .Ok(used) => { @assert(false, "h3 ok"); },
+        .Incomplete => { @assert(false, "h3 incomplete"); },
+        .Err(error) => {},
+    }
+    // 大小写不敏感 + 额外头
+    var h4 = "X-Test: yes\r\ncontent-length: 2\r\n\r\nhi";
+    match parse_frame(ByteView.from_slice(h4), &body) {
+        .Ok(used) => { expect_body(&body, "hi"); },
+        _ => { @assert(false, "h4"); },
+    }
+    ByteBuffer.drop(&body);
+}
+
+fn main() {
+    test_roundtrip();
+    test_multiple_and_leftover();
+    test_incomplete_and_errors();
+}
diff --git a/stage2/utils/hash.spl b/stage2/utils/hash.spl
new file mode 100644
index 0000000..6b31505
--- /dev/null
+++ b/stage2/utils/hash.spl
@@ -0,0 +1,88 @@
+// Expect retcode = 0
+// 哈希工具：FNV-1a 字节哈希 + 整型混合；并给出可直接交给 map 的回调
+// （map 的 hash/eq 回调接收指向键字节的指针，这里按 usize / ByteView 解释）。
+// 仅提供确定性哈希与等值，不承诺密码学性质。
+
+type .{ .ByteView } = @import("byte_view.spl");
+
+// FNV-1a 64 位，逐字节；空输入返回 offset basis。
+fn hash_bytes(data: *u8, len: usize): usize {
+    var h: usize = 14695981039346656037;
+    var i: usize = 0;
+    while i < len {
+        h = (h ^ (data[i] as usize)) * 1099511628211;
+        i = i + 1;
+    }
+    ret h;
+}
+
+fn hash_byte_view(view: ByteView): usize {
+    ret hash_bytes(view.ptr, view.len);
+}
+
+// 整型混合（splitmix 风格的移位-乘法），保证非零与良好扩散。
+fn hash_usize_value(value: usize): usize {
+    var h = value;
+    h = (h ^ (h >> 30)) * 1099511628211;
+    h = h ^ (h >> 27);
+    h = h * 1099511628211;
+    h = h ^ (h >> 31);
+    if h == 0 { ret 1099511628211; } // 0 映射到非零
+    ret h;
+}
+
+// ---- map 回调 ----
+
+// 键按 usize 解释。
+fn hash_usize(key: *u8): usize {
+    ret hash_usize_value((key as *usize).*);
+}
+
+fn eq_usize(a: *u8, b: *u8): bool {
+    ret (a as *usize).* == (b as *usize).*;
+}
+
+// 键按 ByteView 解释，哈希其指向的字节。
+fn hash_view(key: *u8): usize {
+    var view = (key as *ByteView).*;
+    ret hash_byte_view(view);
+}
+
+fn eq_view(a: *u8, b: *u8): bool {
+    ret ByteView.eq((a as *ByteView).*, (b as *ByteView).*);
+}
+
+// ---- 测试辅助 ----
+
+fn expect_hash(view: []u8, expected: usize) {
+    var got = hash_bytes(view.ptr, view.len);
+    @assert(got == expected, "hash value");
+}
+
+fn test_fnv_vectors() {
+    expect_hash("", 14695981039346656037);
+    expect_hash("a", 12638187200555641996);
+    expect_hash("foobar", 9625390261332436968);
+    expect_hash("hello", 11831194018420276491);
+
+    // 同一输入稳定；不同输入不同。
+    var s1: []u8 = "abc";
+    var s2: []u8 = "abd";
+    @assert(hash_bytes(s1.ptr, 3) == hash_bytes(s1.ptr, 3), "deterministic");
+    @assert(hash_bytes(s1.ptr, 3) != hash_bytes(s2.ptr, 3), "distinct");
+
+    // ByteView 与 bytes 一致。
+    var v = ByteView.from_slice("foobar");
+    @assert(hash_byte_view(v) == hash_bytes(v.ptr, v.len), "view == bytes");
+
+    // 整型混合：0 不保持 0，且稳定/区分。
+    @assert(hash_usize_value(0) != 0, "zero mixed");
+    @assert(hash_usize_value(7) == hash_usize_value(7), "int stable");
+    @assert(hash_usize_value(7) != hash_usize_value(8), "int distinct");
+}
+
+// map 回调（hash_usize/eq_usize/hash_view/eq_view）由 map.spl 在自己的测试里验证。
+
+fn main() {
+    test_fnv_vectors();
+}
diff --git a/stage2/utils/io.spl b/stage2/utils/io.spl
new file mode 100644
index 0000000..90058b3
--- /dev/null
+++ b/stage2/utils/io.spl
@@ -0,0 +1,152 @@
+// Expect retcode = 0
+// I/O 便利层（构建在 sys/file 之上）：
+//   - C 字符串路径/模式的 open 包装
+//   - 整块读写（write_all / read_all）
+//   - stdout/stderr 打印
+// 说明：read_all 会按文件大小一次性读入 ByteBuffer；不处理增长中的文件。
+
+type .{ .Error, .Status, .make } = @import("error.spl");
+type .{ .ByteView } = @import("byte_view.spl");
+type .{ .ByteBuffer } = @import("byte_buffer.spl");
+type File = @import("../sys/file.spl");
+
+// 把 view 复制到以 NUL 结尾的临时缓冲，返回其 data（归属 tmp，调用后勿再用）。
+fn cstr(tmp: *ByteBuffer, view: ByteView): *u8 {
+    ByteBuffer.clear(tmp);
+    if view.len != 0 {
+        match ByteBuffer.append(tmp, ByteView.as_slice(view)) {
+            .Ok => {},
+            .Err(error) => { ret null; },
+        }
+    }
+    match ByteBuffer.append_byte(tmp, '\0') {
+        .Ok => {},
+        .Err(error) => { ret null; },
+    }
+    ret tmp.storage.data;
+}
+
+fn open_cstr(path: ByteView, mode: ByteView): File {
+    var pbuf = ByteBuffer.new();
+    var mbuf = ByteBuffer.new();
+    var p = cstr(&pbuf, path);
+    var m = cstr(&mbuf, mode);
+    var f = File.open(p, m);
+    ByteBuffer.drop(&pbuf);
+    ByteBuffer.drop(&mbuf);
+    ret f;
+}
+
+// 反复写入直到写完；返回实际写入字节数（!= bytes.len 表示中途失败）。
+fn write_all(f: *File, bytes: []u8): u64 {
+    var total: u64 = 0;
+    var want = bytes.len as u64;
+    while total < want {
+        var offset = total as usize;
+        var w = File.write(f, bytes.ptr + offset, want - total);
+        if w == 0 { break; }
+        total = total + w;
+    }
+    ret total;
+}
+
+fn print(bytes: []u8) {
+    var out = File.stdout();
+    write_all(&out, bytes);
+}
+
+fn println(bytes: []u8) {
+    print(bytes);
+    print("\n");
+}
+
+fn eprint(bytes: []u8) {
+    var err = File.stderr();
+    write_all(&err, bytes);
+}
+
+// 读取整个文件到 out（先清空）。打开失败或不完整返回 Err(.Io)。
+fn read_all(path: ByteView, out: *ByteBuffer): Status {
+    var f = open_cstr(path, ByteView.from_slice("rb"));
+    if !File.valid(&f) { ret .Err(make(.Io)); }
+
+    var sz = File.size(&f);
+    ByteBuffer.clear(out);
+    if sz != 0 {
+        match ByteBuffer.reserve(out, sz as usize) {
+            .Ok => {},
+            .Err(error) => { File.close(&f); ret .Err(error); },
+        }
+        var got = File.read(&f, out.storage.data, sz);
+        out.storage.len = got as usize;
+        File.close(&f);
+        if got != sz { ret .Err(make(.Io)); }
+        ret .Ok;
+    }
+    File.close(&f);
+    ret .Ok;
+}
+
+// ---- 测试辅助 ----
+
+fn view(text: []u8): ByteView {
+    ret ByteView.from_slice(text);
+}
+
+fn test_roundtrip() {
+    var path = view("build/_io_roundtrip.tmp");
+
+    var f = open_cstr(path, view("wb"));
+    @assert(File.valid(&f), "open wb");
+    var w = write_all(&f, "hello\nworld");
+    @assert(w == 11, "write_all length");
+    File.close(&f);
+
+    var out = ByteBuffer.new();
+    match read_all(path, &out) {
+        .Ok => {},
+        .Err(error) => { @assert(false, "read_all"); },
+    }
+    var got = ByteBuffer.as_slice(out);
+    @assert(got.len == 11, "roundtrip length");
+    @assert(got.ptr[0] == 'h' && got.ptr[5] == '\n' && got.ptr[10] == 'd', "roundtrip bytes");
+    ByteBuffer.drop(&out);
+}
+
+fn test_append_and_missing() {
+    var path = view("build/_io_append.tmp");
+
+    var f = open_cstr(path, view("wb"));
+    @assert(File.valid(&f), "open wb");
+    write_all(&f, "ab");
+    File.close(&f);
+
+    f = open_cstr(path, view("ab"));
+    @assert(File.valid(&f), "open ab");
+    var w2 = write_all(&f, "cd");
+    File.close(&f);
+    @assert(w2 == 2, "append length");
+
+    var out = ByteBuffer.new();
+    match read_all(path, &out) {
+        .Ok => {},
+        .Err(error) => { @assert(false, "read append"); },
+    }
+    var got = ByteBuffer.as_slice(out);
+    @assert(got.len == 4 && got.ptr[2] == 'c' && got.ptr[3] == 'd', "append content");
+
+    // 不存在的文件
+    match read_all(view("build/_io_no_such_file.tmp"), &out) {
+        .Ok => { @assert(false, "missing should fail"); },
+        .Err(error) => {
+            match error.kind { .Io => {}, _ => { @assert(false, "error kind"); } }
+        },
+    }
+    ByteBuffer.drop(&out);
+}
+
+fn main() {
+    test_roundtrip();
+    test_append_and_missing();
+    println("io: ok");
+}
diff --git a/stage2/utils/json.spl b/stage2/utils/json.spl
index 6c873f0..bfaf081 100644
--- a/stage2/utils/json.spl
+++ b/stage2/utils/json.spl
@@ -95,17 +95,24 @@ fn peek(p: *Parser): i32 {
     ret p.src.ptr[p.pos] as i32;
 }
 
+// 当前字符是否等于 ch（EOF 时 false）。
+fn at_char(p: *Parser, ch: u8): bool {
+    var c = peek(p);
+    if c < 0 { ret false; }
+    ret (c as u8) == ch;
+}
+
 fn parse_string_raw(p: *Parser): ViewResult {
-    if peek(p) != 34 { ret .Err(at(.InvalidFormat, p.pos)); }
+    if !at_char(p, '"') { ret .Err(at(.InvalidFormat, p.pos)); }
     p.pos = p.pos + 1;
     var start = p.pos;
     while p.pos < p.src.len {
         var c = p.src.ptr[p.pos];
-        if c == 34 { break; }
-        if c == 92 { // '\\'
+        if c == '"' { break; }
+        if c == '\\' { // '\\'
             p.pos = p.pos + 1;
             if p.pos >= p.src.len { ret .Err(at(.UnexpectedEnd, p.pos)); }
-            if p.src.ptr[p.pos] == 117 { // 'u'
+            if p.src.ptr[p.pos] == 'u' { // 'u'
                 var k: usize = 0;
                 while k < 4 {
                     p.pos = p.pos + 1;
@@ -141,18 +148,18 @@ fn parse_literal(p: *Parser, word: []u8, kind: JsonKind): NodeResult {
 
 fn parse_number(p: *Parser): NodeResult {
     var start = p.pos;
-    if peek(p) == 45 { p.pos = p.pos + 1; } // '-'
+    if at_char(p, '-') { p.pos = p.pos + 1; } // '-'
     if p.pos >= p.src.len || !is_digit(p.src.ptr[p.pos]) {
         ret .Err(at(.InvalidDigit, p.pos));
     }
-    if p.src.ptr[p.pos] == 48 { // 前导 0 后面不能接数字
+    if p.src.ptr[p.pos] == '0' { // 前导 0 后面不能接数字
         p.pos = p.pos + 1;
     } else {
         while p.pos < p.src.len && is_digit(p.src.ptr[p.pos]) {
             p.pos = p.pos + 1;
         }
     }
-    if p.pos < p.src.len && p.src.ptr[p.pos] == 46 { // '.'
+    if p.pos < p.src.len && p.src.ptr[p.pos] == '.' { // '.'
         p.pos = p.pos + 1;
         if p.pos >= p.src.len || !is_digit(p.src.ptr[p.pos]) {
             ret .Err(at(.InvalidDigit, p.pos));
@@ -161,9 +168,9 @@ fn parse_number(p: *Parser): NodeResult {
             p.pos = p.pos + 1;
         }
     }
-    if p.pos < p.src.len && (p.src.ptr[p.pos] == 101 || p.src.ptr[p.pos] == 69) { // e/E
+    if p.pos < p.src.len && (p.src.ptr[p.pos] == 'e' || p.src.ptr[p.pos] == 'E') { // e/E
         p.pos = p.pos + 1;
-        if p.pos < p.src.len && (p.src.ptr[p.pos] == 43 || p.src.ptr[p.pos] == 45) {
+        if p.pos < p.src.len && (p.src.ptr[p.pos] == '+' || p.src.ptr[p.pos] == '-') {
             p.pos = p.pos + 1;
         }
         if p.pos >= p.src.len || !is_digit(p.src.ptr[p.pos]) {
@@ -190,7 +197,7 @@ fn parse_array(p: *Parser): NodeResult {
     var kids = Vec.blank(@sizeof(JsonChild), 8);
 
     skip_ws(p);
-    if peek(p) == 93 { // ']'
+    if at_char(p, ']') { // ']'
         p.pos = p.pos + 1;
         var empty = blank_node(.Array);
         empty.children = kids;
@@ -211,12 +218,11 @@ fn parse_array(p: *Parser): NodeResult {
         }
 
         skip_ws(p);
-        var c = peek(p);
-        if c == 44 { // ','
+        if at_char(p, ',') { // ','
             p.pos = p.pos + 1;
             continue;
         }
-        if c == 93 { // ']'
+        if at_char(p, ']') { // ']'
             p.pos = p.pos + 1;
             break;
         }
@@ -234,7 +240,7 @@ fn parse_object(p: *Parser): NodeResult {
     var kids = Vec.blank(@sizeof(JsonChild), 8);
 
     skip_ws(p);
-    if peek(p) == 125 { // '}'
+    if at_char(p, '}') { // '}'
         p.pos = p.pos + 1;
         var empty = blank_node(.Object);
         empty.children = kids;
@@ -250,7 +256,7 @@ fn parse_object(p: *Parser): NodeResult {
         }
 
         skip_ws(p);
-        if peek(p) != 58 { // ':'
+        if !at_char(p, ':') { // ':'
             Vec.drop(&kids);
             ret .Err(at(.InvalidFormat, p.pos));
         }
@@ -270,12 +276,11 @@ fn parse_object(p: *Parser): NodeResult {
         }
 
         skip_ws(p);
-        var c = peek(p);
-        if c == 44 { // ','
+        if at_char(p, ',') { // ','
             p.pos = p.pos + 1;
             continue;
         }
-        if c == 125 { // '}'
+        if at_char(p, '}') { // '}'
             p.pos = p.pos + 1;
             break;
         }
@@ -294,7 +299,7 @@ fn parse_value(p: *Parser): NodeResult {
     if c < 0 { ret .Err(at(.UnexpectedEnd, p.pos)); }
     var ch = c as u8;
 
-    if ch == 34 { // '"'
+    if ch == '"' { // '"'
         var view = ByteView.empty();
         match parse_string_raw(p) {
             .Ok(v) => { view = v; },
@@ -304,11 +309,11 @@ fn parse_value(p: *Parser): NodeResult {
         node.text = view;
         ret intern_node(p, node);
     }
-    if ch == 123 { ret parse_object(p); } // '{'
-    if ch == 91 { ret parse_array(p); }   // '['
-    if ch == 116 { ret parse_literal(p, "true", .True); }
-    if ch == 102 { ret parse_literal(p, "false", .False); }
-    if ch == 110 { ret parse_literal(p, "null", .Null); }
+    if ch == '{' { ret parse_object(p); } // '{'
+    if ch == '[' { ret parse_array(p); }  // '['
+    if ch == 't' { ret parse_literal(p, "true", .True); }
+    if ch == 'f' { ret parse_literal(p, "false", .False); }
+    if ch == 'n' { ret parse_literal(p, "null", .Null); }
     ret parse_number(p);
 }
 
@@ -431,7 +436,7 @@ fn object_find(doc: *Json, index: usize, key: ByteView): FindResult {
 }
 
 fn emit_quoted(out: *ByteBuffer, text: ByteView): Status {
-    match ByteBuffer.append_byte(out, 34) { // '"'
+    match ByteBuffer.append_byte(out, '"') { // '"'
         .Ok => {},
         .Err(error) => { ret .Err(error); },
     }
@@ -439,11 +444,11 @@ fn emit_quoted(out: *ByteBuffer, text: ByteView): Status {
         .Ok => {},
         .Err(error) => { ret .Err(error); },
     }
-    ret ByteBuffer.append_byte(out, 34);
+    ret ByteBuffer.append_byte(out, '"');
 }
 
 fn emit_indent(out: *ByteBuffer, indent: usize, depth: usize): Status {
-    match ByteBuffer.append_byte(out, 10) { // '\n'
+    match ByteBuffer.append_byte(out, '\n') { // '\n'
         .Ok => {},
         .Err(error) => { ret .Err(error); },
     }
@@ -468,11 +473,11 @@ fn emit(out: *ByteBuffer, doc: *Json, index: usize, indent: usize, depth: usize)
 
 fn emit_children(out: *ByteBuffer, doc: *Json, n: *JsonNode, is_object: bool, indent: usize,
                  depth: usize): Status {
-    var open_byte: u8 = 91;  // '['
-    var close_byte: u8 = 93; // ']'
+    var open_byte: u8 = '[';  // '['
+    var close_byte: u8 = ']'; // ']'
     if is_object {
-        open_byte = 123;  // '{'
-        close_byte = 125; // '}'
+        open_byte = '{';  // '{'
+        close_byte = '}'; // '}'
     }
 
     match ByteBuffer.append_byte(out, open_byte) {
@@ -488,7 +493,7 @@ fn emit_children(out: *ByteBuffer, doc: *Json, n: *JsonNode, is_object: bool, in
     var i: usize = 0;
     while i < count {
         if i != 0 {
-            match ByteBuffer.append_byte(out, 44) { // ','
+            match ByteBuffer.append_byte(out, ',') { // ','
                 .Ok => {},
                 .Err(error) => { ret .Err(error); },
             }
@@ -508,12 +513,12 @@ fn emit_children(out: *ByteBuffer, doc: *Json, n: *JsonNode, is_object: bool, in
                         .Ok => {},
                         .Err(error) => { ret .Err(error); },
                     }
-                    match ByteBuffer.append_byte(out, 58) { // ':'
+                    match ByteBuffer.append_byte(out, ':') { // ':'
                         .Ok => {},
                         .Err(error) => { ret .Err(error); },
                     }
                     if indent != 0 {
-                        match ByteBuffer.append_byte(out, 32) { // ' '
+                        match ByteBuffer.append_byte(out, ' ') { // ' '
                             .Ok => {},
                             .Err(error) => { ret .Err(error); },
                         }
@@ -746,14 +751,14 @@ fn test_construction_helpers() {
 
 fn test_large_array() {
     var out = ByteBuffer.new();
-    expect_ok(ByteBuffer.append_byte(&out, 91)); // '['
+    expect_ok(ByteBuffer.append_byte(&out, '[')); // '['
     var i: usize = 0;
     while i < 1000 {
-        if i != 0 { expect_ok(ByteBuffer.append_byte(&out, 44)); }
-        expect_ok(ByteBuffer.append_byte(&out, 49)); // '1'
+        if i != 0 { expect_ok(ByteBuffer.append_byte(&out, ',')); }
+        expect_ok(ByteBuffer.append_byte(&out, '1')); // '1'
         i = i + 1;
     }
-    expect_ok(ByteBuffer.append_byte(&out, 93)); // ']'
+    expect_ok(ByteBuffer.append_byte(&out, ']')); // ']'
 
     var text = ByteBuffer.as_slice(out);
     var r = parse(ByteView.from_slice(text));
diff --git a/stage2/utils/map.spl b/stage2/utils/map.spl
index bc98544..ed9670e 100644
--- a/stage2/utils/map.spl
+++ b/stage2/utils/map.spl
@@ -6,6 +6,7 @@
 type .{ .Error, .Status, .make } = @import("error.spl");
 type .{ .Vec } = @import("vec.spl");
 type .{ .ByteView } = @import("byte_view.spl");
+type .{ .hash_usize, .eq_usize, .hash_view, .eq_view } = @import("hash.spl");
 type .{ .copy } = @import("../sys/mem.spl");
 
 // 槽位状态：0=EMPTY，1=OCCUPIED，2=DELETED（墓碑，仅在扩容重哈希时清除）。
@@ -239,15 +240,6 @@ fn take(source: *Map): Map {
 
 // ---- 测试辅助 ----
 
-fn hash_usize(key: *u8): usize {
-    var value = (key as *usize).*;
-    ret value * 2654435761;
-}
-
-fn eq_usize(a: *u8, b: *u8): bool {
-    ret (a as *usize).* == (b as *usize).*;
-}
-
 fn expect_ok(result: Status) {
     match result {
         .Ok => {},
@@ -391,21 +383,6 @@ fn test_take_and_self_put() {
     drop(&moved);
 }
 
-fn hash_view(key: *u8): usize {
-    var view = (key as *ByteView).*;
-    var h: usize = 14695981039346656037;
-    var i: usize = 0;
-    while i < view.len {
-        h = (h ^ (view.ptr[i] as usize)) * 1099511628211;
-        i = i + 1;
-    }
-    ret h;
-}
-
-fn eq_view(a: *u8, b: *u8): bool {
-    ret ByteView.eq((a as *ByteView).*, (b as *ByteView).*);
-}
-
 fn test_byte_view_keys() {
     var m = new(@sizeof(ByteView), @alignof(ByteView), @sizeof(i32), @alignof(i32),
                 hash_view, eq_view);
diff --git a/stage2/utils/path.spl b/stage2/utils/path.spl
new file mode 100644
index 0000000..3ba5cbe
--- /dev/null
+++ b/stage2/utils/path.spl
@@ -0,0 +1,231 @@
+// Expect retcode = 0
+// 路径工具（不访问文件系统）：同时接受 '/' 与 '\\'；支持 Windows 盘符 "C:"。
+// 输入为借用 ByteView，输出写入调用方的 ByteBuffer。
+// 注意：normalize 对相对路径的前导 ".." 采取丢弃（不保留），适用于导入路径场景。
+
+type .{ .Error, .Status, .make } = @import("error.spl");
+type .{ .ByteView } = @import("byte_view.spl");
+type .{ .ByteBuffer } = @import("byte_buffer.spl");
+type .{ .Vec } = @import("vec.spl");
+
+fn is_sep(c: u8): bool {
+    ret c == '/' || c == '\\'; // '/' '\\'
+}
+
+fn is_drive(path: ByteView): bool {
+    ret path.len >= 2 && path.ptr[1] == ':'; // ':'
+}
+
+fn is_absolute(path: ByteView): bool {
+    if path.len == 0 { ret false; }
+    if is_sep(path.ptr[0]) { ret true; }
+    ret is_drive(path);
+}
+
+// 最后一个分隔符下标；无返回 ~0。
+fn last_sep(path: ByteView): usize {
+    var i = path.len;
+    while i > 0 {
+        i = i - 1;
+        if is_sep(path.ptr[i]) { ret i; }
+    }
+    ret ~(0 as usize);
+}
+
+// 目录部分（不含末尾分隔符）；根 "/" 返回 "/"；无分隔符返回空。
+fn dirname(path: ByteView): ByteView {
+    var idx = last_sep(path);
+    if idx == ~(0 as usize) { ret ByteView.empty(); }
+    if idx == 0 {
+        var r = ByteView.slice(path, 0, 1);
+        match r {
+            .Ok(v) => { ret v; },
+            .Err(error) => { ret ByteView.empty(); },
+        }
+    }
+    var r = ByteView.slice(path, 0, idx);
+    match r {
+        .Ok(v) => { ret v; },
+        .Err(error) => { ret ByteView.empty(); },
+    }
+    ret ByteView.empty();
+}
+
+fn basename(path: ByteView): ByteView {
+    var idx = last_sep(path);
+    if idx == ~(0 as usize) { ret path; }
+    var r = ByteView.slice(path, idx + 1, path.len);
+    match r {
+        .Ok(v) => { ret v; },
+        .Err(error) => { ret ByteView.empty(); },
+    }
+    ret ByteView.empty();
+}
+
+// 扩展名（不含 '.'）；无扩展名返回空。
+fn extension(path: ByteView): ByteView {
+    var base = basename(path);
+    var i = base.len;
+    while i > 0 {
+        i = i - 1;
+        if base.ptr[i] == '.' { // '.'
+            if i == 0 { ret ByteView.empty(); }
+            var r = ByteView.slice(base, i + 1, base.len);
+            match r {
+                .Ok(v) => { ret v; },
+                .Err(error) => { ret ByteView.empty(); },
+            }
+        }
+    }
+    ret ByteView.empty();
+}
+
+// out = dir + sep + rel（rel 绝对时直接使用 rel）；先清空 out。
+fn join_into(out: *ByteBuffer, dir: ByteView, rel: ByteView): Status {
+    ByteBuffer.clear(out);
+    if is_absolute(rel) {
+        ret ByteBuffer.append(out, ByteView.as_slice(rel));
+    }
+    match ByteBuffer.append(out, ByteView.as_slice(dir)) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    if dir.len != 0 && !is_sep(dir.ptr[dir.len - 1]) {
+        match ByteBuffer.append_byte(out, '/') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    }
+    ret ByteBuffer.append(out, ByteView.as_slice(rel));
+}
+
+// 折叠重复分隔符、'.'，及'..'（回退上一段；相对路径前导 '..' 丢弃）。
+fn normalize_into(out: *ByteBuffer, path: ByteView): Status {
+    ByteBuffer.clear(out);
+    if path.len == 0 { ret .Ok; }
+
+    var starts = Vec.blank(@sizeof(usize), 8);
+    var i: usize = 0;
+
+    if is_sep(path.ptr[0]) {
+        match ByteBuffer.append_byte(out, '/') { .Ok => {}, .Err(e) => { Vec.drop(&starts); ret .Err(e); } }
+    } else if is_drive(path) {
+        match ByteBuffer.append_byte(out, path.ptr[0]) { .Ok => {}, .Err(e) => { Vec.drop(&starts); ret .Err(e); } }
+        match ByteBuffer.append_byte(out, ':') { .Ok => {}, .Err(e) => { Vec.drop(&starts); ret .Err(e); } }
+        i = 2; // 跳过盘符
+    }
+
+    while i < path.len {
+        while i < path.len && is_sep(path.ptr[i]) { i = i + 1; }
+        if i >= path.len { break; }
+        var start = i;
+        while i < path.len && !is_sep(path.ptr[i]) { i = i + 1; }
+
+        var seg_result = ByteView.slice(path, start, i);
+        var seg = ByteView.empty();
+        match seg_result {
+            .Ok(v) => { seg = v; },
+            .Err(error) => { Vec.drop(&starts); ret .Err(error); },
+        }
+
+        if ByteView.eq(seg, ByteView.from_slice(".")) { continue; }
+        if ByteView.eq(seg, ByteView.from_slice("..")) {
+            if starts.len > 0 {
+                var idx = starts.len - 1;
+                match Vec.get(&starts, idx) {
+                    .Ok(p) => { out.storage.len = (p as *usize).*; },
+                    .Err(error) => {},
+                }
+                starts.len = starts.len - 1;
+            }
+            continue;
+        }
+
+        var comp_start = out.storage.len;
+        if comp_start > 0 && !is_sep(out.storage.data[comp_start - 1]) {
+            match ByteBuffer.append_byte(out, '/') { .Ok => {}, .Err(e) => { Vec.drop(&starts); ret .Err(e); } }
+        }
+        var recorded = comp_start;
+        match Vec.push(&starts, &recorded as *u8) {
+            .Ok => {},
+            .Err(error) => { Vec.drop(&starts); ret .Err(error); },
+        }
+        match ByteBuffer.append(out, ByteView.as_slice(seg)) {
+            .Ok => {},
+            .Err(error) => { Vec.drop(&starts); ret .Err(error); },
+        }
+    }
+
+    Vec.drop(&starts);
+    ret .Ok;
+}
+
+// ---- 测试辅助 ----
+
+fn expect_ok(result: Status) {
+    match result {
+        .Ok => {},
+        .Err(error) => { @assert(false, "unexpected path error"); },
+    }
+}
+
+fn view(text: []u8): ByteView {
+    ret ByteView.from_slice(text);
+}
+
+fn expect_view(got: ByteView, expected: []u8) {
+    @assert(ByteView.eq(got, view(expected)), "view value");
+}
+
+fn expect_out(out: *ByteBuffer, expected: []u8) {
+    var actual = ByteBuffer.as_slice(out.*);
+    @assert(actual.len == expected.len, "out length");
+    var i: usize = 0;
+    while i < expected.len {
+        @assert(actual.ptr[i] == expected.ptr[i], "out byte");
+        i = i + 1;
+    }
+}
+
+fn test_parts() {
+    @assert(is_sep('/') && is_sep('\\'), "seps");
+    @assert(is_absolute(view("/a/b")), "posix abs");
+    @assert(is_absolute(view("\\a")), "win abs");
+    @assert(is_absolute(view("C:/a")), "drive abs");
+    @assert(!is_absolute(view("a/b")), "relative");
+
+    expect_view(dirname(view("/a/b/c")), "/a/b");
+    expect_view(dirname(view("/a")), "/");
+    expect_view(dirname(view("abc")), "");
+    expect_view(basename(view("/a/b/c")), "c");
+    expect_view(basename(view("abc")), "abc");
+    expect_view(extension(view("a/b.spl")), "spl");
+    expect_view(extension(view("a/b")), "");
+    expect_view(extension(view("a/.hidden")), "");
+}
+
+fn test_join() {
+    var out = ByteBuffer.new();
+    expect_ok(join_into(&out, view("stage2/utils"), view("vec.spl")));
+    expect_out(&out, "stage2/utils/vec.spl");
+    expect_ok(join_into(&out, view("a/"), view("b.spl")));
+    expect_out(&out, "a/b.spl");
+    expect_ok(join_into(&out, view("ignored"), view("/abs/x")));
+    expect_out(&out, "/abs/x");
+    ByteBuffer.drop(&out);
+}
+
+fn test_normalize() {
+    var out = ByteBuffer.new();
+    expect_ok(normalize_into(&out, view("a/b/../c")));
+    expect_out(&out, "a/c");
+    expect_ok(normalize_into(&out, view("./a//b")));
+    expect_out(&out, "a/b");
+    expect_ok(normalize_into(&out, view("/a/./b/../c")));
+    expect_out(&out, "/a/c");
+    expect_ok(normalize_into(&out, view("C:/a/../b")));
+    expect_out(&out, "C:/b");
+    expect_ok(normalize_into(&out, view("a/..")));
+    expect_out(&out, "");
+    ByteBuffer.drop(&out);
+}
+
+fn main() {
+    test_parts();
+    test_join();
+    test_normalize();
+}
diff --git a/stage2/utils/source.spl b/stage2/utils/source.spl
new file mode 100644
index 0000000..7a7c388
--- /dev/null
+++ b/stage2/utils/source.spl
@@ -0,0 +1,258 @@
+// Expect retcode = 0
+// 源文件与位置映射：fname + text 借用调用方存储；行起始表由 Source 拥有。
+// 提供字节偏移 <-> (行,列)（行列均为 1-based、按字节计），以及取整行文本。
+// Sources 是按 fname 索引的多文件集合。
+
+type .{ .Error, .Status, .make } = @import("error.spl");
+type .{ .ByteView, .FindResult } = @import("byte_view.spl");
+type .{ .Vec } = @import("vec.spl");
+type .{ .Map } = @import("map.spl");
+type .{ .hash_view, .eq_view } = @import("hash.spl");
+
+type Position = struct {
+    line: usize, // 1-based
+    col: usize,  // 1-based，按字节
+}
+
+type Source = @self;
+
+fname: ByteView,    // 借用
+text: ByteView,     // 借用
+line_starts: Vec,   // usize：每行起始字节偏移（含首行 0）
+
+type SourceResult = enum {
+    Ok: Source,
+    Err: Error,
+}
+
+type AddResult = enum {
+    Ok: usize,
+    Err: Error,
+}
+
+fn blank(): Source {
+    ret Source {
+        .fname = ByteView.empty(),
+        .text = ByteView.empty(),
+        .line_starts = Vec.blank(@sizeof(usize), 8),
+    };
+}
+
+fn new(fname: ByteView, text: ByteView): SourceResult {
+    var starts = Vec.blank(@sizeof(usize), 8);
+    var zero: usize = 0;
+    match Vec.push(&starts, &zero as *u8) {
+        .Ok => {},
+        .Err(error) => { ret .Err(error); },
+    }
+
+    var i: usize = 0;
+    while i < text.len {
+        if text.ptr[i] == '\n' {
+            var next = i + 1;
+            match Vec.push(&starts, &next as *u8) {
+                .Ok => {},
+                .Err(error) => { Vec.drop(&starts); ret .Err(error); },
+            }
+        }
+        i = i + 1;
+    }
+
+    ret .Ok(Source { .fname = fname, .text = text, .line_starts = starts });
+}
+
+fn line_count(self: Source): usize {
+    ret self.line_starts.len;
+}
+
+fn line_start_offset(self: Source, line: usize): usize {
+    if line == 0 || line > self.line_starts.len { ret ~(0 as usize); }
+    match Vec.get(&self.line_starts, line - 1) {
+        .Ok(ptr) => { ret (ptr as *usize).*; },
+        .Err(error) => { ret ~(0 as usize); },
+    }
+    ret ~(0 as usize);
+}
+
+// offset 所在行（0-based）；超出范围夹到末行。
+fn line_index_at(self: Source, offset: usize): usize {
+    var off = offset;
+    if off > self.text.len { off = self.text.len; }
+    var lo: usize = 0;
+    var hi: usize = self.line_starts.len;
+    while lo + 1 < hi {
+        var mid = (lo + hi) / 2;
+        var start = line_start_offset(self, mid + 1); // 1-based -> 内部
+        if start != ~(0 as usize) && start <= off { lo = mid; } else { hi = mid; }
+    }
+    ret lo;
+}
+
+fn line_col(self: Source, offset: usize): Position {
+    var off = offset;
+    if off > self.text.len { off = self.text.len; }
+    var line0 = line_index_at(self, off);
+    var start = line_start_offset(self, line0 + 1); // 1-based
+    if start == ~(0 as usize) { start = 0; }
+    ret Position { .line = line0 + 1, .col = off - start + 1 };
+}
+
+// line 为 1-based；返回整行（去掉行尾 '\n' 与 '\r'）。
+fn line_text(self: Source, line: usize): ByteView {
+    var n = self.line_starts.len;
+    if line == 0 || line > n { ret ByteView.empty(); }
+    var start = line_start_offset(self, line);
+    if start == ~(0 as usize) { ret ByteView.empty(); }
+    var end = self.text.len;
+    if line < n {
+        var next = line_start_offset(self, line + 1);
+        if next != ~(0 as usize) { end = next; }
+    }
+    while end > start && (self.text.ptr[end - 1] == '\n' || self.text.ptr[end - 1] == '\r') {
+        end = end - 1;
+    }
+    var r = ByteView.slice(self.text, start, end);
+    match r {
+        .Ok(v) => { ret v; },
+        .Err(error) => { ret ByteView.empty(); },
+    }
+    ret ByteView.empty();
+}
+
+fn drop(self: *Source) {
+    Vec.drop(&self.line_starts);
+}
+
+// ---- 多文件集合 ----
+
+type Sources = struct {
+    items: Vec,   // Source
+    by_name: Map, // ByteView(fname) -> usize
+}
+
+fn set_new(): Sources {
+    ret Sources {
+        .items = Vec.blank(@sizeof(Source), 8),
+        .by_name = Map.blank(@sizeof(ByteView), @alignof(ByteView), @sizeof(usize),
+                             @alignof(usize), hash_view, eq_view),
+    };
+}
+
+fn set_add(set: *Sources, fname: ByteView, text: ByteView): AddResult {
+    var src = new(fname, text);
+    var s = blank();
+    match src {
+        .Ok(v) => { s = v; },
+        .Err(error) => { ret .Err(error); },
+    }
+
+    var index = set.items.len;
+    match Vec.push(&set.items, &s as *u8) {
+        .Ok => {},
+        .Err(error) => { drop(&s); ret .Err(error); },
+    }
+    match Map.put(&set.by_name, &fname as *u8, &index as *u8) {
+        .Ok => {},
+        .Err(error) => {
+            // 回滚最后一项
+            set.items.len = index;
+            drop(&s);
+            ret .Err(error);
+        },
+    }
+    ret .Ok(index);
+}
+
+fn set_at(set: *Sources, index: usize): *Source {
+    match Vec.get(&set.items, index) {
+        .Ok(ptr) => { ret ptr as *Source; },
+        .Err(error) => { ret null; },
+    }
+    ret null;
+}
+
+fn set_find(set: *Sources, fname: ByteView): FindResult {
+    var out: usize = 0;
+    if Map.get(&set.by_name, &fname as *u8, &out as *u8) { ret .Found(out); }
+    ret .NotFound;
+}
+
+fn set_drop(set: *Sources) {
+    var i: usize = 0;
+    while i < set.items.len {
+        var s = set_at(set, i);
+        if s != null { drop(s); }
+        i = i + 1;
+    }
+    Vec.drop(&set.items);
+    Map.drop(&set.by_name);
+}
+
+// ---- 测试辅助 ----
+
+fn view(text: []u8): ByteView {
+    ret ByteView.from_slice(text);
+}
+
+fn test_single() {
+    var text = "line1\nline2\n\nline4";
+    var r = new(view("a.spl"), view(text));
+    var src = blank();
+    match r {
+        .Ok(v) => { src = v; },
+        .Err(error) => { @assert(false, "source new"); },
+    }
+
+    @assert(line_count(src) == 4, "line count");
+
+    var p0 = line_col(src, 0);
+    @assert(p0.line == 1 && p0.col == 1, "start");
+    var p6 = line_col(src, 6);
+    @assert(p6.line == 2 && p6.col == 1, "line2 start");
+    var p8 = line_col(src, 8);
+    @assert(p8.line == 2 && p8.col == 3, "line2 col3");
+    var p12 = line_col(src, 12);
+    @assert(p12.line == 3 && p12.col == 1, "empty line3");
+    var p13 = line_col(src, 13);
+    @assert(p13.line == 4 && p13.col == 1, "line4 start");
+    var p18 = line_col(src, 18);
+    @assert(p18.line == 4 && p18.col == 6, "past end clamps");
+
+    var l1 = line_text(src, 1);
+    var l3 = line_text(src, 3);
+    var l4 = line_text(src, 4);
+    @assert(ByteView.eq(l1, view("line1")), "line1");
+    @assert(ByteView.eq(l3, view("")), "line3 empty");
+    @assert(ByteView.eq(l4, view("line4")), "line4");
+
+    drop(&src);
+}
+
+fn test_sources() {
+    var set = set_new();
+    var r1 = set_add(&set, view("a.spl"), view("x\ny"));
+    var r2 = set_add(&set, view("b.spl"), view("z"));
+    var a: usize = 0;
+    var b: usize = 0;
+    match r1 { .Ok(v) => { a = v; }, .Err(e) => { @assert(false, "add a"); } }
+    match r2 { .Ok(v) => { b = v; }, .Err(e) => { @assert(false, "add b"); } }
+    @assert(a == 0 && b == 1, "indices");
+
+    match set_find(&set, view("b.spl")) {
+        .Found(i) => { @assert(i == 1, "find b"); },
+        .NotFound => { @assert(false, "b missing"); },
+    }
+    match set_find(&set, view("c.spl")) {
+        .Found(i) => { @assert(false, "c unexpected"); },
+        .NotFound => {},
+    }
+
+    var sb = set_at(&set, b);
+    @assert(sb != null && line_count(sb.*) == 1, "at b");
+    set_drop(&set);
+}
+
+fn main() {
+    test_single();
+    test_sources();
+}
diff --git a/stage2/utils/str.spl b/stage2/utils/str.spl
new file mode 100644
index 0000000..889126b
--- /dev/null
+++ b/stage2/utils/str.spl
@@ -0,0 +1,238 @@
+// Expect retcode = 0
+// 拥有型字节字符串：在 ByteBuffer 之上提供字符串常用操作。
+// 普通赋值复制句柄（共享存储），所有权转移用 take；失败不产生部分状态。
+
+type .{ .Error, .Status, .make } = @import("error.spl");
+type .{ .ByteView, .FindResult } = @import("byte_view.spl");
+type .{ .ByteBuffer } = @import("byte_buffer.spl");
+
+type Str = @self;
+
+storage: ByteBuffer,
+
+type StrResult = enum {
+    Ok: Str,
+    Err: Error,
+}
+
+fn new(): Str {
+    ret Str { .storage = ByteBuffer.new() };
+}
+
+// 复制 view 指向的字节。
+fn from_view(view: ByteView): StrResult {
+    var s = new();
+    match append_view(&s, view) {
+        .Ok => {},
+        .Err(error) => { drop(&s); ret .Err(error); },
+    }
+    ret .Ok(s);
+}
+
+fn from_slice(bytes: []u8): StrResult {
+    ret from_view(ByteView.from_slice(bytes));
+}
+
+fn size(self: Str): usize {
+    ret ByteBuffer.size(self.storage);
+}
+
+fn is_empty(self: Str): bool {
+    ret ByteBuffer.is_empty(self.storage);
+}
+
+fn as_view(self: *Str): ByteView {
+    ret ByteView.from_slice(ByteBuffer.as_slice(self.storage));
+}
+
+fn as_slice(self: *Str): []u8 {
+    ret ByteBuffer.as_slice(self.storage);
+}
+
+fn reserve(self: *Str, capacity: usize): Status {
+    ret ByteBuffer.reserve(&self.storage, capacity);
+}
+
+fn append_view(self: *Str, view: ByteView): Status {
+    ret ByteBuffer.append(&self.storage, ByteView.as_slice(view));
+}
+
+fn append(self: *Str, bytes: []u8): Status {
+    ret ByteBuffer.append(&self.storage, bytes);
+}
+
+fn append_str(self: *Str, other: Str): Status {
+    var o = other;
+    ret append_view(self, as_view(&o));
+}
+
+fn append_byte(self: *Str, byte: u8): Status {
+    ret ByteBuffer.append_byte(&self.storage, byte);
+}
+
+fn clear(self: *Str) {
+    ByteBuffer.clear(&self.storage);
+}
+
+fn eq(a: Str, b: Str): bool {
+    var x = a;
+    var y = b;
+    ret ByteView.eq(as_view(&x), as_view(&y));
+}
+
+fn eq_slice(a: Str, bytes: []u8): bool {
+    var x = a;
+    ret ByteView.eq(as_view(&x), ByteView.from_slice(bytes));
+}
+
+fn cmp(a: Str, b: Str): i32 {
+    var x = a;
+    var y = b;
+    ret ByteView.cmp(as_view(&x), as_view(&y));
+}
+
+fn starts_with(self: Str, prefix: Str): bool {
+    var x = self;
+    var p = prefix;
+    ret ByteView.starts_with(as_view(&x), as_view(&p));
+}
+
+fn ends_with(self: Str, suffix: Str): bool {
+    var x = self;
+    var s = suffix;
+    ret ByteView.ends_with(as_view(&x), as_view(&s));
+}
+
+fn find(self: Str, needle: Str): FindResult {
+    var x = self;
+    var n = needle;
+    ret ByteView.find(as_view(&x), as_view(&n));
+}
+
+// 深复制；允许 out 与 source 为同一对象。
+fn clone(out: *Str, source: Str): Status {
+    ret ByteBuffer.clone(&out.storage, source.storage);
+}
+
+fn take(source: *Str): Str {
+    ret Str { .storage = ByteBuffer.take(&source.storage) };
+}
+
+fn drop(self: *Str) {
+    ByteBuffer.drop(&self.storage);
+}
+
+// ---- 测试辅助 ----
+
+fn expect_ok(result: Status) {
+    match result {
+        .Ok => {},
+        .Err(error) => { @assert(false, "unexpected str error"); },
+    }
+}
+
+fn expect_text(s: *Str, expected: []u8) {
+    var view = as_view(s);
+    @assert(ByteView.eq(view, ByteView.from_slice(expected)), "str text");
+}
+
+fn test_construction() {
+    var s = new();
+    @assert(is_empty(s) && size(s) == 0, "empty new");
+    expect_ok(append(&s, "hello"));
+    expect_ok(append_byte(&s, ' '));
+    expect_ok(append(&s, "world"));
+    expect_text(&s, "hello world");
+    @assert(size(s) == 11, "size");
+
+    var r = from_slice("copy");
+    match r {
+        .Ok(copy) => {
+            expect_text(&copy, "copy");
+            @assert(eq(copy, copy), "self eq");
+            drop(&copy);
+        },
+        .Err(error) => { @assert(false, "from_slice"); },
+    }
+    drop(&s);
+}
+
+fn test_compare_and_search() {
+    var hello_r = from_slice("hello");
+    var hello: Str = new();
+    match hello_r {
+        .Ok(v) => { hello = v; },
+        .Err(error) => { @assert(false, "hello"); },
+    }
+    var help_r = from_slice("help");
+    var help: Str = new();
+    match help_r {
+        .Ok(v) => { help = v; },
+        .Err(error) => { @assert(false, "help"); },
+    }
+
+    @assert(eq(hello, hello), "eq");
+    @assert(!eq(hello, help), "ne");
+    @assert(eq_slice(hello, "hello"), "eq_slice");
+    @assert(cmp(hello, help) < 0, "cmp lt");
+    @assert(cmp(help, hello) > 0, "cmp gt");
+
+    @assert(starts_with(hello, hello), "starts_with");
+    @assert(ends_with(hello, hello), "ends_with");
+    var pre_r = from_slice("he");
+    var pre: Str = new();
+    match pre_r { .Ok(v) => { pre = v; }, .Err(e) => { @assert(false, "pre"); } }
+    @assert(starts_with(hello, pre), "starts he");
+    @assert(!ends_with(hello, pre), "!ends he");
+
+    var needle_r = from_slice("ll");
+    var needle: Str = new();
+    match needle_r { .Ok(v) => { needle = v; }, .Err(e) => { @assert(false, "needle"); } }
+    match find(hello, needle) {
+        .Found(index) => { @assert(index == 2, "find ll"); },
+        .NotFound => { @assert(false, "ll not found"); },
+    }
+    match find(hello, pre) {
+        .Found(index) => { @assert(index == 0, "find he"); },
+        .NotFound => { @assert(false, "he not found"); },
+    }
+
+    drop(&needle);
+    drop(&pre);
+    drop(&help);
+    drop(&hello);
+}
+
+fn test_clone_take_and_failures() {
+    var source = new();
+    expect_ok(append(&source, "owned"));
+
+    var copy = new();
+    expect_ok(clone(&copy, source));
+    expect_text(&copy, "owned");
+    expect_ok(append_byte(&source, '!')); // '!'
+    expect_text(&copy, "owned");         // 独立
+
+    var before = ByteBuffer.as_slice(copy.storage).ptr;
+    expect_ok(clone(&copy, copy));
+    @assert(ByteBuffer.as_slice(copy.storage).ptr != before, "self clone replaces allocation");
+    expect_text(&copy, "owned");
+
+    var moved = take(&copy);
+    @assert(ByteBuffer.size(copy.storage) == 0 && ByteBuffer.capacity(copy.storage) == 0,
+            "take empties source");
+    expect_text(&moved, "owned");
+
+    clear(&moved);
+    @assert(is_empty(moved) && ByteBuffer.capacity(moved.storage) != 0, "clear keeps capacity");
+
+    drop(&moved);
+    drop(&copy);
+    drop(&source);
+}
+
+fn main() {
+    test_construction();
+    test_compare_and_search();
+    test_clone_take_and_failures();
+}
diff --git a/stage2/utils/utf8.spl b/stage2/utils/utf8.spl
new file mode 100644
index 0000000..e4ba7ec
--- /dev/null
+++ b/stage2/utils/utf8.spl
@@ -0,0 +1,251 @@
+// Expect retcode = 0
+// UTF-8 工具：解码/编码、校验、码点数、以及 UTF-8 字节偏移 <-> UTF-16 code unit 偏移
+// （LSP 位置基于 UTF-16）。严格校验过长编码与代理区。
+
+type .{ .Error, .Status, .make } = @import("error.spl");
+type .{ .ByteBuffer } = @import("byte_buffer.spl");
+
+// 由首字节推断序列长度；非法/续字节返回 0。
+fn seq_len(b: u8): usize {
+    if b < 0x80 { ret 1; }
+    if b < 0xC0 { ret 0; } // 续字节
+    if b < 0xE0 { ret 2; }
+    if b < 0xF0 { ret 3; }
+    if b < 0xF8 { ret 4; }
+    ret 0;
+}
+
+// 从 index 解码一个码点到 out_cp，返回消耗字节数；非法返回 0。
+fn decode(data: *u8, len: usize, index: usize, out_cp: *u32): usize {
+    if index >= len { ret 0; }
+    var b0 = data[index];
+
+    if b0 < 0x80 {
+        out_cp.* = b0 as u32;
+        ret 1;
+    }
+    if b0 < 0xC2 { ret 0; } // 续字节或过长编码的前导
+
+    if b0 < 0xE0 {
+        if index + 1 >= len { ret 0; }
+        var b1 = data[index + 1];
+        if (b1 & 0xC0) != 0x80 { ret 0; }
+        out_cp.* = (((b0 & 0x1F) as u32) << 6) | ((b1 & 0x3F) as u32);
+        ret 2;
+    }
+    if b0 < 0xF0 {
+        if index + 2 >= len { ret 0; }
+        var b1 = data[index + 1];
+        var b2 = data[index + 2];
+        if (b1 & 0xC0) != 0x80 || (b2 & 0xC0) != 0x80 { ret 0; }
+        var cp = (((b0 & 0x0F) as u32) << 12) | (((b1 & 0x3F) as u32) << 6) | ((b2 & 0x3F) as u32);
+        if cp < 0x800 { ret 0; }                            // 过长
+        if cp >= 0xD800 && cp <= 0xDFFF { ret 0; }          // 代理区
+        out_cp.* = cp;
+        ret 3;
+    }
+    if b0 < 0xF5 {
+        if index + 3 >= len { ret 0; }
+        var b1 = data[index + 1];
+        var b2 = data[index + 2];
+        var b3 = data[index + 3];
+        if (b1 & 0xC0) != 0x80 || (b2 & 0xC0) != 0x80 || (b3 & 0xC0) != 0x80 { ret 0; }
+        var cp = (((b0 & 0x07) as u32) << 18) | (((b1 & 0x3F) as u32) << 12) |
+                 (((b2 & 0x3F) as u32) << 6) | ((b3 & 0x3F) as u32);
+        if cp < 0x10000 || cp > 0x10FFFF { ret 0; }
+        out_cp.* = cp;
+        ret 4;
+    }
+    ret 0;
+}
+
+fn valid(data: *u8, len: usize): bool {
+    var i: usize = 0;
+    var cp: u32 = 0;
+    while i < len {
+        var n = decode(data, len, i, &cp);
+        if n == 0 { ret false; }
+        i = i + n;
+    }
+    ret true;
+}
+
+fn encode_into(out: *ByteBuffer, cp: u32): Status {
+    if cp < 0x80 {
+        ret ByteBuffer.append_byte(out, cp as u8);
+    }
+    if cp < 0x800 {
+        var b0 = (0xC0 | ((cp >> 6) as u8));
+        var b1 = (0x80 | ((cp & 0x3F) as u8));
+        match ByteBuffer.append_byte(out, b0) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        ret ByteBuffer.append_byte(out, b1);
+    }
+    if cp >= 0xD800 && cp <= 0xDFFF { ret .Err(make(.InvalidArgument)); }
+    if cp < 0x10000 {
+        var b0 = (0xE0 | ((cp >> 12) as u8));
+        var b1 = (0x80 | (((cp >> 6) & 0x3F) as u8));
+        var b2 = (0x80 | ((cp & 0x3F) as u8));
+        match ByteBuffer.append_byte(out, b0) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match ByteBuffer.append_byte(out, b1) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        ret ByteBuffer.append_byte(out, b2);
+    }
+    if cp <= 0x10FFFF {
+        var b0 = (0xF0 | ((cp >> 18) as u8));
+        var b1 = (0x80 | (((cp >> 12) & 0x3F) as u8));
+        var b2 = (0x80 | (((cp >> 6) & 0x3F) as u8));
+        var b3 = (0x80 | ((cp & 0x3F) as u8));
+        match ByteBuffer.append_byte(out, b0) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match ByteBuffer.append_byte(out, b1) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match ByteBuffer.append_byte(out, b2) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        ret ByteBuffer.append_byte(out, b3);
+    }
+    ret .Err(make(.InvalidArgument));
+}
+
+fn count_codepoints(data: *u8, len: usize): usize {
+    var i: usize = 0;
+    var n: usize = 0;
+    var cp: u32 = 0;
+    while i < len {
+        var used = decode(data, len, i, &cp);
+        if used == 0 { ret n; }
+        i = i + used;
+        n = n + 1;
+    }
+    ret n;
+}
+
+// 码点占用的 UTF-16 code unit 数：BMP 为 1，补充平面为 2。
+fn utf16_units(data: *u8, len: usize): usize {
+    var i: usize = 0;
+    var n: usize = 0;
+    var cp: u32 = 0;
+    while i < len {
+        var used = decode(data, len, i, &cp);
+        if used == 0 { ret n; }
+        if cp >= 0x10000 { n = n + 2; } else { n = n + 1; }
+        i = i + used;
+    }
+    ret n;
+}
+
+// 字节偏移 -> UTF-16 偏移（夹取到 [0, utf16_units]）。
+fn byte_to_utf16(data: *u8, len: usize, byte_off: usize): usize {
+    var i: usize = 0;
+    var n: usize = 0;
+    var cp: u32 = 0;
+    while i < len && i < byte_off {
+        var used = decode(data, len, i, &cp);
+        if used == 0 { ret n; }
+        if cp >= 0x10000 { n = n + 2; } else { n = n + 1; }
+        i = i + used;
+    }
+    ret n;
+}
+
+// UTF-16 偏移 -> 字节偏移（夹取；落在码点内部时返回该码点起始字节）。
+fn utf16_to_byte(data: *u8, len: usize, utf16_off: usize): usize {
+    var i: usize = 0;
+    var n: usize = 0;
+    var cp: u32 = 0;
+    while i < len {
+        if n >= utf16_off { ret i; }
+        var used = decode(data, len, i, &cp);
+        if used == 0 { ret i; }
+        var units: usize = 1;
+        if cp >= 0x10000 { units = 2; }
+        if n + units > utf16_off { ret i; }
+        n = n + units;
+        i = i + used;
+    }
+    ret i;
+}
+
+// ---- 测试辅助 ----
+
+fn expect_ok(result: Status) {
+    match result {
+        .Ok => {},
+        .Err(error) => { @assert(false, "unexpected utf8 error"); },
+    }
+}
+
+fn test_decode_encode_roundtrip() {
+    var out = ByteBuffer.new();
+    expect_ok(encode_into(&out, 0x41));    // 'A'
+    expect_ok(encode_into(&out, 0xE9));    // é
+    expect_ok(encode_into(&out, 0x20AC));  // €
+    expect_ok(encode_into(&out, 0x1F600)); // 😀
+    var bytes = ByteBuffer.as_slice(out);
+    @assert(bytes.len == 1 + 2 + 3 + 4, "encoded length");
+    @assert(bytes.ptr[0] == 0x41, "A");
+    @assert(bytes.ptr[1] == 0xC3 && bytes.ptr[2] == 0xA9, "e-acute");
+    @assert(bytes.ptr[3] == 0xE2 && bytes.ptr[4] == 0x82 && bytes.ptr[5] == 0xAC, "euro");
+    @assert(bytes.ptr[6] == 0xF0 && bytes.ptr[9] == 0x80, "emoji head/tail");
+
+    @assert(valid(bytes.ptr, bytes.len), "valid");
+    @assert(count_codepoints(bytes.ptr, bytes.len) == 4, "codepoints");
+    @assert(utf16_units(bytes.ptr, bytes.len) == 1 + 1 + 1 + 2, "utf16 units");
+
+    var cp: u32 = 0;
+    @assert(decode(bytes.ptr, bytes.len, 0, &cp) == 1 && cp == 0x41, "dec A");
+    @assert(decode(bytes.ptr, bytes.len, 1, &cp) == 2 && cp == 0xE9, "dec e");
+    @assert(decode(bytes.ptr, bytes.len, 3, &cp) == 3 && cp == 0x20AC, "dec euro");
+    @assert(decode(bytes.ptr, bytes.len, 6, &cp) == 4 && cp == 0x1F600, "dec emoji");
+    ByteBuffer.drop(&out);
+}
+
+fn test_invalid() {
+    var bad1: [2]u8 = .[0xC0, 0x80];       // 过长
+    var bad2: [3]u8 = .[0xED, 0xA0, 0x80]; // 代理区
+    var bad3: [2]u8 = .[0xC3, 0x28];       // 坏续字节
+    var bad4: [1]u8 = .[0x80];             // 孤立续字节
+    var bad5: [3]u8 = .[0xE2, 0x82];       // 截断
+    var s1 = bad1[0..2];
+    var s2 = bad2[0..3];
+    var s3 = bad3[0..2];
+    var s4 = bad4[0..1];
+    var s5 = bad5[0..2];
+    @assert(!valid(s1.ptr, s1.len), "overlong");
+    @assert(!valid(s2.ptr, s2.len), "surrogate");
+    @assert(!valid(s3.ptr, s3.len), "bad cont");
+    @assert(!valid(s4.ptr, s4.len), "lone cont");
+    @assert(!valid(s5.ptr, s5.len), "truncated");
+
+    var out = ByteBuffer.new();
+    match encode_into(&out, 0xD800) {
+        .Ok => { @assert(false, "surrogate encode"); },
+        .Err(error) => {},
+    }
+    match encode_into(&out, 0x110000) {
+        .Ok => { @assert(false, "out of range encode"); },
+        .Err(error) => {},
+    }
+    ByteBuffer.drop(&out);
+}
+
+fn test_offsets() {
+    var out = ByteBuffer.new();
+    expect_ok(encode_into(&out, 0x41));    // byte 0, utf16 0
+    expect_ok(encode_into(&out, 0x1F600)); // bytes 1..4, utf16 1..2
+    expect_ok(encode_into(&out, 0x42));    // byte 5, utf16 3
+    var b = ByteBuffer.as_slice(out);
+
+    @assert(byte_to_utf16(b.ptr, b.len, 0) == 0, "b0");
+    @assert(byte_to_utf16(b.ptr, b.len, 1) == 1, "b1");
+    @assert(byte_to_utf16(b.ptr, b.len, 5) == 3, "b5");
+    @assert(byte_to_utf16(b.ptr, b.len, 6) == 4, "clamp");
+
+    @assert(utf16_to_byte(b.ptr, b.len, 0) == 0, "u0");
+    @assert(utf16_to_byte(b.ptr, b.len, 1) == 1, "u1");
+    @assert(utf16_to_byte(b.ptr, b.len, 2) == 1, "u2 stays in emoji");
+    @assert(utf16_to_byte(b.ptr, b.len, 3) == 5, "u3");
+    @assert(utf16_to_byte(b.ptr, b.len, 99) == 6, "clamp");
+    ByteBuffer.drop(&out);
+}
+
+fn main() {
+    test_decode_encode_roundtrip();
+    test_invalid();
+    test_offsets();
+}