stage2 迁移ast解析器 stage1 防止错误死循环

gramma.pest

@@ -214,7 +214,9 @@ TypePostfixExpr = {
 }
 
 TypePrimary = {
-    IDENTIFIER
+    AT ~ IDENTIFIER          // @self 等自引用/内置类型
+  | ComptimeCallExpr         // @import("...") 等导入即类型
+  | IDENTIFIER
   | KEYWORD_i8
   | KEYWORD_u8
   | KEYWORD_i16

nob.c

@@ -133,12 +133,33 @@ static bool collect_headers(void) {
     return true;
 }
 
+// stage2 的所有 .spl：作为 splc1.sir 的依赖（不解析 @import，保守全量重建）。
+static Nob_Files g_spl2 = {0};
+
+static bool collect_spl2_entry(Nob_Walk_Entry entry) {
+    if (entry.type != NOB_FILE_REGULAR)
+        return true;
+    const char *dot = strrchr(entry.path, '.');
+    if (!dot || strcmp(dot, ".spl") != 0)
+        return true;
+    da_append(&g_spl2, nob_temp_strdup(entry.path));
+    return true;
+}
+
+static bool collect_spl2(void) {
+    if (g_spl2.count > 0)
+        return true;
+    return nob_walk_dir("stage2", collect_spl2_entry);
+}
+
 static bool prepare_build(void) {
     if (!nob_mkdir_if_not_exists(BUILD_FOLDER))
         return false;
     if (!sync_cflags_stamp())
         return false;
-    return collect_headers();
+    if (!collect_headers())
+        return false;
+    return collect_spl2();
 }
 
 static bool build_object(const char *src) {
@@ -320,6 +341,105 @@ static bool test_once_with_dump(const char *fname, void *user_data) {
     return nob_cmd_run(&cmd);
 }
 
+// 用 splc0 编译 stage2/splc1.spl -> build/splc1.sir（依赖 stage2 全部 .spl）。
+static bool build_splc1(void) {
+    if (!build_target("splc0"))
+        return false;
+
+    const char *src = "stage2/splc1.spl";
+    const char *sir = BUILD_FOLDER "splc1.sir";
+
+    size_t ninputs = g_spl2.count + 2;
+    const char **inputs = (const char **)nob_temp_alloc(sizeof(const char *) * ninputs);
+    inputs[0] = src;
+    for (size_t i = 0; i < g_spl2.count; ++i)
+        inputs[i + 1] = g_spl2.items[i];
+    inputs[ninputs - 1] = CFLAGS_STAMP;
+
+    int needs = nob_needs_rebuild(sir, inputs, ninputs);
+    if (needs < 0)
+        return false;
+    if (needs == 0)
+        return true;
+
+    Nob_Cmd cmd = {0};
+    nob_cmd_append(&cmd, exe_path("splc0"), src, sir);
+    return nob_cmd_run(&cmd);
+}
+
+static bool run_capture(Nob_Cmd *cmd, Nob_Cmd_Result *result) {
+    Nob_String_Builder rendered = {0};
+    nob_cmd_render(*cmd, &rendered);
+    Nob_String_View sv = nob_sb_to_sv(rendered);
+    bool ok = nob_cmd_capture_popen(nob_temp_sv_to_cstr(sv), result);
+    nob_sb_free(rendered);
+    return ok;
+}
+
+// diff 测试：对同一文件跑 `splc0 --dump <mode>` 与 `spl_cli splc1.sir --dump <mode>`，
+// 逐字节比较 stdout。
+static bool test_once_diff(const char *fname, void *user_data) {
+    const char *mode = (const char *)user_data;
+    if (!mode) {
+        nob_log(NOB_ERROR, "diff test requires a mode: lex|ast|sema|ir|vm");
+        return false;
+    }
+    if (!build_target("splc0"))
+        return false;
+    if (!build_target("spl_cli"))
+        return false;
+    if (!build_splc1())
+        return false;
+
+    Nob_Cmd c_cmd = {0};
+    nob_cmd_append(&c_cmd, exe_path("splc0"), "--dump", (char *)mode, (char *)fname,
+                   BUILD_FOLDER "diff_c.sir");
+    Nob_Cmd_Result c_res = {0};
+    if (!run_capture(&c_cmd, &c_res)) {
+        nob_cmd_result_free(c_res);
+        return false;
+    }
+
+    Nob_Cmd s_cmd = {0};
+    nob_cmd_append(&s_cmd, exe_path("spl_cli"), BUILD_FOLDER "splc1.sir", "--dump", (char *)mode,
+                   (char *)fname);
+    Nob_Cmd_Result s_res = {0};
+    if (!run_capture(&s_cmd, &s_res)) {
+        nob_cmd_result_free(c_res);
+        nob_cmd_result_free(s_res);
+        return false;
+    }
+
+    Nob_String_View c_out = nob_sb_to_sv(c_res.out);
+    Nob_String_View s_out = nob_sb_to_sv(s_res.out);
+    bool same = nob_sv_eq(c_out, s_out);
+    if (!same) {
+        nob_log(NOB_ERROR, "diff mismatch: %s (c=%zu bytes, spl=%zu bytes)", fname, c_out.count,
+                s_out.count);
+        // 打印首个不相同行（按行比较）。
+        Nob_String_View ca = c_out;
+        Nob_String_View sa = s_out;
+        size_t lineno = 1;
+        while (ca.count > 0 || sa.count > 0) {
+            Nob_String_View cl = nob_sv_chop_by_delim(&ca, '\n');
+            Nob_String_View sl = nob_sv_chop_by_delim(&sa, '\n');
+            if (!nob_sv_eq(cl, sl)) {
+                nob_log(NOB_ERROR, "  line %zu:", lineno);
+                nob_log(NOB_ERROR, "    c  : %.*s", (int)cl.count, cl.data);
+                nob_log(NOB_ERROR, "    spl: %.*s", (int)sl.count, sl.data);
+                break;
+            }
+            lineno++;
+        }
+        nob_log(NOB_ERROR, "--- c stderr/stdout merged above; exit=%d vs spl exit=%d ---",
+                c_res.exit_code, s_res.exit_code);
+    }
+
+    nob_cmd_result_free(c_res);
+    nob_cmd_result_free(s_res);
+    return same;
+}
+
 static bool is_stage_name(const char *s) {
     return strcmp(s, "stage1") == 0 || strcmp(s, "stage2") == 0 || strcmp(s, "diff") == 0;
 }
@@ -481,6 +601,15 @@ int main(int argc, char **argv) {
             }
         }
 
+        // diff 需要显式 mode，且使用专门的双编译器比较函数。
+        if (strcmp(stage, "diff") == 0) {
+            if (!mode_name) {
+                nob_log(NOB_ERROR, "diff requires a mode: lex|ast|sema|ir|vm");
+                return 1;
+            }
+            tfn = test_once_diff;
+        }
+
         // 3. 剩下的都当 pattern
         Nob_Files files = {0};
         while (argc > 0) {
@@ -489,6 +618,7 @@ int main(int argc, char **argv) {
 
         return run_test(stage, &files, tfn, (void *)mode_name) ? 0 : 1;
     }
+
     if (strcmp(command_name, "clean") == 0) {
         if (nob_file_exists(BUILD_FOLDER)) {
             if (!delete_directory_recursively(BUILD_FOLDER))

stage0/spl_vm.c

@@ -724,9 +724,15 @@ int spl_vm_prepare(spl_vm_t *vm, const char *entry, int argc, const char **argv,
     if (!vm || !vm->prog) {
         return -1;
     }
-    spl_vm_func_t *fn = spl_prog_get_func(vm->prog, entry ? entry : "main");
+    spl_vm_func_t *fn = NULL;
+    const char *entry_point = "_start";
+    fn = spl_prog_get_func(vm->prog, entry_point);
+    if (fn == NULL) {
+        entry_point = entry ? entry : "main";
+        fn = spl_prog_get_func(vm->prog, entry_point);
+    }
     if (!fn) {
-        fprintf(stderr, "vm: entry point '%s' not found\n", entry ? entry : "main");
+        fprintf(stderr, "vm: entry point '%s' not found\n", entry_point);
         return -1;
     }
     vm->fp = 0;

stage1/spl_ast2ir.c

@@ -691,6 +691,69 @@ static spl_ir_node_ref_t emit_cast(spl_ast2ir_t *a, spl_ast_node_t *n) {
     return ast2ir_emit_cast_builtin(a, n, in, ft, tt, v);
 }
 
+/* 字段访问的基对象是 rvalue（如函数调用结果）时的回退路径：
+ * 先求值整个聚合值，再用 agg.extract 取出字段值（无需左值/临时槽）。 */
+static spl_ir_node_ref_t ast2ir_field_from_value(spl_ast2ir_t *a, spl_ast_node_t *n) {
+    spl_ast_node_t *obj = spl_ast_node(a->ast, n->field.obj);
+    if (!obj)
+        return 0;
+    spl_lang_type_id_t obj_lang = node_lang(a, obj);
+    spl_lang_type_node_t *obj_ln = spl_lang_type_node(a->sema->type, obj_lang);
+    if (!obj_ln)
+        return 0;
+
+    spl_type_id_t agg_ir = to_ir_type(a, obj_lang);
+    if (!agg_ir)
+        return 0;
+    usize idx = 0;
+    if (!spl_type_agg_member_idx(&a->sema->type->ir, agg_ir, n->field.field, &idx))
+        return 0;
+
+    usize field_idx = idx;
+    if (obj_ln->kind == SPL_LANG_TYPE_ENUM)
+        field_idx = idx + 1; /* enum: field 0 是 tag */
+
+    spl_type_id_t field_ir = 0;
+    spl_type_node_t *agg_tn = spl_type_node(&a->sema->type->ir, agg_ir);
+    if (agg_tn && field_idx < vec_size(agg_tn->agg_type.variants))
+        field_ir = vec_at(agg_tn->agg_type.variants, field_idx).type_id;
+
+    spl_ir_node_ref_t val = emit_expr(a, n->field.obj);
+    if (!val)
+        return 0;
+    return spl_ir_builder_agg_extract(a->ir, agg_ir, field_ir, field_idx, val);
+}
+
+/* 限定枚举变体 Type.Variant：产出枚举常量（tag=变体下标；无载荷变体不带 payload）。 */
+static spl_ir_node_ref_t ast2ir_enum_variant(spl_ast2ir_t *a, spl_ast_node_t *n) {
+    spl_type_id_t t = node_ir(a, n);
+    if (!t)
+        return 0;
+    spl_type_node_t *tn = spl_type_node(&a->sema->type->ir, t);
+    if (!tn)
+        return 0;
+
+    usize vidx = 0;
+    if (!spl_type_agg_member_idx(&a->sema->type->ir, t, n->field.field, &vidx))
+        return 0;
+
+    spl_type_id_t tag_ir = tn->agg_type.tag_type;
+    spl_ir_node_ref_t eptr = spl_ir_builder_mem_alloca(a->ir, t, const_usize(a, 1));
+    spl_ir_node_ref_t tptr = spl_ir_builder_mem_field(a->ir, t, eptr, SPL_ENUM_TAG_FIELD_IDX);
+    spl_ir_node_ref_t tag_val = spl_ir_builder_type_const_int(a->ir, tag_ir, vidx);
+    spl_ir_builder_mem_store(a->ir, tag_ir, tptr, tag_val);
+
+    if (vidx < vec_size(tn->agg_type.variants)) {
+        spl_type_id_t payload_ir = vec_at(tn->agg_type.variants, vidx).type_id;
+        if (payload_ir) {
+            spl_ir_node_ref_t pptr = spl_ir_builder_mem_field(a->ir, t, eptr, vidx + 1);
+            spl_ir_node_ref_t zero = emit_zero_value(a, payload_ir);
+            spl_ir_builder_mem_store(a->ir, payload_ir, pptr, zero);
+        }
+    }
+    return eptr;
+}
+
 static spl_ir_node_ref_t emit_expr(spl_ast2ir_t *a, spl_ast_node_ref_t ref) {
     spl_ir_node_ref_t val = emit_expr_raw(a, ref);
     if (!val)
@@ -1022,10 +1085,20 @@ static spl_ir_node_ref_t emit_expr_raw(spl_ast2ir_t *a, spl_ast_node_ref_t ref)
             }
         }
 
+        /* 限定枚举变体 Type.Variant：基对象是聚合类型名且为 enum。 */
+        spl_ast_node_t *objn = spl_ast_node(a->ast, n->field.obj);
+        if (objn && objn->kind == SPL_AST_EXPR_IDENT) {
+            spl_def_node_t *od = spl_def_node(a->sema->type, objn->resolved_def_id);
+            spl_lang_type_id_t et = node_lang(a, objn);
+            spl_lang_type_node_t *etn = spl_lang_type_node(a->sema->type, et);
+            if (od && od->kind == SPL_DEF_AGG && etn && etn->kind == SPL_LANG_TYPE_ENUM)
+                return ast2ir_enum_variant(a, n);
+        }
+
         spl_ir_node_ref_t ptr = emit_lvalue(a, ref);
-        if (!ptr)
-            return 0;
+        if (ptr)
             return spl_ir_builder_mem_load(a->ir, node_ir(a, n), ptr);
+        return ast2ir_field_from_value(a, n);
     }
     case SPL_AST_EXPR_INDEX: {
         spl_ir_node_ref_t ptr = emit_lvalue(a, ref);
@@ -1065,6 +1138,46 @@ static spl_ir_node_ref_t emit_expr_raw(spl_ast2ir_t *a, spl_ast_node_ref_t ref)
 
         usize nelems = vec_size(n->array_literal.elems);
 
+        /* [N]T -> []T：物化数组到局部存储，再构造 {ptr, len} 切片。 */
+        spl_lang_type_node_t *sln = spl_lang_type_node(a->sema->type, node_lang(a, n));
+        if (sln && sln->kind == SPL_LANG_TYPE_SLICE) {
+            spl_type_id_t elem_ir = to_ir_type(a, sln->extra.elem);
+            if (!elem_ir)
+                Panic("array literal slice element has no ir type");
+            spl_lang_type_id_t usize_lang = spl_lang_make_int(a->sema->type, sizeof(void *) * 8, 0);
+            spl_type_id_t usize_ir = to_ir_type(a, usize_lang);
+            spl_type_id_t ptr_ir = spl_type_builder_ptr(&a->sema->type->ir, elem_ir);
+
+            spl_ir_node_ref_t base = 0;
+            if (nelems == 0) {
+                base = spl_ir_builder_type_const_ptr(a->ir, ptr_ir, NULL);
+            } else {
+                spl_type_id_t arr_ir =
+                    spl_type_builder_array(&a->sema->type->ir, elem_ir, nelems);
+                spl_ir_node_ref_t slot =
+                    spl_ir_builder_mem_alloca(a->ir, arr_ir, const_usize(a, 1));
+                for (usize i = 0; i < nelems; ++i) {
+                    spl_ir_node_ref_t fptr =
+                        spl_ir_builder_mem_field(a->ir, arr_ir, slot, i);
+                    spl_ir_node_ref_t v = emit_expr(a, vec_at(n->array_literal.elems, i));
+                    if (!v)
+                        Panic("array element emit failed");
+                    spl_ir_builder_mem_store(a->ir, elem_ir, fptr, v);
+                }
+                base = spl_ir_builder_mem_field(a->ir, arr_ir, slot, 0);
+            }
+            if (!base)
+                Panic("array literal slice base failed");
+
+            spl_ir_node_ref_t len = spl_ir_builder_type_const_int(a->ir, usize_ir, nelems);
+            spl_ir_node_ref_t node = spl_ir_builder_agg_construct(a->ir, t);
+            if (!node)
+                Panic("agg_construct failed for array-to-slice");
+            spl_ir_builder_agg_construct_field(a->ir, node, base);
+            spl_ir_builder_agg_construct_field(a->ir, node, len);
+            return node;
+        }
+
         spl_ir_node_ref_vec_t field_refs;
         vec_init(field_refs);
         for (usize i = 0; i < nelems; ++i) {

stage1/spl_lexer.c

@@ -411,7 +411,7 @@ spl_tok_vec_t spl_lex(const char *source, const char *fname) {
     }
 
         /* Three-char operators first */
-        TRY_OP3('<', '-', '>', TOK_ARROW_BOTH, TOK_RIGHT_ARROW, TOK_LT)
+        TRY_OP3('<', '-', '>', TOK_ARROW_BOTH, TOK_LEFT_ARROW, TOK_LT)
         TRY_OP3('<', '=', '>', TOK_FAT_ARROW_BOTH, TOK_LE, TOK_LT)
         TRY_OP3('<', '<', '=', TOK_ASSIGN_L_SH, TOK_L_SH, TOK_LT)
         TRY_OP3('>', '>', '=', TOK_ASSIGN_R_SH, TOK_R_SH, TOK_GT)
@@ -433,7 +433,6 @@ spl_tok_vec_t spl_lex(const char *source, const char *fname) {
         TRY_OP2('&', '=', TOK_ASSIGN_AND, TOK_AND)
         TRY_OP2('|', '=', TOK_ASSIGN_OR, TOK_OR)
         TRY_OP2('^', '=', TOK_ASSIGN_XOR, TOK_XOR)
-        TRY_OP2('<', '-', TOK_LEFT_ARROW, TOK_LT)
         TRY_OP2('-', '>', TOK_RIGHT_ARROW, TOK_SUB)
         TRY_OP2(':', '=', TOK_COLON_ASSIGN, TOK_COLON)
 

stage1/spl_parser.c

@@ -849,6 +849,7 @@ static spl_ast_node_ref_t parse_container_declaration(spl_parser_t *p) {
     }
     default:
         PARSE_ERROR(p, SPL_DIAG_KIND_PARSE_UNRECOGNIZED, t, "unrecognized declaration");
+        advance(p); /* 消费该 token，避免调用方在错误恢复时原地死循环 */
         vec_free(tag_list);
         return 0;
     }

stage1/spl_sema.c

@@ -1520,6 +1520,15 @@ static spl_lang_type_id_t sema_coerce(spl_sema_t *sema, spl_ast_node_ref_t ref,
             n->resolved_lang_tid = expected;
             return expected;
         }
+        /* [N]T -> []T：数组字面量在需要切片处隐式物化为切片。 */
+        if (en && en->kind == SPL_LANG_TYPE_SLICE) {
+            spl_lang_type_id_t et = en->extra.elem;
+            for (usize i = 0; i < vec_size(n->array_literal.elems); ++i)
+                if (!sema_coerce(sema, vec_at(n->array_literal.elems, i), et))
+                    return 0;
+            n->resolved_lang_tid = expected;
+            return expected;
+        }
         break;
     }
 
@@ -2024,6 +2033,20 @@ static spl_lang_type_id_t sema_infer(spl_sema_t *sema, spl_ast_node_ref_t ref) {
         if (!obj)
             return 0;
 
+        /* 限定枚举变体 Type.Variant：基对象是聚合类型名且类型为 enum。 */
+        spl_ast_node_t *objn = spl_ast_node(sema->ast, n->field.obj);
+        if (objn && objn->kind == SPL_AST_EXPR_IDENT) {
+            spl_def_node_t *od = spl_def_node(sema->type, objn->resolved_def_id);
+            if (od && od->kind == SPL_DEF_AGG && obj->kind == SPL_LANG_TYPE_ENUM) {
+                spl_type_id_t eir = spl_lang_type_get_ir_type(sema->type, obj_tid);
+                usize vidx = 0;
+                if (eir && spl_type_agg_member_idx(&sema->type->ir, eir, n->field.field, &vidx)) {
+                    n->resolved_lang_tid = obj_tid;
+                    return obj_tid;
+                }
+            }
+        }
+
         spl_intrin_t in = spl_intrin_lookup_field(obj->kind, n->field.field);
         if (in != SPL_INTRIN_NONE) {
             spl_lang_type_id_t t = sema_intrin_field(sema, n, in);

stage1/splc0.c

@@ -81,6 +81,39 @@ static void dump_tokens(spl_tok_vec_t toks) {
     vec_free(toks);
 }
 
+/* 折叠重复分隔符并去掉 "." 路径分量，使同一模块的 import 解析结果唯一。
+   就地重写，'\' 统一成 '/'；保留前导分隔符（绝对路径）。 */
+static void normalize_import_path(char *p) {
+    usize n = strlen(p);
+    char *out = malloc(n + 1);
+    if (!out)
+        return;
+    char *w = out;
+    if (p[0] == '/' || p[0] == '\\')
+        *w++ = '/';
+
+    char *r = p;
+    while (*r) {
+        while (*r == '/' || *r == '\\')
+            r++;
+        if (!*r)
+            break;
+        char *seg = r;
+        while (*r && *r != '/' && *r != '\\')
+            r++;
+        usize len = (usize)(r - seg);
+        if (len == 1 && seg[0] == '.')
+            continue;
+        if (w != out && w[-1] != '/')
+            *w++ = '/';
+        memcpy(w, seg, len);
+        w += len;
+    }
+    *w = 0;
+    strcpy(p, out);
+    free(out);
+}
+
 /* 把 import_path 拼接到 importing_file 所在目录。
    from_fname 为空时按相对 cwd 处理。绝对路径直接透传。 */
 static void resolve_relative(const char *from, const char *path, char *out, usize cap) {
@@ -91,6 +124,7 @@ static void resolve_relative(const char *from, const char *path, char *out, usiz
     }
     if (path[0] == '/' || path[0] == '\\' || (path[0] && path[1] == ':')) {
         snprintf(out, cap, "%s", path);
+        normalize_import_path(out);
         return;
     }
     const char *last_sep = NULL;
@@ -99,10 +133,12 @@ static void resolve_relative(const char *from, const char *path, char *out, usiz
             last_sep = p;
     if (!last_sep) {
         snprintf(out, cap, "%s", path);
+        normalize_import_path(out);
         return;
     }
     usize dir_len = (usize)(last_sep - from) + 1;
     snprintf(out, cap, "%.*s%s", (int)dir_len, from, path);
+    normalize_import_path(out);
 }
 
 static spl_ast_t *driver_load(void *user, const char *from_fname, const char *path,

stage2/ast.spl

@@ -0,0 +1,642 @@
+// Expect retcode = 0
+
+// AST：用 ADT（enum 带载荷）表达节点，节点引用是 arena 序号 usize。
+// stage1 的 C 结构只是「手写 union 模拟 ADT」；这里直接用 SPL 的 enum 载荷，
+// 不照抄 union 布局，只要求最终 dump / 语义与 stage1 一致。
+// 索引 0 为 NONE 哨兵（与 stage1 一致）。
+
+type .{ .Error } = @import("utils/error.spl");
+type .{ .Vec } = @import("utils/vec.spl");
+type .{ .ByteView } = @import("utils/byte_view.spl");
+type .{ .ByteBuffer } = @import("utils/byte_buffer.spl");
+type .{ .append_usize } = @import("utils/fmt.spl");
+type .{ .Punct } = @import("token.spl");
+type BigInt = @import("utils/mp.spl");
+type .{ .Span, .span_none, .span_make, .span_is_none } = @import("./diag.spl");
+type .{ .Sources, .Source, .set_at } = @import("utils/source.spl");
+
+// 算子不另立 enum：AST 直接存词法 Punct。一元/二元由节点 kind 区分
+// （一元：Sub=负 / Not=! / BitNot=~ / And=& / Mul=.*；二元即对应 Punct）。
+
+// ---- 节点体（ADT；variant 名 = stage1 的 kind 名；载荷用内联匿名 struct）----
+
+type Body = enum {
+    None,
+    Root: struct { members: usize },
+    ContainerMembers: struct { external_tags: Vec, declarations: Vec },
+    ExternalTagItem: struct { is_at: bool, is_external: bool, name: ByteView, args: Vec },
+    TagItem: struct { is_at: bool, is_external: bool, name: ByteView, args: Vec },
+    ComptimeDecl: struct { body: usize },
+    TopLevelBlock: struct { tag_list: Vec, block: usize },
+    FnDecl: struct {
+        tag_list: Vec,
+        is_comptime: bool,
+        name: ByteView,
+        params: Vec,
+        ret_type: usize,
+        body: usize,
+    },
+    VarDecl: struct { attr_list: Vec, pattern: usize, type_expr: usize, init_expr: usize },
+    ConstDecl: struct { attr_list: Vec, pattern: usize, type_expr: usize, init_expr: usize },
+    TypeDecl: struct { attr_list: Vec, pattern: usize, type_expr: usize },
+    MemberDecl: struct {
+        attr_list: Vec,
+        name: ByteView,
+        type_expr: usize,
+        init_expr: usize,
+    },
+    ParamDecl: struct { attr_list: Vec, name: ByteView, type_expr: usize },
+    StatementRet: struct { expr: usize },
+    StatementBreak: struct { label: usize, expr: usize },
+    StatementContinue: struct { label: usize },
+    StatementDefer: struct { body: usize },
+    StatementComptime: struct { body: usize },
+    StatementDecl: struct { decl: usize },
+    StatementExpr: struct { expr: usize },
+    AggregateMember: struct {
+        attr_list: Vec,
+        name: ByteView,
+        type_expr: usize,
+        init_expr: usize,
+    },
+    StructInitItem: struct { name: ByteView, expr: usize },
+    MatchArm: struct { pattern: usize, body: usize },
+    Label: struct { name: ByteView },
+    ExprIntLit: struct { value: BigInt }, // 任意宽度（mp），解析后由 Ast 拥有
+    ExprFloatLit: struct { text: ByteView },
+    ExprCharLit: struct { value: u8 },
+    ExprStringLit: struct { data: ByteView },
+    ExprBoolLit: struct { value: bool },
+    ExprNullLit,
+    ExprUndefinedLit,
+    ExprIdent: struct { name: ByteView },
+    ExprUnary: struct { op: Punct, operand: usize },
+    ExprBinary: struct { op: Punct, left: usize, right: usize },
+    ExprCall: struct { callee: usize, args: Vec },
+    ExprField: struct { obj: usize, name: ByteView },
+    ExprIndex: struct { obj: usize, index: usize },
+    ExprSlice: struct { obj: usize, begin: usize, end: usize },
+    ExprCast: struct { expr: usize, type_expr: usize },
+    ExprTuple: struct { items: Vec },
+    ExprArrayLit: struct {
+        has_len: bool,
+        len: usize,
+        is_underscore_len: bool,
+        elem_type: usize,
+        elems: Vec,
+    },
+    ExprAggregateLit: struct { name: ByteView, fields: Vec },
+    ExprComptimeCall: struct { name: ByteView, args: Vec },
+    ExprBlock: struct { stmts: Vec, tail_expr: usize },
+    ExprIf: struct { cond: usize, then_block: usize, else_branch: usize },
+    ExprIfvar: struct { var_decl: usize, then_block: usize, else_branch: usize },
+    ExprWhile: struct { label: usize, cond: usize, body: usize, else_branch: usize },
+    ExprLoop: struct { label: usize, body: usize },
+    ExprFor: struct {
+        label: usize,
+        iter_exprs: Vec,
+        pattern: usize,
+        body: usize,
+        else_branch: usize,
+    },
+    ExprMatch: struct { matched_expr: usize, arms: Vec },
+    ExprFnType: struct { is_comptime: bool, params: Vec, ret_type: usize },
+    TypePath: struct { name: ByteView, generic_args: Vec },
+    TypePointer: struct { pointee: usize },
+    TypeArray: struct {
+        has_len: bool,
+        len: isize,
+        is_underscore_len: bool,
+        elem_type: usize,
+    },
+    TypeSlice: struct { elem_type: usize },
+    TypeFn: struct { is_comptime: bool, params: Vec, ret_type: usize },
+    TypeParen: struct { types: Vec },
+    TypeAggregate: struct { kind_name: ByteView, members: Vec },
+    TypeComptimeCall: struct { name: ByteView, args: Vec },
+    PatternIdent: struct { name: ByteView },
+    PatternUnderscore,
+    PatternDot: struct { name: ByteView, sub_pattern: usize },
+    PatternTuple: struct { items: Vec },
+    PatternList: struct { items: Vec },
+    PatternAggregate: struct { items: Vec },
+    PatternField: struct { name: ByteView, pattern: usize },
+}
+
+fn kind_name(body: Body): []u8 {
+    match body {
+        .None => { ret "none"; },
+        .Root(_) => { ret "root"; },
+        .ContainerMembers(_) => { ret "container_members"; },
+        .ExternalTagItem(_) => { ret "external_tag_item"; },
+        .TagItem(_) => { ret "tag_item"; },
+        .ComptimeDecl(_) => { ret "comptime_decl"; },
+        .TopLevelBlock(_) => { ret "top_level_block"; },
+        .FnDecl(_) => { ret "fn_decl"; },
+        .VarDecl(_) => { ret "var_decl"; },
+        .ConstDecl(_) => { ret "const_decl"; },
+        .TypeDecl(_) => { ret "type_decl"; },
+        .MemberDecl(_) => { ret "member_decl"; },
+        .ParamDecl(_) => { ret "param_decl"; },
+        .StatementRet(_) => { ret "statement_ret"; },
+        .StatementBreak(_) => { ret "statement_break"; },
+        .StatementContinue(_) => { ret "statement_continue"; },
+        .StatementDefer(_) => { ret "statement_defer"; },
+        .StatementComptime(_) => { ret "statement_comptime"; },
+        .StatementDecl(_) => { ret "statement_decl"; },
+        .StatementExpr(_) => { ret "statement_expr"; },
+        .AggregateMember(_) => { ret "aggregate_member"; },
+        .StructInitItem(_) => { ret "struct_init_item"; },
+        .MatchArm(_) => { ret "match_arm"; },
+        .Label(_) => { ret "label"; },
+        .ExprIntLit(_) => { ret "int_lit"; },
+        .ExprFloatLit(_) => { ret "float_lit"; },
+        .ExprCharLit(_) => { ret "char_lit"; },
+        .ExprStringLit(_) => { ret "string_lit"; },
+        .ExprBoolLit(_) => { ret "bool_lit"; },
+        .ExprNullLit => { ret "null_lit"; },
+        .ExprUndefinedLit => { ret "undefined_lit"; },
+        .ExprIdent(_) => { ret "ident"; },
+        .ExprUnary(_) => { ret "unary"; },
+        .ExprBinary(_) => { ret "binary"; },
+        .ExprCall(_) => { ret "call"; },
+        .ExprField(_) => { ret "field"; },
+        .ExprIndex(_) => { ret "index"; },
+        .ExprSlice(_) => { ret "slice"; },
+        .ExprCast(_) => { ret "cast"; },
+        .ExprTuple(_) => { ret "tuple"; },
+        .ExprArrayLit(_) => { ret "array_literal"; },
+        .ExprAggregateLit(_) => { ret "aggregate_literal"; },
+        .ExprComptimeCall(_) => { ret "comptime_call"; },
+        .ExprBlock(_) => { ret "block"; },
+        .ExprIf(_) => { ret "if_expr"; },
+        .ExprIfvar(_) => { ret "ifvar_expr"; },
+        .ExprWhile(_) => { ret "while_expr"; },
+        .ExprLoop(_) => { ret "loop_expr"; },
+        .ExprFor(_) => { ret "for_expr"; },
+        .ExprMatch(_) => { ret "match_expr"; },
+        .ExprFnType(_) => { ret "fn_type"; },
+        .TypePath(_) => { ret "type_path"; },
+        .TypePointer(_) => { ret "type_pointer"; },
+        .TypeArray(_) => { ret "type_array"; },
+        .TypeSlice(_) => { ret "type_slice"; },
+        .TypeFn(_) => { ret "type_fn"; },
+        .TypeParen(_) => { ret "type_paren"; },
+        .TypeAggregate(_) => { ret "type_aggregate"; },
+        .TypeComptimeCall(_) => { ret "type_comptime_call"; },
+        .PatternIdent(_) => { ret "pattern_ident"; },
+        .PatternUnderscore => { ret "pattern_underscore"; },
+        .PatternDot(_) => { ret "pattern_dot"; },
+        .PatternTuple(_) => { ret "pattern_tuple"; },
+        .PatternList(_) => { ret "pattern_list"; },
+        .PatternAggregate(_) => { ret "pattern_aggregate"; },
+        .PatternField(_) => { ret "pattern_field"; },
+    }
+    ret "?";
+}
+
+type Node = struct {
+    span: Span,
+    resolved_def_id: usize,
+    resolved_lang_tid: usize,
+    expected_lang_tid: usize,
+    attrs: Vec,
+    body: Body,
+}
+
+type Ast = @self;
+
+nodes: Vec, // Node
+root: usize,
+sources: *Sources, // 借用；可为 null（dump 时不解析行列）
+
+type AstResult = enum {
+    Ok: Ast,
+    Err: Error,
+}
+
+fn new_node(body: Body, span: Span): Node {
+    ret Node {
+        .span = span,
+        .resolved_def_id = 0,
+        .resolved_lang_tid = 0,
+        .expected_lang_tid = 0,
+        .attrs = Vec.blank(@sizeof(usize), @alignof(usize)),
+        .body = body,
+    };
+}
+
+fn new(sources: *Sources): AstResult {
+    var ast = Ast {
+        .nodes = Vec.blank(@sizeof(Node), @alignof(Node)),
+        .root = 0,
+        .sources = sources,
+    };
+    var sentinel = new_node(.None, span_none());
+    match Vec.push(&ast.nodes, &sentinel as *u8) {
+        .Ok => {},
+        .Err(error) => {
+            Vec.drop(&ast.nodes);
+            ret .Err(error);
+        },
+    }
+    ret .Ok(ast);
+}
+
+fn drop(self: *Ast) {
+    var i: usize = 0;
+    while i < self.nodes.len {
+        var n = node(self, i);
+        if n != null {
+            Vec.drop(&n.attrs);
+            match n.body {
+                .ExprIntLit(x) => { BigInt.drop(&x.value); },
+                _ => {},
+            }
+        }
+        i = i + 1;
+    }
+    Vec.drop(&self.nodes);
+    self.root = 0;
+}
+
+fn node(self: *Ast, ref: usize): *Node {
+    match Vec.get(&self.nodes, ref) {
+        .Ok(ptr) => { ret ptr as *Node; },
+        .Err(error) => { ret null; },
+    }
+    ret null;
+}
+
+// 追加节点，返回其引用（失败返回 0）。
+fn alloc(self: *Ast, body: Body, span: Span): usize {
+    var n = new_node(body, span);
+    match Vec.push(&self.nodes, &n as *u8) {
+        .Ok => {},
+        .Err(error) => {
+            Vec.drop(&n.attrs);
+            ret 0;
+        },
+    }
+    ret self.nodes.len - 1;
+}
+
+fn count(self: *Ast): usize {
+    ret self.nodes.len;
+}
+
+fn ref_vec(): Vec {
+    ret Vec.blank(@sizeof(usize), @alignof(usize));
+}
+
+// ---- dump（镜像 stage1 spl_ast_dump）----
+//
+// AST:
+// <indent>`-/|-kind #ref [line:col]
+// indent：栈上每层 last 为真输出 "  "，否则 "| "。
+
+fn push(buf: *ByteBuffer, bytes: []u8): bool {
+    match ByteBuffer.append(buf, bytes) {
+        .Ok => { ret true; },
+        .Err(error) => { ret false; },
+    }
+    ret false;
+}
+
+fn push_byte(buf: *ByteBuffer, byte: u8): bool {
+    match ByteBuffer.append_byte(buf, byte) {
+        .Ok => { ret true; },
+        .Err(error) => { ret false; },
+    }
+    ret false;
+}
+
+fn push_usize(buf: *ByteBuffer, value: usize): bool {
+    match append_usize(buf, value, 10, false) {
+        .Ok => { ret true; },
+        .Err(error) => { ret false; },
+    }
+    ret false;
+}
+
+fn ref_at(list: Vec, index: usize): usize {
+    match Vec.get(&list, index) {
+        .Ok(ptr) => { ret (ptr as *usize).*; },
+        .Err(error) => { ret 0; },
+    }
+    ret 0;
+}
+
+fn dump_opt(ast: *Ast, buf: *ByteBuffer, stack: *Vec, ref: usize, last: bool): bool {
+    if ref == 0 { ret true; }
+    ret dump_node(ast, buf, stack, ref, last);
+}
+
+fn dump_list(ast: *Ast, buf: *ByteBuffer, stack: *Vec, list: Vec): bool {
+    var i: usize = 0;
+    while i < list.len {
+        var last = (i + 1 == list.len);
+        if !dump_node(ast, buf, stack, ref_at(list, i), last) { ret false; }
+        i = i + 1;
+    }
+    ret true;
+}
+
+fn dump_node(ast: *Ast, buf: *ByteBuffer, stack: *Vec, ref: usize, last: bool): bool {
+    var n = node(ast, ref);
+    if n == null { ret false; }
+
+    var i: usize = 0;
+    while i < stack.len {
+        var flag: u8 = 0;
+        match Vec.get(stack, i) {
+            .Ok(ptr) => { flag = ptr.*; },
+            .Err(error) => { ret false; },
+        }
+        if !push(buf, if flag != 0 { "  " } else { "| " }) { ret false; }
+        i = i + 1;
+    }
+    if !push(buf, if last { "`-" } else { "|-" }) { ret false; }
+    if !push(buf, kind_name(n.body)) { ret false; }
+    if !push(buf, " #") { ret false; }
+    if !push_usize(buf, ref) { ret false; }
+    if !span_is_none(n.span) && ast.sources != null {
+        var src = set_at(ast.sources, n.span.file);
+        if src != null {
+            var pos = Source.line_col(src.*, n.span.start);
+            if !push(buf, " [") { ret false; }
+            if !push_usize(buf, pos.line) { ret false; }
+            if !push_byte(buf, ':') { ret false; }
+            if !push_usize(buf, pos.col) { ret false; }
+            if !push_byte(buf, ']') { ret false; }
+        }
+    }
+    if !push_byte(buf, '\n') { ret false; }
+
+    var l: u8 = 0;
+    if last { l = 1; }
+    match Vec.push(stack, &l as *u8) {
+        .Ok => {},
+        .Err(error) => { ret false; },
+    }
+
+    var ok = true;
+    match n.body {
+        .None => {},
+        .Root(r) => { ok = dump_opt(ast, buf, stack, r.members, true); },
+        .ContainerMembers(c) => {
+            ok = dump_list(ast, buf, stack, c.external_tags) &&
+                 dump_list(ast, buf, stack, c.declarations);
+        },
+        .ExternalTagItem(t) => { ok = dump_list(ast, buf, stack, t.args); },
+        .TagItem(t) => { ok = dump_list(ast, buf, stack, t.args); },
+        .ComptimeDecl(x) => { ok = dump_opt(ast, buf, stack, x.body, true); },
+        .TopLevelBlock(t) => {
+            ok = dump_list(ast, buf, stack, t.tag_list) &&
+                 dump_opt(ast, buf, stack, t.block, true);
+        },
+        .FnDecl(f) => {
+            ok = dump_list(ast, buf, stack, f.tag_list) &&
+                 dump_list(ast, buf, stack, f.params) &&
+                 dump_opt(ast, buf, stack, f.ret_type, false) &&
+                 dump_opt(ast, buf, stack, f.body, true);
+        },
+        .VarDecl(d) => {
+            ok = dump_list(ast, buf, stack, d.attr_list) &&
+                 dump_opt(ast, buf, stack, d.pattern, false) &&
+                 dump_opt(ast, buf, stack, d.type_expr, false) &&
+                 dump_opt(ast, buf, stack, d.init_expr, true);
+        },
+        .ConstDecl(d) => {
+            ok = dump_list(ast, buf, stack, d.attr_list) &&
+                 dump_opt(ast, buf, stack, d.pattern, false) &&
+                 dump_opt(ast, buf, stack, d.type_expr, false) &&
+                 dump_opt(ast, buf, stack, d.init_expr, true);
+        },
+        .TypeDecl(d) => {
+            ok = dump_list(ast, buf, stack, d.attr_list) &&
+                 dump_opt(ast, buf, stack, d.pattern, false) &&
+                 dump_opt(ast, buf, stack, d.type_expr, true);
+        },
+        .MemberDecl(d) => {
+            ok = dump_list(ast, buf, stack, d.attr_list) &&
+                 dump_opt(ast, buf, stack, d.type_expr, false) &&
+                 dump_opt(ast, buf, stack, d.init_expr, true);
+        },
+        .ParamDecl(p) => {
+            ok = dump_list(ast, buf, stack, p.attr_list) &&
+                 dump_opt(ast, buf, stack, p.type_expr, true);
+        },
+        .StatementRet(e) => { ok = dump_opt(ast, buf, stack, e.expr, true); },
+        .StatementBreak(b) => {
+            ok = dump_opt(ast, buf, stack, b.label, false) &&
+                 dump_opt(ast, buf, stack, b.expr, true);
+        },
+        .StatementContinue(x) => { ok = dump_opt(ast, buf, stack, x.label, true); },
+        .StatementDefer(x) => { ok = dump_opt(ast, buf, stack, x.body, true); },
+        .StatementComptime(x) => { ok = dump_opt(ast, buf, stack, x.body, true); },
+        .StatementDecl(d) => { ok = dump_opt(ast, buf, stack, d.decl, true); },
+        .StatementExpr(e) => { ok = dump_opt(ast, buf, stack, e.expr, true); },
+        .AggregateMember(d) => {
+            ok = dump_list(ast, buf, stack, d.attr_list) &&
+                 dump_opt(ast, buf, stack, d.type_expr, false) &&
+                 dump_opt(ast, buf, stack, d.init_expr, true);
+        },
+        .StructInitItem(s) => { ok = dump_opt(ast, buf, stack, s.expr, true); },
+        .MatchArm(m) => {
+            ok = dump_opt(ast, buf, stack, m.pattern, false) &&
+                 dump_opt(ast, buf, stack, m.body, true);
+        },
+        .Label(x) => {},
+        .ExprIntLit(x) => {},
+        .ExprFloatLit(x) => {},
+        .ExprCharLit(x) => {},
+        .ExprStringLit(x) => {},
+        .ExprBoolLit(x) => {},
+        .ExprNullLit => {},
+        .ExprUndefinedLit => {},
+        .ExprIdent(x) => {},
+        .ExprUnary(u) => { ok = dump_opt(ast, buf, stack, u.operand, true); },
+        .ExprBinary(b) => {
+            ok = dump_opt(ast, buf, stack, b.left, false) &&
+                 dump_opt(ast, buf, stack, b.right, true);
+        },
+        .ExprCall(c) => {
+            ok = dump_opt(ast, buf, stack, c.callee, false) &&
+                 dump_list(ast, buf, stack, c.args);
+        },
+        .ExprField(f) => { ok = dump_opt(ast, buf, stack, f.obj, true); },
+        .ExprIndex(x) => {
+            ok = dump_opt(ast, buf, stack, x.obj, false) &&
+                 dump_opt(ast, buf, stack, x.index, true);
+        },
+        .ExprSlice(s) => {
+            ok = dump_opt(ast, buf, stack, s.obj, false) &&
+                 dump_opt(ast, buf, stack, s.begin, false) &&
+                 dump_opt(ast, buf, stack, s.end, true);
+        },
+        .ExprCast(c) => {
+            ok = dump_opt(ast, buf, stack, c.expr, false) &&
+                 dump_opt(ast, buf, stack, c.type_expr, true);
+        },
+        .ExprTuple(v) => { ok = dump_list(ast, buf, stack, v.items); },
+        .ExprArrayLit(a) => {
+            ok = dump_opt(ast, buf, stack, a.elem_type, false) &&
+                 dump_list(ast, buf, stack, a.elems);
+        },
+        .ExprAggregateLit(a) => { ok = dump_list(ast, buf, stack, a.fields); },
+        .ExprComptimeCall(c) => { ok = dump_list(ast, buf, stack, c.args); },
+        .ExprBlock(b) => {
+            ok = dump_list(ast, buf, stack, b.stmts) &&
+                 dump_opt(ast, buf, stack, b.tail_expr, true);
+        },
+        .ExprIf(x) => {
+            ok = dump_opt(ast, buf, stack, x.cond, false) &&
+                 dump_opt(ast, buf, stack, x.then_block, false) &&
+                 dump_opt(ast, buf, stack, x.else_branch, true);
+        },
+        .ExprIfvar(x) => {
+            ok = dump_opt(ast, buf, stack, x.var_decl, false) &&
+                 dump_opt(ast, buf, stack, x.then_block, false) &&
+                 dump_opt(ast, buf, stack, x.else_branch, true);
+        },
+        .ExprWhile(w) => {
+            ok = dump_opt(ast, buf, stack, w.label, false) &&
+                 dump_opt(ast, buf, stack, w.cond, false) &&
+                 dump_opt(ast, buf, stack, w.body, false) &&
+                 dump_opt(ast, buf, stack, w.else_branch, true);
+        },
+        .ExprLoop(x) => {
+            ok = dump_opt(ast, buf, stack, x.label, false) &&
+                 dump_opt(ast, buf, stack, x.body, true);
+        },
+        .ExprFor(f) => {
+            ok = dump_opt(ast, buf, stack, f.label, false) &&
+                 dump_list(ast, buf, stack, f.iter_exprs) &&
+                 dump_opt(ast, buf, stack, f.pattern, false) &&
+                 dump_opt(ast, buf, stack, f.body, false) &&
+                 dump_opt(ast, buf, stack, f.else_branch, true);
+        },
+        .ExprMatch(m) => {
+            ok = dump_opt(ast, buf, stack, m.matched_expr, false) &&
+                 dump_list(ast, buf, stack, m.arms);
+        },
+        .ExprFnType(f) => {
+            ok = dump_list(ast, buf, stack, f.params) &&
+                 dump_opt(ast, buf, stack, f.ret_type, true);
+        },
+        .TypePath(p) => {},
+        .TypePointer(p) => { ok = dump_opt(ast, buf, stack, p.pointee, true); },
+        .TypeArray(a) => { ok = dump_opt(ast, buf, stack, a.elem_type, true); },
+        .TypeSlice(s) => { ok = dump_opt(ast, buf, stack, s.elem_type, true); },
+        .TypeFn(f) => {
+            ok = dump_list(ast, buf, stack, f.params) &&
+                 dump_opt(ast, buf, stack, f.ret_type, true);
+        },
+        .TypeParen(p) => { ok = dump_list(ast, buf, stack, p.types); },
+        .TypeAggregate(a) => { ok = dump_list(ast, buf, stack, a.members); },
+        .TypeComptimeCall(c) => { ok = dump_list(ast, buf, stack, c.args); },
+        .PatternIdent(x) => {},
+        .PatternUnderscore => {},
+        .PatternDot(p) => { ok = dump_opt(ast, buf, stack, p.sub_pattern, true); },
+        .PatternTuple(v) => { ok = dump_list(ast, buf, stack, v.items); },
+        .PatternList(v) => { ok = dump_list(ast, buf, stack, v.items); },
+        .PatternAggregate(v) => { ok = dump_list(ast, buf, stack, v.items); },
+        .PatternField(p) => { ok = dump_opt(ast, buf, stack, p.pattern, true); },
+    }
+
+    stack.len = stack.len - 1;
+    ret ok;
+}
+
+// 把整棵树 dump 到 out（追加，不 clear）。
+fn dump(ast: *Ast, out: *ByteBuffer): bool {
+    if !push(out, "AST:\n") { ret false; }
+    var stack = Vec.blank(1, 1);
+    defer { Vec.drop(&stack); }
+    ret dump_node(ast, out, &stack, ast.root, true);
+}
+
+// ---- 测试 ----
+
+fn test_kind_names() {
+    @assert(ByteView.eq(ByteView.from_slice(kind_name(.None)), ByteView.from_slice("none")),
+            "none name");
+    @assert(ByteView.eq(ByteView.from_slice(kind_name(.ExprBinary(.{
+                        .op = .Add,
+                        .left = 0,
+                        .right = 0,
+                    }))),
+                        ByteView.from_slice("binary")),
+            "binary name");
+}
+
+fn test_alloc_and_payload() {
+    var ast: Ast = undefined;
+    match new(null) {
+        .Ok(v) => { ast = v; },
+        .Err(error) => { @assert(false, "ast new failed"); },
+    }
+    defer { drop(&ast); }
+
+    @assert(count(&ast) == 1, "sentinel only");
+
+    var r = alloc(&ast, .ExprIdent(.{ .name = ByteView.from_slice("main") }), span_make(0, 5, 9));
+    @assert(r == 1, "first alloc is 1");
+
+    var n = node(&ast, r);
+    n.resolved_def_id = 99;
+    @assert(n.span.file == 0 && n.span.start == 5 && n.span.end == 9, "span stored");
+    @assert(n.resolved_def_id == 99, "header field");
+
+    match n.body {
+        .ExprIdent(i) => {
+            @assert(ByteView.eq(i.name, ByteView.from_slice("main")), "payload name");
+        },
+        _ => { @assert(false, "expected ident"); },
+    }
+}
+
+fn expect_bytes(buf: *ByteBuffer, expected: []u8) {
+    var actual = ByteBuffer.as_slice(buf.*);
+    @assert(actual.len == expected.len, "dump length");
+    var i: usize = 0;
+    while i < expected.len {
+        @assert(actual.ptr[i] == expected.ptr[i], "dump byte");
+        i = i + 1;
+    }
+}
+
+fn test_dump() {
+    var ast: Ast = undefined;
+    match new(null) {
+        .Ok(v) => { ast = v; },
+        .Err(error) => { @assert(false, "ast new failed"); },
+    }
+    defer { drop(&ast); }
+
+    var id_ref = alloc(&ast, .ExprIdent(.{ .name = ByteView.from_slice("main") }), span_none()); // #1
+    var decls = ref_vec();
+    match Vec.push(&decls, &id_ref as *u8) {
+        .Ok => {},
+        .Err(error) => { @assert(false, "push decl"); },
+    }
+    var cm_ref = alloc(&ast, .ContainerMembers(.{
+        .external_tags = ref_vec(),
+        .declarations = decls,
+    }), span_none()); // #2
+    var root_ref = alloc(&ast, .Root(.{ .members = cm_ref }), span_none()); // #3
+    ast.root = root_ref;
+
+    var buf = ByteBuffer.new();
+    defer { ByteBuffer.drop(&buf); }
+    @assert(dump(&ast, &buf), "dump ok");
+    // sources=null → 不解析行列；无 span 时也不打印。
+    expect_bytes(&buf, "AST:\n`-root #3\n  `-container_members #2\n    `-ident #1\n");
+}
+
+fn main() {
+    test_kind_names();
+    test_alloc_and_payload();
+    test_dump();
+}

stage2/diag.spl

@@ -0,0 +1,791 @@
+// Expect retcode = 0
+
+// 丰富诊断（参考 rustc / Clang / Swift 的设计，不照搬 stage1 C 实现）：
+//   Severity  Note/Help/Warning/Error/Fatal
+//   Code      固定语义枚举 + 稳定 E 码；集中管理，利于 diff
+//   Span      字节区间 {file,start,end}；file 是 Sources 的下标，NO_FILE 表示无位置
+//   Label     主/次标签（span + 文本），可多个
+//   Child     子诊断（note/help，可带 span），挂在条目上
+// 渲染：人类可读（源码行 + caret 下划线）与 JSON（字段稳定）双格式。
+// 位置只存下标，渲染时经 Sources 解析行列/取行，避免复制文件名。
+// 消息按拥有型 Str 保存；条目拥有其 labels/children；clear/drop 释放全部内部资源。
+
+type .{ .Error, .Status, .make } = @import("./utils/error.spl");
+type .{ .Vec } = @import("./utils/vec.spl");
+type .{ .ByteBuffer } = @import("./utils/byte_buffer.spl");
+type .{ .Str } = @import("./utils/str.spl");
+type .{ .ByteView } = @import("./utils/byte_view.spl");
+type .{ .append_usize } = @import("./utils/fmt.spl");
+type .{ .Sources, .Source, .set_new, .set_add, .set_at, .set_drop } = @import("./utils/source.spl");
+
+type Severity = enum {
+    Note,
+    Help,
+    Warning,
+    Error,
+    Fatal,
+}
+
+type Stage = enum {
+    None,
+    Lex,
+    Parse,
+    Sema,
+    Lower,
+    Vm,
+    Driver,
+}
+
+type Code = enum {
+    None,
+    // 词法 / 语法
+    LexUnexpected,
+    ParseExpected,
+    ParseUnrecognized,
+    // sema
+    UndefinedSymbol,
+    DuplicateSymbol,
+    TypeMismatch,
+    ArgCountMismatch,
+    NotAFunction,
+    NoField,
+    NoMethod,
+    IndexBaseInvalid,
+    TupleIndexOob,
+    FitsOverflow,
+    CircularDep,
+    BreakOutsideLoop,
+    ContinueOutsideLoop,
+    UndefinedLabel,
+    CastInvalid,
+    UnknownAttr,
+    UnknownIntrin,
+    IntrinArgCount,
+    VarUninitialized,
+    CannotInfer,
+    MustReturnValue,
+    // lowering
+    UnhandledNode,
+    ImplicitCastFail,
+    NoLvalue,
+    BadFieldIndex,
+    BadBlock,
+    // vm
+    IrUnknownKind,
+    FrameOverflow,
+    // 通用
+    Internal,
+}
+
+// file 下标不可用时的哨兵。
+fn no_file(): usize {
+    ret ~(0 as usize);
+}
+
+type Span = struct {
+    file: usize,
+    start: usize,
+    end: usize,
+}
+
+fn span_none(): Span {
+    ret Span { .file = no_file(), .start = 0, .end = 0 };
+}
+
+fn span_make(file: usize, start: usize, end: usize): Span {
+    ret Span { .file = file, .start = start, .end = end };
+}
+
+fn span_is_none(span: Span): bool {
+    ret span.file == no_file();
+}
+
+type Label = struct {
+    span: Span,
+    message: Str,
+    primary: bool,
+}
+
+type Child = struct {
+    severity: Severity,
+    span: Span,
+    message: Str,
+}
+
+type Item = struct {
+    severity: Severity,
+    stage: Stage,
+    code: Code,
+    span: Span,
+    message: Str,
+    labels: Vec,   // Label
+    children: Vec, // Child
+}
+
+type Diag = @self;
+
+items: Vec,        // Item
+sources: *Sources, // 借用；可为 null（则只输出消息）
+errors: usize,
+warnings: usize,
+fatals: usize,
+
+fn new(sources: *Sources): Diag {
+    ret Diag {
+        .items = Vec.blank(@sizeof(Item), @alignof(Item)),
+        .sources = sources,
+        .errors = 0,
+        .warnings = 0,
+        .fatals = 0,
+    };
+}
+
+fn item_at(self: *Diag, index: usize): *Item {
+    match Vec.get(&self.items, index) {
+        .Ok(ptr) => { ret ptr as *Item; },
+        .Err(error) => { ret null; },
+    }
+    ret null;
+}
+
+fn label_at(it: *Item, index: usize): *Label {
+    match Vec.get(&it.labels, index) {
+        .Ok(ptr) => { ret ptr as *Label; },
+        .Err(error) => { ret null; },
+    }
+    ret null;
+}
+
+fn child_at(it: *Item, index: usize): *Child {
+    match Vec.get(&it.children, index) {
+        .Ok(ptr) => { ret ptr as *Child; },
+        .Err(error) => { ret null; },
+    }
+    ret null;
+}
+
+fn last(self: *Diag): usize {
+    if self.items.len == 0 { ret no_file(); }
+    ret self.items.len - 1;
+}
+
+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 fatal_count(self: Diag): usize { ret self.fatals; }
+
+fn has_error(self: Diag): bool { ret self.errors > 0; }
+
+fn has_fatal(self: Diag): bool { ret self.fatals > 0; }
+
+// ---- 采集 ----
+
+// 追加一条诊断，返回其下标（失败返回 no_file()）。span 为主位置。
+fn emit(self: *Diag, severity: Severity, stage: Stage, code: Code, span: Span,
+        msg: []u8): usize {
+    var message = Str.new();
+    match Str.from_slice(msg) {
+        .Ok(v) => { message = v; },
+        .Err(error) => { ret no_file(); },
+    }
+    var item = Item {
+        .severity = severity,
+        .stage = stage,
+        .code = code,
+        .span = span,
+        .message = message,
+        .labels = Vec.blank(@sizeof(Label), @alignof(Label)),
+        .children = Vec.blank(@sizeof(Child), @alignof(Child)),
+    };
+    match Vec.push(&self.items, &item as *u8) {
+        .Ok => {},
+        .Err(error) => { Str.drop(&item.message); ret no_file(); },
+    }
+    match severity {
+        .Error => { self.errors = self.errors + 1; },
+        .Fatal => { self.errors = self.errors + 1; self.fatals = self.fatals + 1; },
+        .Warning => { self.warnings = self.warnings + 1; },
+        _ => {},
+    }
+    ret self.items.len - 1;
+}
+
+fn err(self: *Diag, span: Span, code: Code, msg: []u8): usize {
+    ret emit(self, .Error, .None, code, span, msg);
+}
+
+fn warn(self: *Diag, span: Span, code: Code, msg: []u8): usize {
+    ret emit(self, .Warning, .None, code, span, msg);
+}
+
+fn note(self: *Diag, span: Span, code: Code, msg: []u8): usize {
+    ret emit(self, .Note, .None, code, span, msg);
+}
+
+// 给最近一条诊断附加标签（primary 决定主/次）。
+fn attach_label(self: *Diag, span: Span, msg: []u8, primary: bool): Status {
+    var idx = last(self);
+    if idx == no_file() { ret .Err(make(.OutOfRange)); }
+    var it = item_at(self, idx);
+    if it == null { ret .Err(make(.OutOfRange)); }
+
+    var message = Str.new();
+    match Str.from_slice(msg) {
+        .Ok(v) => { message = v; },
+        .Err(error) => { ret .Err(error); },
+    }
+    var lab = Label { .span = span, .message = message, .primary = primary };
+    match Vec.push(&it.labels, &lab as *u8) {
+        .Ok => {},
+        .Err(error) => { Str.drop(&lab.message); ret .Err(error); },
+    }
+    ret .Ok;
+}
+
+// 给最近一条诊断附加子诊断（note/help 等）。
+fn attach_child(self: *Diag, severity: Severity, span: Span, msg: []u8): Status {
+    var idx = last(self);
+    if idx == no_file() { ret .Err(make(.OutOfRange)); }
+    var it = item_at(self, idx);
+    if it == null { ret .Err(make(.OutOfRange)); }
+
+    var message = Str.new();
+    match Str.from_slice(msg) {
+        .Ok(v) => { message = v; },
+        .Err(error) => { ret .Err(error); },
+    }
+    var ch = Child { .severity = severity, .span = span, .message = message };
+    match Vec.push(&it.children, &ch as *u8) {
+        .Ok => {},
+        .Err(error) => { Str.drop(&ch.message); ret .Err(error); },
+    }
+    ret .Ok;
+}
+
+// ---- 名称 ----
+
+fn severity_name(severity: Severity): []u8 {
+    match severity {
+        .Note => { ret "note"; },
+        .Help => { ret "help"; },
+        .Warning => { ret "warning"; },
+        .Error => { ret "error"; },
+        .Fatal => { ret "fatal"; },
+    }
+    ret "?";
+}
+
+fn stage_name(stage: Stage): []u8 {
+    match stage {
+        .None => { ret "none"; },
+        .Lex => { ret "lex"; },
+        .Parse => { ret "parse"; },
+        .Sema => { ret "sema"; },
+        .Lower => { ret "lower"; },
+        .Vm => { ret "vm"; },
+        .Driver => { ret "driver"; },
+    }
+    ret "?";
+}
+
+// 稳定 E 码（diff 依赖；新增 Code 时分配新码，勿改旧值）。
+fn code_str(code: Code): []u8 {
+    match code {
+        .None => { ret "E0000"; },
+        .LexUnexpected => { ret "E0001"; },
+        .ParseExpected => { ret "E0002"; },
+        .ParseUnrecognized => { ret "E0003"; },
+        .UndefinedSymbol => { ret "E0101"; },
+        .DuplicateSymbol => { ret "E0102"; },
+        .TypeMismatch => { ret "E0103"; },
+        .ArgCountMismatch => { ret "E0104"; },
+        .NotAFunction => { ret "E0105"; },
+        .NoField => { ret "E0106"; },
+        .NoMethod => { ret "E0107"; },
+        .IndexBaseInvalid => { ret "E0108"; },
+        .TupleIndexOob => { ret "E0109"; },
+        .FitsOverflow => { ret "E0110"; },
+        .CircularDep => { ret "E0111"; },
+        .BreakOutsideLoop => { ret "E0112"; },
+        .ContinueOutsideLoop => { ret "E0113"; },
+        .UndefinedLabel => { ret "E0114"; },
+        .CastInvalid => { ret "E0115"; },
+        .UnknownAttr => { ret "E0116"; },
+        .UnknownIntrin => { ret "E0117"; },
+        .IntrinArgCount => { ret "E0118"; },
+        .VarUninitialized => { ret "E0119"; },
+        .CannotInfer => { ret "E0120"; },
+        .MustReturnValue => { ret "E0121"; },
+        .UnhandledNode => { ret "E0201"; },
+        .ImplicitCastFail => { ret "E0202"; },
+        .NoLvalue => { ret "E0203"; },
+        .BadFieldIndex => { ret "E0204"; },
+        .BadBlock => { ret "E0205"; },
+        .IrUnknownKind => { ret "E0301"; },
+        .FrameOverflow => { ret "E0302"; },
+        .Internal => { ret "E9000"; },
+    }
+    ret "E0000";
+}
+
+fn code_name(code: Code): []u8 {
+    match code {
+        .None => { ret "none"; },
+        .LexUnexpected => { ret "lex_unexpected"; },
+        .ParseExpected => { ret "parse_expected"; },
+        .ParseUnrecognized => { ret "parse_unrecognized"; },
+        .UndefinedSymbol => { ret "undefined_symbol"; },
+        .DuplicateSymbol => { ret "duplicate_symbol"; },
+        .TypeMismatch => { ret "type_mismatch"; },
+        .ArgCountMismatch => { ret "arg_count_mismatch"; },
+        .NotAFunction => { ret "not_a_function"; },
+        .NoField => { ret "no_field"; },
+        .NoMethod => { ret "no_method"; },
+        .IndexBaseInvalid => { ret "index_base_invalid"; },
+        .TupleIndexOob => { ret "tuple_index_oob"; },
+        .FitsOverflow => { ret "fits_overflow"; },
+        .CircularDep => { ret "circular_dep"; },
+        .BreakOutsideLoop => { ret "break_outside_loop"; },
+        .ContinueOutsideLoop => { ret "continue_outside_loop"; },
+        .UndefinedLabel => { ret "undefined_label"; },
+        .CastInvalid => { ret "cast_invalid"; },
+        .UnknownAttr => { ret "unknown_attr"; },
+        .UnknownIntrin => { ret "unknown_intrin"; },
+        .IntrinArgCount => { ret "intrin_arg_count"; },
+        .VarUninitialized => { ret "var_uninitialized"; },
+        .CannotInfer => { ret "cannot_infer"; },
+        .MustReturnValue => { ret "must_return_value"; },
+        .UnhandledNode => { ret "unhandled_node"; },
+        .ImplicitCastFail => { ret "implicit_cast_fail"; },
+        .NoLvalue => { ret "no_lvalue"; },
+        .BadFieldIndex => { ret "bad_field_index"; },
+        .BadBlock => { ret "bad_block"; },
+        .IrUnknownKind => { ret "ir_unknown_kind"; },
+        .FrameOverflow => { ret "frame_overflow"; },
+        .Internal => { ret "internal"; },
+    }
+    ret "none";
+}
+
+// ---- 渲染：公共小工具 ----
+
+fn append_num(out: *ByteBuffer, value: usize): Status {
+    ret append_usize(out, value, 10, false);
+}
+
+fn append_str(out: *ByteBuffer, s: *Str): Status {
+    ret ByteBuffer.append(out, Str.as_slice(s));
+}
+
+fn append_view(out: *ByteBuffer, view: ByteView): Status {
+    ret ByteBuffer.append(out, ByteView.as_slice(view));
+}
+
+type Resolved = struct {
+    ok: bool,
+    fname: ByteView,
+    line: usize,
+    col: usize,
+    text: ByteView,
+    width: usize,
+}
+
+fn resolve(self: *Diag, span: Span): Resolved {
+    var r = Resolved {
+        .ok = false,
+        .fname = ByteView.empty(),
+        .line = 0,
+        .col = 0,
+        .text = ByteView.empty(),
+        .width = 1,
+    };
+    if self.sources == null { ret r; }
+    if span_is_none(span) { ret r; }
+
+    var src = set_at(self.sources, span.file);
+    if src == null { ret r; }
+
+    var pos = Source.line_col(src.*, span.start);
+    var width: usize = 1;
+    if span.end > span.start { width = span.end - span.start; }
+    r.ok = true;
+    r.fname = src.fname;
+    r.line = pos.line;
+    r.col = pos.col;
+    r.text = Source.line_text(src.*, pos.line);
+    r.width = width;
+    ret r;
+}
+
+fn render_header(it: *Item, r: *Resolved, out: *ByteBuffer): Status {
+    if r.ok {
+        match append_view(out, r.fname) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match ByteBuffer.append_byte(out, ':') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match append_num(out, r.line) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match ByteBuffer.append_byte(out, ':') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match append_num(out, r.col) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match ByteBuffer.append(out, ": ") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    }
+    match ByteBuffer.append(out, severity_name(it.severity)) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match ByteBuffer.append_byte(out, '[') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match ByteBuffer.append(out, code_str(it.code)) { .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); } }
+    ret ByteBuffer.append_byte(out, '\n');
+}
+
+// 输出一行源码 + caret 下划线（列范围按字节；制表符不特殊处理）。
+fn render_span_visual(self: *Diag, span: Span, out: *ByteBuffer): Status {
+    var r = resolve(self, span);
+    if !r.ok { ret .Ok; }
+
+    match ByteBuffer.append(out, "    ") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match append_view(out, r.text) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match ByteBuffer.append_byte(out, '\n') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+
+    match ByteBuffer.append(out, "    ") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    var pad: usize = 0;
+    if r.col > 0 { pad = r.col - 1; }
+    match ByteBuffer.append_repeat(out, ' ', pad) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+
+    var width = r.width;
+    if width == 0 { width = 1; }
+    if r.col > 0 {
+        var room: usize = 0;
+        if r.text.len >= pad { room = r.text.len - pad; }
+        if room == 0 { room = 1; }
+        if width > room { width = room; }
+    }
+    var k: usize = 0;
+    while k < width {
+        var ch: u8 = '~';
+        if k == 0 { ch = '^'; }
+        match ByteBuffer.append_byte(out, ch) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        k = k + 1;
+    }
+    ret ByteBuffer.append_byte(out, '\n');
+}
+
+fn render_child_line(ch: *Child, out: *ByteBuffer): Status {
+    match ByteBuffer.append(out, "    ") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match ByteBuffer.append(out, severity_name(ch.severity)) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match ByteBuffer.append(out, ": ") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match append_str(out, &ch.message) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    ret ByteBuffer.append_byte(out, '\n');
+}
+
+fn render_human(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)); }
+
+        var r = resolve(self, it.span);
+        match render_header(it, &r, out) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        match render_span_visual(self, it.span, out) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+
+        var j: usize = 0;
+        while j < it.labels.len {
+            var lab = label_at(it, j);
+            if lab != null {
+                match render_span_visual(self, lab.span, out) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+                if Str.size(lab.message) != 0 {
+                    match ByteBuffer.append(out, "    ") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+                    match append_str(out, &lab.message) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+                    match ByteBuffer.append_byte(out, '\n') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+                }
+            }
+            j = j + 1;
+        }
+
+        var c: usize = 0;
+        while c < it.children.len {
+            var ch = child_at(it, c);
+            if ch != null {
+                match render_child_line(ch, out) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+                match render_span_visual(self, ch.span, out) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            }
+            c = c + 1;
+        }
+
+        if i + 1 < self.items.len {
+            match ByteBuffer.append_byte(out, '\n') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        }
+        i = i + 1;
+    }
+    ret .Ok;
+}
+
+// ---- JSON 渲染（字段稳定，供 LSP / diff 工具）----
+
+fn append_json_hex4(out: *ByteBuffer, b: u8): Status {
+    var hex: []u8 = "0123456789abcdef";
+    match ByteBuffer.append(out, "\\u00") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match ByteBuffer.append_byte(out, hex[(b >> 4) as usize]) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    ret ByteBuffer.append_byte(out, hex[(b & 15) as usize]);
+}
+
+fn append_json_str(out: *ByteBuffer, bytes: []u8): Status {
+    match ByteBuffer.append_byte(out, '"') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    var i: usize = 0;
+    while i < bytes.len {
+        var b = bytes[i];
+        if b == '"' {
+            match ByteBuffer.append(out, "\\\"") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        } else if b == '\\' {
+            match ByteBuffer.append(out, "\\\\") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        } else if b == '\n' {
+            match ByteBuffer.append(out, "\\n") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        } else if b == '\r' {
+            match ByteBuffer.append(out, "\\r") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        } else if b == '\t' {
+            match ByteBuffer.append(out, "\\t") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        } else if b < ' ' {
+            match append_json_hex4(out, b) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        } else {
+            match ByteBuffer.append_byte(out, b) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        }
+        i = i + 1;
+    }
+    ret ByteBuffer.append_byte(out, '"');
+}
+
+fn append_str_json(out: *ByteBuffer, s: *Str): Status {
+    ret append_json_str(out, Str.as_slice(s));
+}
+
+fn append_bool(out: *ByteBuffer, value: bool): Status {
+    if value { ret ByteBuffer.append(out, "true"); }
+    ret ByteBuffer.append(out, "false");
+}
+
+fn append_loc_json(self: *Diag, span: Span, out: *ByteBuffer): Status {
+    var r = resolve(self, span);
+    match ByteBuffer.append(out, "{\"file\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    if r.ok {
+        match append_json_str(out, ByteView.as_slice(r.fname)) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    } else {
+        match ByteBuffer.append(out, "null") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    }
+    match ByteBuffer.append(out, ",\"start\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match append_num(out, span.start) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match ByteBuffer.append(out, ",\"end\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match append_num(out, span.end) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match ByteBuffer.append(out, ",\"line\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    if r.ok {
+        match append_num(out, r.line) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    } else {
+        match ByteBuffer.append(out, "null") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    }
+    match ByteBuffer.append(out, ",\"col\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    if r.ok {
+        match append_num(out, r.col) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    } else {
+        match ByteBuffer.append(out, "null") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    }
+    ret ByteBuffer.append_byte(out, '}');
+}
+
+fn render_item_json(self: *Diag, it: *Item, out: *ByteBuffer): Status {
+    match ByteBuffer.append(out, "{\"severity\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match append_json_str(out, severity_name(it.severity)) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match ByteBuffer.append(out, ",\"stage\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match append_json_str(out, stage_name(it.stage)) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match ByteBuffer.append(out, ",\"code\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match append_json_str(out, code_str(it.code)) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match ByteBuffer.append(out, ",\"name\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match append_json_str(out, code_name(it.code)) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match ByteBuffer.append(out, ",\"message\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match append_str_json(out, &it.message) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+
+    match ByteBuffer.append(out, ",\"span\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match append_loc_json(self, it.span, out) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+
+    match ByteBuffer.append(out, ",\"labels\":[") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    var j: usize = 0;
+    while j < it.labels.len {
+        var lab = label_at(it, j);
+        if lab != null {
+            if j != 0 {
+                match ByteBuffer.append_byte(out, ',') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            }
+            match ByteBuffer.append(out, "{\"message\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            match append_str_json(out, &lab.message) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            match ByteBuffer.append(out, ",\"primary\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            match append_bool(out, lab.primary) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            match ByteBuffer.append(out, ",\"span\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            match append_loc_json(self, lab.span, out) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            match ByteBuffer.append_byte(out, '}') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        }
+        j = j + 1;
+    }
+    match ByteBuffer.append_byte(out, ']') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+
+    match ByteBuffer.append(out, ",\"children\":[") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    var c: usize = 0;
+    while c < it.children.len {
+        var ch = child_at(it, c);
+        if ch != null {
+            if c != 0 {
+                match ByteBuffer.append_byte(out, ',') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            }
+            match ByteBuffer.append(out, "{\"severity\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            match append_json_str(out, severity_name(ch.severity)) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            match ByteBuffer.append(out, ",\"message\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            match append_str_json(out, &ch.message) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            match ByteBuffer.append(out, ",\"span\":") { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            match append_loc_json(self, ch.span, out) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+            match ByteBuffer.append_byte(out, '}') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        }
+        c = c + 1;
+    }
+    match ByteBuffer.append_byte(out, ']') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    ret ByteBuffer.append_byte(out, '}');
+}
+
+fn render_json(self: *Diag, out: *ByteBuffer): Status {
+    match ByteBuffer.append_byte(out, '[') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    var i: usize = 0;
+    while i < self.items.len {
+        var it = item_at(self, i);
+        if it == null { ret .Err(make(.OutOfRange)); }
+        if i != 0 {
+            match ByteBuffer.append_byte(out, ',') { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        }
+        match render_item_json(self, it, out) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+        i = i + 1;
+    }
+    ret ByteBuffer.append_byte(out, ']');
+}
+
+// ---- 生命周期 ----
+
+fn drop_item(it: *Item) {
+    Str.drop(&it.message);
+    var j: usize = 0;
+    while j < it.labels.len {
+        var lab = label_at(it, j);
+        if lab != null { Str.drop(&lab.message); }
+        j = j + 1;
+    }
+    Vec.drop(&it.labels);
+    var c: usize = 0;
+    while c < it.children.len {
+        var ch = child_at(it, c);
+        if ch != null { Str.drop(&ch.message); }
+        c = c + 1;
+    }
+    Vec.drop(&it.children);
+}
+
+fn clear(self: *Diag) {
+    var i: usize = 0;
+    while i < self.items.len {
+        var it = item_at(self, i);
+        if it != null { drop_item(it); }
+        i = i + 1;
+    }
+    Vec.clear(&self.items);
+    self.errors = 0;
+    self.warnings = 0;
+    self.fatals = 0;
+}
+
+fn drop(self: *Diag) {
+    clear(self);
+    Vec.drop(&self.items);
+    self.sources = null;
+}
+
+// ---- 测试辅助 ----
+
+fn view(text: []u8): ByteView {
+    ret ByteView.from_slice(text);
+}
+
+fn expect_ok(result: Status) {
+    match result {
+        .Ok => {},
+        .Err(error) => { @assert(false, "unexpected diag error"); },
+    }
+}
+
+fn has_bytes(out: *ByteBuffer, needle: []u8): bool {
+    ret ByteView.contains(ByteView.from_slice(ByteBuffer.as_slice(out.*)), view(needle));
+}
+
+fn test_collect_render_human() {
+    var sources = set_new();
+    match set_add(&sources, view("a.spl"), view("fn main() {\n    y;\n}\n")) {
+        .Ok(index) => {},
+        .Err(error) => { @assert(false, "add source"); },
+    }
+
+    var d = new(&sources);
+    var idx = err(&d, span_make(0, 16, 17), .UndefinedSymbol, "undefined symbol `y`");
+    @assert(idx != no_file(), "emit");
+    expect_ok(attach_label(&d, span_make(0, 16, 17), "not found in scope", false));
+    expect_ok(attach_child(&d, .Note, span_none(), "did you mean `x`?"));
+
+    @assert(count(d) == 1, "count");
+    @assert(error_count(d) == 1 && has_error(d), "errors");
+    @assert(warning_count(d) == 0 && !has_fatal(d), "warnings/fatal");
+
+    var out = ByteBuffer.new();
+    expect_ok(render_human(&d, &out));
+    @assert(has_bytes(&out, "a.spl:2:5: error[E0101]: undefined symbol `y`"), "human header");
+    @assert(has_bytes(&out, "    y;"), "human source line");
+    @assert(has_bytes(&out, "    ^"), "human caret");
+    @assert(has_bytes(&out, "not found in scope"), "human label");
+    @assert(has_bytes(&out, "note: did you mean `x`?"), "human child");
+    ByteBuffer.drop(&out);
+
+    clear(&d);
+    @assert(count(d) == 0 && error_count(d) == 0, "clear");
+    drop(&d);
+    set_drop(&sources);
+}
+
+fn test_render_json() {
+    var sources = set_new();
+    match set_add(&sources, view("b.spl"), view("let z = 1;\n")) {
+        .Ok(index) => {},
+        .Err(error) => { @assert(false, "add source"); },
+    }
+
+    var d = new(&sources);
+    err(&d, span_make(0, 4, 5), .TypeMismatch, "expected i32, got bool");
+    expect_ok(attach_child(&d, .Help, span_none(), "annotate the type"));
+
+    var out = ByteBuffer.new();
+    expect_ok(render_json(&d, &out));
+    @assert(has_bytes(&out, "\"severity\":\"error\""), "json severity");
+    @assert(has_bytes(&out, "\"code\":\"E0103\""), "json code");
+    @assert(has_bytes(&out, "\"name\":\"type_mismatch\""), "json name");
+    @assert(has_bytes(&out, "\"line\":1"), "json line");
+    @assert(has_bytes(&out, "\"col\":5"), "json col");
+    @assert(has_bytes(&out, "\"children\":[{\"severity\":\"help\""), "json child");
+    ByteBuffer.drop(&out);
+
+    drop(&d);
+    set_drop(&sources);
+}
+
+fn test_empty_and_headerless() {
+    var d = new(null);
+    err(&d, span_none(), .Internal, "no location");
+    var out = ByteBuffer.new();
+    expect_ok(render_human(&d, &out));
+    @assert(has_bytes(&out, "error[E9000]: no location"), "headerless");
+    ByteBuffer.drop(&out);
+    drop(&d);
+}
+
+fn main() {
+    test_collect_render_human();
+    test_render_json();
+    test_empty_and_headerless();
+}

stage2/lexer.spl

@@ -0,0 +1,719 @@
+// Expect retcode = 0
+
+// 词法分析器：字节流 -> Vec<Token>。镜像 stage1/spl_lexer.c 的行为：
+// 行注释作为 token 产出；块注释（可嵌套）整体跳过；换行产出 Endline；EOF 产出 Eof。
+// 复用 utils/ascii 做字符分类；整型/字符在词法期解码后放进 TokenKind 载荷。
+// Token.text 借用输入字节，不复制。
+
+type .{ .Vec, .VecResult } = @import("utils/vec.spl");
+type .{ .ByteView } = @import("utils/byte_view.spl");
+type .{ .is_digit, .is_alpha, .is_alnum, .is_hex_digit, .digit_value } = @import("utils/ascii.spl");
+type .{
+    .Token, .TokenKind, .Radix, .Keyword, .Punct, .keyword_of
+} = @import("token.spl");
+type .{ .Span, .span_make } = @import("./diag.spl");
+type Diag = @import("./diag.spl");
+
+type Lexer = @self;
+
+src: ByteView,
+pos: usize,
+line: usize,
+col: usize,
+diag: *Diag, // 借用；可为 null（合法输入不产生诊断）
+file: usize,
+
+fn new(src: ByteView, diag: *Diag, file: usize): Lexer {
+    ret Lexer { .src = src, .pos = 0, .line = 1, .col = 1, .diag = diag, .file = file };
+}
+
+// 记录一条词法诊断（非法输入路径）。合法输入不应触发。
+fn report(self: *Lexer, start: usize, end: usize, message: []u8) {
+    if self.diag == null { ret; }
+    Diag.err(self.diag, span_make(self.file, start, end), .LexUnexpected, message);
+}
+
+fn is_ident_start(c: u8): bool { ret is_alpha(c) || c == '_'; }
+
+fn is_ident_cont(c: u8): bool { ret is_alnum(c) || c == '_'; }
+
+fn hex_val(c: u8): u8 {
+    match digit_value(c, 16) {
+        .Ok(v) => { ret v; },
+        .Err(error) => { ret 0; },
+    }
+    ret 0;
+}
+
+// 文本形如 'a' / '\n' / '\x41'；返回解码后的字节。
+fn decode_char(text: ByteView): u8 {
+    var start: usize = 1;
+    var end = text.len;
+    if text.len >= 2 && text.ptr[text.len - 1] == '\'' { end = text.len - 1; }
+    if end <= start { ret 0; }
+
+    var b = text.ptr[start];
+    if b != '\\' { ret b; }
+
+    start = start + 1;
+    if start >= end { ret 0; }
+    var e = text.ptr[start];
+    if e == 'n' { ret '\n'; }
+    if e == 't' { ret '\t'; }
+    if e == 'r' { ret '\r'; }
+    if e == '\\' { ret '\\'; }
+    if e == '"' { ret '"'; }
+    if e == '\'' { ret '\''; }
+    if e == '0' { ret 0; }
+    if e == 'x' || e == 'X' {
+        var q = start + 1;
+        var value: u8 = 0;
+        var digits: usize = 0;
+        while q < end && digits < 2 && is_hex_digit(text.ptr[q]) {
+            value = value * 16 + hex_val(text.ptr[q]);
+            q = q + 1;
+            digits = digits + 1;
+        }
+        if digits == 0 { ret e; }
+        ret value;
+    }
+    ret e;
+}
+
+// ---- 读取辅助 ----
+
+fn has2(self: *Lexer, ch: u8): bool {
+    ret self.pos + 1 < self.src.len && self.src.ptr[self.pos + 1] == ch;
+}
+
+fn has3(self: *Lexer, ch1: u8, ch2: u8): bool {
+    ret self.pos + 2 < self.src.len && self.src.ptr[self.pos + 1] == ch1 &&
+        self.src.ptr[self.pos + 2] == ch2;
+}
+
+fn sub(self: *Lexer, start: usize, end: usize): ByteView {
+    match ByteView.slice(self.src, start, end) {
+        .Ok(v) => { ret v; },
+        .Err(error) => { ret ByteView.empty(); },
+    }
+    ret ByteView.empty();
+}
+
+// 产出 token 并把 pos/col 推进到 end。Vec.push 失败视为致命（同 vec.make_vec 风格）。
+fn emit_to(self: *Lexer, out: *Vec, kind: TokenKind, start: usize, end: usize, line: usize,
+           col: usize) {
+    var t = Token.make(kind, sub(self, start, end), line, col, start);
+    match Vec.push(out, &t as *u8) {
+        .Ok => {},
+        .Err(error) => { @assert(false, "lexer: Vec.push failed"); },
+    }
+    self.pos = end;
+    self.col = col + (end - start);
+}
+
+fn op1(self: *Lexer, out: *Vec, punct: Punct, start: usize, line: usize, col: usize) {
+    emit_to(self, out, .Punct(punct), start, start + 1, line, col);
+}
+
+fn op2(self: *Lexer, out: *Vec, punct: Punct, start: usize, line: usize, col: usize) {
+    emit_to(self, out, .Punct(punct), start, start + 2, line, col);
+}
+
+fn op3(self: *Lexer, out: *Vec, punct: Punct, start: usize, line: usize, col: usize) {
+    emit_to(self, out, .Punct(punct), start, start + 3, line, col);
+}
+
+// ---- 复合扫描 ----
+
+fn scan_int(self: *Lexer, out: *Vec, start: usize, line: usize, col: usize) {
+    var p = self.pos;
+    if self.src.ptr[p] == '0' && p + 1 < self.src.len {
+        var nc = self.src.ptr[p + 1];
+        if nc == 'x' || nc == 'X' {
+            p = p + 2;
+            while p < self.src.len && (is_hex_digit(self.src.ptr[p]) || self.src.ptr[p] == '_') {
+                p = p + 1;
+            }
+            emit_to(self, out, .Int(.Hex), start, p, line, col);
+            ret;
+        }
+        if nc == 'b' || nc == 'B' {
+            p = p + 2;
+            while p < self.src.len && (self.src.ptr[p] == '0' || self.src.ptr[p] == '1' ||
+                                       self.src.ptr[p] == '_') {
+                p = p + 1;
+            }
+            emit_to(self, out, .Int(.Bin), start, p, line, col);
+            ret;
+        }
+        if nc == 'o' || nc == 'O' {
+            p = p + 2;
+            while p < self.src.len &&
+                   ((self.src.ptr[p] >= '0' && self.src.ptr[p] <= '7') ||
+                    self.src.ptr[p] == '_') {
+                p = p + 1;
+            }
+            emit_to(self, out, .Int(.Oct), start, p, line, col);
+            ret;
+        }
+    }
+
+    while p < self.src.len && (is_digit(self.src.ptr[p]) || self.src.ptr[p] == '_') { p = p + 1; }
+    if p + 1 < self.src.len && self.src.ptr[p] == '.' && is_digit(self.src.ptr[p + 1]) {
+        p = p + 1;
+        // 小数部分允许下划线分隔（Rust 风格）。
+        while p < self.src.len && (is_digit(self.src.ptr[p]) || self.src.ptr[p] == '_') {
+            p = p + 1;
+        }
+        // 可选指数：e/E [+-] 数字（数字允许下划线）；仅当确有数字时纳入。
+        if p < self.src.len && (self.src.ptr[p] == 'e' || self.src.ptr[p] == 'E') {
+            var q = p + 1;
+            if q < self.src.len && (self.src.ptr[q] == '+' || self.src.ptr[q] == '-') { q = q + 1; }
+            if q < self.src.len && is_digit(self.src.ptr[q]) {
+                p = q + 1;
+                while p < self.src.len && (is_digit(self.src.ptr[p]) || self.src.ptr[p] == '_') {
+                    p = p + 1;
+                }
+            }
+        }
+        emit_to(self, out, .Float, start, p, line, col);
+        ret;
+    }
+    emit_to(self, out, .Int(.Dec), start, p, line, col);
+}
+
+fn scan_char(self: *Lexer, out: *Vec, start: usize, line: usize, col: usize) {
+    var p = self.pos + 1;
+    var closed = false;
+    var escape_ok = true;
+    if p < self.src.len {
+        if self.src.ptr[p] == '\\' {
+            p = p + 1;
+            if p < self.src.len && (self.src.ptr[p] == 'x' || self.src.ptr[p] == 'X') {
+                p = p + 1;
+                var digits: usize = 0;
+                while p < self.src.len && digits < 2 && is_hex_digit(self.src.ptr[p]) {
+                    p = p + 1;
+                    digits = digits + 1;
+                }
+                if digits == 0 { escape_ok = false; }
+            } else if p < self.src.len {
+                var e = self.src.ptr[p];
+                if !(e == 'n' || e == 't' || e == 'r' || e == '\\' || e == '"' || e == '\'' ||
+                     e == '0') {
+                    escape_ok = false;
+                }
+                p = p + 1;
+            } else {
+                escape_ok = false;
+            }
+        } else {
+            p = p + 1;
+        }
+    }
+    if p < self.src.len && self.src.ptr[p] == '\'' {
+        p = p + 1;
+        closed = true;
+    }
+    if !closed {
+        report(self, start, p, "unterminated character literal");
+    } else if !escape_ok {
+        report(self, start, p, "invalid escape in character literal");
+    }
+    emit_to(self, out, .Char(decode_char(sub(self, start, p))), start, p, line, col);
+}
+
+fn scan_string(self: *Lexer, out: *Vec, start: usize, line: usize, col: usize) {
+    var p = self.pos + 1;
+    var l = line;
+    var cc = col + 1;
+    var closed = false;
+    while p < self.src.len {
+        var b = self.src.ptr[p];
+        if b == '\\' {
+            p = p + 1;
+            cc = cc + 1;
+            if p < self.src.len {
+                p = p + 1;
+                cc = cc + 1;
+            }
+        } else if b == '"' {
+            p = p + 1;
+            cc = cc + 1;
+            closed = true;
+            break;
+        } else if b == '\n' {
+            l = l + 1;
+            cc = 1;
+            p = p + 1;
+        } else {
+            p = p + 1;
+            cc = cc + 1;
+        }
+    }
+    if !closed { report(self, start, p, "unterminated string literal"); }
+    emit_to(self, out, .Str, start, p, line, col);
+    self.line = l;
+    self.col = cc;
+}
+
+fn scan_ident(self: *Lexer, out: *Vec, start: usize, line: usize, col: usize) {
+    var p = self.pos;
+    while p < self.src.len && is_ident_cont(self.src.ptr[p]) { p = p + 1; }
+    var text = sub(self, start, p);
+    match keyword_of(text) {
+        .Yes(kw) => { emit_to(self, out, .Keyword(kw), start, p, line, col); },
+        .No => { emit_to(self, out, .Ident, start, p, line, col); },
+    }
+}
+
+// 块注释：可嵌套，整体跳过（不产出 token）。未闭合时报诊断。
+fn skip_block_comment(self: *Lexer, start: usize) {
+    self.pos = self.pos + 2;
+    self.col = self.col + 2;
+    var depth: usize = 1;
+    while self.pos + 1 < self.src.len && depth > 0 {
+        if self.src.ptr[self.pos] == '*' && self.src.ptr[self.pos + 1] == '/' {
+            depth = depth - 1;
+            self.pos = self.pos + 2;
+            self.col = self.col + 2;
+        } else if self.src.ptr[self.pos] == '/' && self.src.ptr[self.pos + 1] == '*' {
+            depth = depth + 1;
+            self.pos = self.pos + 2;
+            self.col = self.col + 2;
+        } else {
+            if self.src.ptr[self.pos] == '\n' {
+                self.line = self.line + 1;
+                self.col = 1;
+            } else {
+                self.col = self.col + 1;
+            }
+            self.pos = self.pos + 1;
+        }
+    }
+    if depth > 0 {
+        report(self, start, self.src.len, "unterminated block comment");
+    }
+}
+
+fn run(self: *Lexer, out: *Vec) {
+    while self.pos < self.src.len {
+        var start = self.pos;
+        var line = self.line;
+        var col = self.col;
+        var c = self.src.ptr[self.pos];
+
+        if c == ' ' || c == '\t' || c == '\r' {
+            self.pos = self.pos + 1;
+            self.col = self.col + 1;
+            continue;
+        }
+        if c == '\n' {
+            emit_to(self, out, .Endline, start, start + 1, line, col);
+            self.line = line + 1;
+            self.col = 1;
+            continue;
+        }
+        if c == '/' && has2(self, '/') {
+            var p = self.pos;
+            while p < self.src.len && self.src.ptr[p] != '\n' { p = p + 1; }
+            emit_to(self, out, .LineComment, start, p, line, col);
+            continue;
+        }
+        if c == '/' && has2(self, '*') {
+            skip_block_comment(self, start);
+            continue;
+        }
+        if c == '\'' {
+            scan_char(self, out, start, line, col);
+            continue;
+        }
+        if c == '"' {
+            scan_string(self, out, start, line, col);
+            continue;
+        }
+        if is_ident_start(c) {
+            scan_ident(self, out, start, line, col);
+            continue;
+        }
+        if is_digit(c) {
+            scan_int(self, out, start, line, col);
+            continue;
+        }
+
+        if c == '<' {
+            if has3(self, '-', '>') { op3(self, out, .ArrowBoth, start, line, col); continue; }
+            if has3(self, '=', '>') { op3(self, out, .FatArrowBoth, start, line, col); continue; }
+            if has3(self, '<', '=') { op3(self, out, .LShAssign, start, line, col); continue; }
+            if has2(self, '<') { op2(self, out, .LSh, start, line, col); continue; }
+            if has2(self, '-') { op2(self, out, .LeftArrow, start, line, col); continue; }
+            if has2(self, '=') { op2(self, out, .Le, start, line, col); continue; }
+            op1(self, out, .Lt, start, line, col);
+            continue;
+        }
+        if c == '>' {
+            if has3(self, '>', '=') { op3(self, out, .RShAssign, start, line, col); continue; }
+            if has2(self, '>') { op2(self, out, .RSh, start, line, col); continue; }
+            if has2(self, '=') { op2(self, out, .Ge, start, line, col); continue; }
+            op1(self, out, .Gt, start, line, col);
+            continue;
+        }
+        if c == '.' {
+            if has3(self, '.', '.') { op3(self, out, .Ellipsis, start, line, col); continue; }
+            if has2(self, '.') { op2(self, out, .Range, start, line, col); continue; }
+            op1(self, out, .Dot, start, line, col);
+            continue;
+        }
+        if c == '=' {
+            if has2(self, '=') { op2(self, out, .Eq, start, line, col); continue; }
+            if has2(self, '>') { op2(self, out, .RightFatArrow, start, line, col); continue; }
+            op1(self, out, .Assign, start, line, col);
+            continue;
+        }
+        if c == '!' {
+            if has2(self, '=') { op2(self, out, .Neq, start, line, col); continue; }
+            op1(self, out, .Not, start, line, col);
+            continue;
+        }
+        if c == '&' {
+            if has2(self, '&') { op2(self, out, .AndAnd, start, line, col); continue; }
+            if has2(self, '=') { op2(self, out, .AndAssign, start, line, col); continue; }
+            op1(self, out, .And, start, line, col);
+            continue;
+        }
+        if c == '|' {
+            if has2(self, '|') { op2(self, out, .OrOr, start, line, col); continue; }
+            if has2(self, '=') { op2(self, out, .OrAssign, start, line, col); continue; }
+            op1(self, out, .Or, start, line, col);
+            continue;
+        }
+        if c == '+' {
+            if has2(self, '=') { op2(self, out, .AddAssign, start, line, col); continue; }
+            op1(self, out, .Add, start, line, col);
+            continue;
+        }
+        if c == '-' {
+            if has2(self, '=') { op2(self, out, .SubAssign, start, line, col); continue; }
+            if has2(self, '>') { op2(self, out, .RightArrow, start, line, col); continue; }
+            op1(self, out, .Sub, start, line, col);
+            continue;
+        }
+        if c == '*' {
+            if has2(self, '=') { op2(self, out, .MulAssign, start, line, col); continue; }
+            op1(self, out, .Mul, start, line, col);
+            continue;
+        }
+        if c == '/' {
+            if has2(self, '=') { op2(self, out, .DivAssign, start, line, col); continue; }
+            op1(self, out, .Div, start, line, col);
+            continue;
+        }
+        if c == '%' {
+            if has2(self, '=') { op2(self, out, .ModAssign, start, line, col); continue; }
+            op1(self, out, .Mod, start, line, col);
+            continue;
+        }
+        if c == '^' {
+            if has2(self, '=') { op2(self, out, .XorAssign, start, line, col); continue; }
+            op1(self, out, .Xor, start, line, col);
+            continue;
+        }
+        if c == ':' {
+            if has2(self, '=') { op2(self, out, .ColonAssign, start, line, col); continue; }
+            op1(self, out, .Colon, start, line, col);
+            continue;
+        }
+        if c == '~' { op1(self, out, .BitNot, start, line, col); continue; }
+        if c == ',' { op1(self, out, .Comma, start, line, col); continue; }
+        if c == ';' { op1(self, out, .Semicolon, start, line, col); continue; }
+        if c == '(' { op1(self, out, .LParen, start, line, col); continue; }
+        if c == ')' { op1(self, out, .RParen, start, line, col); continue; }
+        if c == '[' { op1(self, out, .LBracket, start, line, col); continue; }
+        if c == ']' { op1(self, out, .RBracket, start, line, col); continue; }
+        if c == '{' { op1(self, out, .LBrace, start, line, col); continue; }
+        if c == '}' { op1(self, out, .RBrace, start, line, col); continue; }
+        if c == '#' { op1(self, out, .Sharp, start, line, col); continue; }
+        if c == '@' { op1(self, out, .At, start, line, col); continue; }
+        if c == '$' { op1(self, out, .Dollar, start, line, col); continue; }
+        if c == '?' { op1(self, out, .Cond, start, line, col); continue; }
+
+        report(self, start, start + 1, "unexpected character");
+        emit_to(self, out, .Unknown(c), start, start + 1, line, col);
+    }
+
+    var end = self.src.len;
+    emit_to(self, out, .Eof, end, end, self.line, self.col);
+}
+
+fn lex(src: ByteView, diag: *Diag, file: usize): VecResult {
+    var out = Vec.blank(@sizeof(Token), @alignof(Token));
+    match Vec.new(@sizeof(Token), @alignof(Token)) {
+        .Ok(v) => { out = v; },
+        .Err(error) => { ret .Err(error); },
+    }
+    var self = new(src, diag, file);
+    run(&self, &out);
+    ret .Ok(out);
+}
+
+// ---- 测试辅助 ----
+
+fn view(text: []u8): ByteView {
+    ret ByteView.from_slice(text);
+}
+
+fn lex_ok(src: []u8): Vec {
+    var out = Vec.blank(@sizeof(Token), @alignof(Token));
+    match lex(view(src), null, 0) {
+        .Ok(v) => { out = v; },
+        .Err(error) => { @assert(false, "lex failed"); },
+    }
+    ret out;
+}
+
+fn at(v: *Vec, index: usize): Token {
+    match Vec.get(v, index) {
+        .Ok(ptr) => { ret (ptr as *Token).*; },
+        .Err(error) => { ret Token.make(.Unknown(0), ByteView.empty(), 0, 0, 0); },
+    }
+    ret Token.make(.Unknown(0), ByteView.empty(), 0, 0, 0);
+}
+
+fn expect_punct(tok: Token, expected: Punct) {
+    match tok.kind {
+        .Punct(p) => { @assert(@enum2int(p) == @enum2int(expected), "punct kind"); },
+        _ => { @assert(false, "expected punct"); },
+    }
+}
+
+fn expect_keyword(tok: Token, expected: Keyword) {
+    match tok.kind {
+        .Keyword(k) => { @assert(@enum2int(k) == @enum2int(expected), "keyword kind"); },
+        _ => { @assert(false, "expected keyword"); },
+    }
+}
+
+fn expect_ident(tok: Token) {
+    match tok.kind {
+        .Ident => {},
+        _ => { @assert(false, "expected ident"); },
+    }
+}
+
+fn expect_int(tok: Token, radix: Radix) {
+    match tok.kind {
+        .Int(r) => {
+            @assert(@enum2int(r) == @enum2int(radix), "int radix");
+        },
+        _ => { @assert(false, "expected int"); },
+    }
+}
+
+fn expect_char(tok: Token, value: u8) {
+    match tok.kind {
+        .Char(v) => { @assert(v == value, "char value"); },
+        _ => { @assert(false, "expected char"); },
+    }
+}
+
+fn expect_endline(tok: Token) {
+    match tok.kind {
+        .Endline => {},
+        _ => { @assert(false, "expected endline"); },
+    }
+}
+
+fn expect_eof(tok: Token) {
+    match tok.kind {
+        .Eof => {},
+        _ => { @assert(false, "expected eof"); },
+    }
+}
+
+fn expect_text(tok: Token, literal: []u8) {
+    @assert(ByteView.eq(tok.text, view(literal)), "token text");
+}
+
+fn expect_pos(tok: Token, line: usize, col: usize) {
+    @assert(tok.line == line && tok.col == col, "token position");
+}
+
+fn test_basic() {
+    var v = lex_ok("fn main() {\n    ret 0;\n}\n");
+    defer { Vec.drop(&v); }
+    @assert(v.len == 13, "basic token count");
+
+    var t0 = at(&v, 0);
+    expect_keyword(t0, .Fn);
+    expect_text(t0, "fn");
+    expect_pos(t0, 1, 1);
+    expect_ident(at(&v, 1));
+    expect_text(at(&v, 1), "main");
+    expect_punct(at(&v, 2), .LParen);
+    expect_punct(at(&v, 3), .RParen);
+    expect_punct(at(&v, 4), .LBrace);
+    expect_endline(at(&v, 5));
+    expect_keyword(at(&v, 6), .Ret);
+    expect_pos(at(&v, 6), 2, 5);
+    expect_int(at(&v, 7), .Dec);
+    expect_punct(at(&v, 8), .Semicolon);
+    expect_endline(at(&v, 9));
+    expect_punct(at(&v, 10), .RBrace);
+    expect_endline(at(&v, 11));
+    expect_eof(at(&v, 12));
+}
+
+fn test_operators() {
+    var v = lex_ok("<-> <- -> <=> <<= << >>= ... .. . := == != < <= > >= && || & | ^ + - * / % ~ ! ? # @ $ , ; : ( ) [ ] { }");
+    defer { Vec.drop(&v); }
+    @assert(v.len == 43, "operator token count");
+
+    expect_punct(at(&v, 0), .ArrowBoth);
+    expect_text(at(&v, 0), "<->");
+    expect_punct(at(&v, 1), .LeftArrow);
+    expect_punct(at(&v, 2), .RightArrow);
+    expect_punct(at(&v, 3), .FatArrowBoth);
+    expect_punct(at(&v, 4), .LShAssign);
+    expect_punct(at(&v, 5), .LSh);
+    expect_text(at(&v, 5), "<<");
+    expect_punct(at(&v, 6), .RShAssign);
+    expect_punct(at(&v, 7), .Ellipsis);
+    expect_punct(at(&v, 8), .Range);
+    expect_punct(at(&v, 9), .Dot);
+    expect_punct(at(&v, 10), .ColonAssign);
+    expect_punct(at(&v, 11), .Eq);
+    expect_punct(at(&v, 12), .Neq);
+    expect_punct(at(&v, 13), .Lt);
+    expect_punct(at(&v, 14), .Le);
+    expect_punct(at(&v, 15), .Gt);
+    expect_punct(at(&v, 16), .Ge);
+    expect_punct(at(&v, 17), .AndAnd);
+    expect_punct(at(&v, 18), .OrOr);
+    expect_punct(at(&v, 19), .And);
+    expect_punct(at(&v, 20), .Or);
+    expect_punct(at(&v, 21), .Xor);
+    expect_punct(at(&v, 22), .Add);
+    expect_punct(at(&v, 23), .Sub);
+    expect_punct(at(&v, 24), .Mul);
+    expect_punct(at(&v, 25), .Div);
+    expect_punct(at(&v, 26), .Mod);
+    expect_punct(at(&v, 27), .BitNot);
+    expect_punct(at(&v, 28), .Not);
+    expect_punct(at(&v, 29), .Cond);
+    expect_punct(at(&v, 30), .Sharp);
+    expect_punct(at(&v, 31), .At);
+    expect_punct(at(&v, 32), .Dollar);
+    expect_punct(at(&v, 33), .Comma);
+    expect_punct(at(&v, 34), .Semicolon);
+    expect_punct(at(&v, 35), .Colon);
+    expect_punct(at(&v, 36), .LParen);
+    expect_punct(at(&v, 37), .RParen);
+    expect_punct(at(&v, 38), .LBracket);
+    expect_punct(at(&v, 39), .RBracket);
+    expect_punct(at(&v, 40), .LBrace);
+    expect_punct(at(&v, 41), .RBrace);
+    expect_eof(at(&v, 42));
+}
+
+fn test_comments_and_literals() {
+    var v = lex_ok("// c\n/* a /* nested */ b */\n'a' '\\n' '\\x41' \"hi\\n\" 0xFF 0b1010 0o17 1_000 3.14");
+    defer { Vec.drop(&v); }
+    @assert(v.len == 13, "literal token count");
+
+    var comment = at(&v, 0);
+    match comment.kind {
+        .LineComment => {},
+        _ => { @assert(false, "expected line comment"); },
+    }
+    expect_text(comment, "// c");
+    expect_pos(comment, 1, 1);
+    expect_endline(at(&v, 1));
+    expect_endline(at(&v, 2));
+
+    var ch3 = at(&v, 3);
+    expect_char(ch3, 'a');
+    @assert(ByteView.eq(ch3.text, view("'a'")), "char text");
+    expect_pos(ch3, 3, 1);
+    expect_char(at(&v, 4), '\n');
+    expect_char(at(&v, 5), 0x41);
+
+    var s = at(&v, 6);
+    match s.kind {
+        .Str => {},
+        _ => { @assert(false, "expected string"); },
+    }
+    expect_text(s, "\"hi\\n\"");
+
+    expect_int(at(&v, 7), .Hex);
+    expect_int(at(&v, 8), .Bin);
+    expect_int(at(&v, 9), .Oct);
+    expect_int(at(&v, 10), .Dec);
+
+    var fl = at(&v, 11);
+    match fl.kind {
+        .Float => {},
+        _ => { @assert(false, "expected float"); },
+    }
+    expect_eof(at(&v, 12));
+}
+
+// 非法输入走 diag：仍产出 token 以便继续收集错误，但 diag 必须记录。
+// Rust 风格数字字面量：分隔符下划线 + 多进制 + 小数/指数字面量。
+fn test_number_forms() {
+    var v = lex_ok("1_000_000 0xFF_FF 0x_ff 0b_1010 0o_7 1_000.5 1.5_0 1_000.500 1.5e-3 1_0.5_0E2");
+    defer { Vec.drop(&v); }
+    @assert(v.len == 11, "number token count");
+
+    expect_int(at(&v, 0), .Dec);
+    expect_text(at(&v, 0), "1_000_000");
+    expect_int(at(&v, 1), .Hex);
+    expect_text(at(&v, 1), "0xFF_FF");
+    expect_int(at(&v, 2), .Hex);
+    expect_text(at(&v, 2), "0x_ff");
+    expect_int(at(&v, 3), .Bin);
+    expect_text(at(&v, 3), "0b_1010");
+    expect_int(at(&v, 4), .Oct);
+    expect_text(at(&v, 4), "0o_7");
+
+    var i: usize = 5;
+    while i < 10 {
+        match at(&v, i).kind {
+            .Float => {},
+            _ => { @assert(false, "expected float"); },
+        }
+        i = i + 1;
+    }
+    expect_text(at(&v, 5), "1_000.5");
+    expect_text(at(&v, 6), "1.5_0");
+    expect_text(at(&v, 7), "1_000.500");
+    expect_text(at(&v, 8), "1.5e-3");
+    expect_text(at(&v, 9), "1_0.5_0E2");
+    expect_eof(at(&v, 10));
+}
+
+fn test_diag_errors() {
+    var diag = Diag.new(null);
+    defer { Diag.drop(&diag); }
+
+    match lex(view("'\\q' \"abc"), &diag, 0) {
+        .Ok(v) => { Vec.drop(&v); },
+        .Err(error) => { @assert(false, "lex should return tokens"); },
+    }
+    @assert(Diag.has_error(diag), "lex errors recorded");
+
+    var clean = Diag.new(null);
+    defer { Diag.drop(&clean); }
+    match lex(view("fn main() { ret 0; }"), &clean, 0) {
+        .Ok(v) => { Vec.drop(&v); },
+        .Err(error) => { @assert(false, "legal lex failed"); },
+    }
+    @assert(!Diag.has_error(clean), "legal input has no diagnostics");
+}
+
+fn main() {
+    test_basic();
+    test_operators();
+    test_comments_and_literals();
+    test_number_forms();
+    test_diag_errors();
+}

stage2/splc1.spl

@@ -0,0 +1,239 @@
+// Expect retcode = 0
+
+// splc1 (stage2 自举编译器入口)。
+// 当前仅实现 `--dump lex <file> [out]`：读取源文件，词法分析，按与 stage1
+// splc0 完全一致的格式打印 token 列表，供 diff 测试逐字节比对。
+
+type .{ .ByteView } = @import("./utils/byte_view.spl");
+type .{ .ByteBuffer } = @import("./utils/byte_buffer.spl");
+type .{ .Vec } = @import("./utils/vec.spl");
+type .{ .Token, .kind_name } = @import("./token.spl");
+type .{ .Lexer } = @import("./lexer.spl");
+type Ast = @import("./ast.spl");
+type Parser = @import("./parser.spl");
+type .{ .append_usize } = @import("./utils/fmt.spl");
+type .{ .read_all, .print, .eprint } = @import("./utils/io.spl");
+type .{ .set_new, .set_add, .set_drop } = @import("./utils/source.spl");
+type Diag = @import("./diag.spl");
+type Logger = @import("./utils/log.spl");
+
+fn token_at(tokens: *Vec, index: usize): Token {
+    match Vec.get(tokens, index) {
+        .Ok(ptr) => { ret (ptr as *Token).*; },
+        .Err(error) => { ret Token.make(.Unknown(0), ByteView.empty(), 0, 0, 0); },
+    }
+    ret Token.make(.Unknown(0), ByteView.empty(), 0, 0, 0);
+}
+
+// 追加成功返回 true；失败（仅 OOM）返回 false。
+fn push(out: *ByteBuffer, bytes: []u8): bool {
+    match ByteBuffer.append(out, bytes) {
+        .Ok => { ret true; },
+        .Err(error) => { ret false; },
+    }
+    ret false;
+}
+
+fn push_byte(out: *ByteBuffer, byte: u8): bool {
+    match ByteBuffer.append_byte(out, byte) {
+        .Ok => { ret true; },
+        .Err(error) => { ret false; },
+    }
+    ret false;
+}
+
+fn push_usize(out: *ByteBuffer, value: usize): bool {
+    match append_usize(out, value, 10, false) {
+        .Ok => { ret true; },
+        .Err(error) => { ret false; },
+    }
+    ret false;
+}
+
+// 输出格式与 splc0.c 的 dump_tokens 一致：
+//   tokens got (N)\n
+//   [TOK_TYPE] <lexeme> (line:col)\n
+fn dump_tokens(tokens: *Vec): bool {
+    var out = ByteBuffer.new();
+    defer { ByteBuffer.drop(&out); }
+
+    if !push(&out, "tokens got (") { ret false; }
+    if !push_usize(&out, tokens.len) { ret false; }
+    if !push(&out, ")\n") { ret false; }
+
+    var i: usize = 0;
+    while i < tokens.len {
+        var tok = token_at(tokens, i);
+        if !push(&out, "[") { ret false; }
+        if !push(&out, kind_name(tok.kind)) { ret false; }
+        if !push(&out, "] ") { ret false; }
+        if !push(&out, ByteView.as_slice(tok.text)) { ret false; }
+        if !push(&out, " (") { ret false; }
+        if !push_usize(&out, tok.line) { ret false; }
+        if !push_byte(&out, ':') { ret false; }
+        if !push_usize(&out, tok.col) { ret false; }
+        if !push(&out, ")\n") { ret false; }
+        i = i + 1;
+    }
+
+    match print(ByteBuffer.as_slice(out)) {
+        .Ok => { ret true; },
+        .Err(error) => { ret false; },
+    }
+    ret false;
+}
+
+fn dump_lex(log: *Logger, path: ByteView): i32 {
+    var content = ByteBuffer.new();
+    defer { ByteBuffer.drop(&content); }
+    match read_all(path, &content) {
+        .Ok => {},
+        .Err(error) => {
+            Logger.error(log, "cannot read input file");
+            ret 1;
+        },
+    }
+
+    var text = ByteView.from_slice(ByteBuffer.as_slice(content));
+
+    var sources = set_new();
+    defer { set_drop(&sources); }
+    match set_add(&sources, path, text) {
+        .Ok(index) => {},
+        .Err(error) => {
+            Logger.error(log, "cannot register source");
+            ret 1;
+        },
+    }
+
+    var diag = Diag.new(&sources);
+    defer { Diag.drop(&diag); }
+
+    var tokens = Vec.blank(@sizeof(Token), @alignof(Token));
+    match Lexer.lex(text, &diag, 0) {
+        .Ok(v) => { tokens = v; },
+        .Err(error) => {
+            Logger.error(log, "lex failed");
+            ret 1;
+        },
+    }
+    defer { Vec.drop(&tokens); }
+
+    // 非法输入：走诊断（stage1 在此处是宽容无视，stage2 显式报错）。
+    if Diag.has_error(diag) {
+        var rendered = ByteBuffer.new();
+        defer { ByteBuffer.drop(&rendered); }
+        match Diag.render_human(&diag, &rendered) {
+            .Ok => {},
+            .Err(error) => {},
+        }
+        eprint(ByteBuffer.as_slice(rendered));
+        ret 1;
+    }
+
+    if !dump_tokens(&tokens) {
+        Logger.fatal(log, "out of memory while dumping tokens");
+        ret 1;
+    }
+    ret 0;
+}
+
+fn dump_ast(log: *Logger, path: ByteView): i32 {
+    var content = ByteBuffer.new();
+    defer { ByteBuffer.drop(&content); }
+    match read_all(path, &content) {
+        .Ok => {},
+        .Err(error) => {
+            Logger.error(log, "cannot read input file");
+            ret 1;
+        },
+    }
+    var text = ByteView.from_slice(ByteBuffer.as_slice(content));
+
+    var sources = set_new();
+    defer { set_drop(&sources); }
+    match set_add(&sources, path, text) {
+        .Ok(index) => {},
+        .Err(error) => {
+            Logger.error(log, "cannot register source");
+            ret 1;
+        },
+    }
+    var diag = Diag.new(&sources);
+    defer { Diag.drop(&diag); }
+
+    var tokens = Vec.blank(@sizeof(Token), @alignof(Token));
+    match Lexer.lex(text, &diag, 0) {
+        .Ok(v) => { tokens = v; },
+        .Err(error) => {
+            Logger.error(log, "lex failed");
+            ret 1;
+        },
+    }
+    defer { Vec.drop(&tokens); }
+
+    var ast: Ast = undefined;
+    match Ast.new(&sources) {
+        .Ok(v) => { ast = v; },
+        .Err(error) => {
+            Logger.error(log, "ast init failed");
+            ret 1;
+        },
+    }
+    defer { Ast.drop(&ast); }
+
+    var p = Parser.new(&ast, tokens, &diag);
+    Parser.parse(&p);
+
+    if Diag.has_error(diag) || p.failed {
+        var rendered = ByteBuffer.new();
+        defer { ByteBuffer.drop(&rendered); }
+        match Diag.render_human(&diag, &rendered) {
+            .Ok => {},
+            .Err(error) => {},
+        }
+        eprint(ByteBuffer.as_slice(rendered));
+        ret 1;
+    }
+
+    var out = ByteBuffer.new();
+    defer { ByteBuffer.drop(&out); }
+    if !Ast.dump(&ast, &out) {
+        Logger.fatal(log, "out of memory while dumping ast");
+        ret 1;
+    }
+    match print(ByteBuffer.as_slice(out)) {
+        .Ok => {},
+        .Err(error) => { ret 1; },
+    }
+    ret 0;
+}
+
+fn main(argc: i32, argv: **u8): i32 {
+    // 无参数：自测运行（`nob test stage2`）视为成功。
+    if argc <= 1 { ret 0; }
+
+    var log = Logger.new(ByteView.from_slice("splc1"));
+
+    if argc < 4 {
+        Logger.fatal(&log, "usage: splc1 --dump <lex|ast> <file> [out]");
+        ret 1;
+    }
+
+    var arg1 = ByteView.from_cstr(argv[1]);
+    var mode = ByteView.from_cstr(argv[2]);
+    var path = ByteView.from_cstr(argv[3]);
+
+    if !ByteView.eq(arg1, ByteView.from_slice("--dump")) {
+        Logger.fatal(&log, "usage: splc1 --dump <lex|ast> <file> [out]");
+        ret 1;
+    }
+    if ByteView.eq(mode, ByteView.from_slice("lex")) {
+        ret dump_lex(&log, path);
+    }
+    if ByteView.eq(mode, ByteView.from_slice("ast")) {
+        ret dump_ast(&log, path);
+    }
+    Logger.fatal(&log, "dump mode not implemented");
+    ret 1;
+}

stage2/sys/file.spl

@@ -34,12 +34,12 @@ fn write_byte(f: *File, b: u8): u64 {
     ret write(f, &tmp, 1);
 }
 
-fn read(f: *File, buf: *u8, n: u64): u64 {
+fn read(f: *File, buf: *u8, n: usize): usize {
     if f.handle == null { ret 0; }
     ret vm_fread(buf as *any, 1, n, f.handle);
 }
 
-fn size(f: *File): u64 {
+fn size(f: *File): usize {
     if f.handle == null { ret 0; }
     ret vm_fsize(f.handle);
 }

stage2/token.spl

@@ -0,0 +1,399 @@
+// Expect retcode = 0
+
+// 词法单元：用带载荷 ADT 表达 TokenKind；关键词/算符各自独立 enum。
+// 整型与字符在词法期解码并随载荷携带；文本一律借用源切片（不复制）。
+
+type .{ .ByteView } = @import("utils/byte_view.spl");
+
+type Keyword = enum {
+    Any, As, Bool, Break, Catch, Comptime, Const, Continue, Defer, Else, Enum,
+    Errdefer, False, Fn, For, If, Loop, Match, Null, Ret, Shape, Struct, True,
+    Try, Type, Undefined, Union, Var, While, Underscore,
+}
+
+type KeywordResult = enum {
+    Yes: Keyword,
+    No,
+}
+
+type Punct = enum {
+    Eq, Assign, AddAssign, Add, SubAssign, Sub, MulAssign, Mul, DivAssign, Div,
+    ModAssign, Mod, RightArrow, LeftArrow, ArrowBoth, FatArrowBoth, RightFatArrow,
+    AndAnd, AndAssign, And, OrOr, OrAssign, Or, XorAssign, Xor,
+    LShAssign, LSh, Le, Lt, RShAssign, RSh, Ge, Gt, Not, Neq, BitNot,
+    LBracket, RBracket, LParen, RParen, LBrace, RBrace,
+    Semicolon, Comma, Colon, ColonAssign, Dot, Range, Ellipsis, Cond, Sharp, At, Dollar,
+}
+
+type Radix = enum {
+    Dec,
+    Bin,
+    Oct,
+    Hex,
+}
+
+type TokenKind = enum {
+    Keyword: Keyword,
+    Punct: Punct,
+    Ident,
+    Int: Radix, // 仅记进制；数值由 parser 按任意宽度（mp）解析
+    Float,
+    Char: u8,
+    Str,
+    LineComment,
+    Endline,
+    Eof,
+    Unknown: u8,
+}
+
+type Token = @self;
+
+kind: TokenKind,
+text: ByteView,
+line: usize,
+col: usize,
+offset: usize, // 源文件字节偏移（供 AST Span 使用）
+
+fn make(kind: TokenKind, text: ByteView, line: usize, col: usize, offset: usize): Token {
+    ret Token { .kind = kind, .text = text, .line = line, .col = col, .offset = offset };
+}
+
+fn is_trivia(kind: TokenKind): bool {
+    match kind {
+        .Endline => { ret true; },
+        .LineComment => { ret true; },
+        _ => { ret false; },
+    }
+}
+
+fn keyword_of(text: ByteView): KeywordResult {
+    if is(text, "any") { ret .Yes(.Any); }
+    if is(text, "as") { ret .Yes(.As); }
+    if is(text, "bool") { ret .Yes(.Bool); }
+    if is(text, "break") { ret .Yes(.Break); }
+    if is(text, "catch") { ret .Yes(.Catch); }
+    if is(text, "comptime") { ret .Yes(.Comptime); }
+    if is(text, "const") { ret .Yes(.Const); }
+    if is(text, "continue") { ret .Yes(.Continue); }
+    if is(text, "defer") { ret .Yes(.Defer); }
+    if is(text, "else") { ret .Yes(.Else); }
+    if is(text, "enum") { ret .Yes(.Enum); }
+    if is(text, "errdefer") { ret .Yes(.Errdefer); }
+    if is(text, "false") { ret .Yes(.False); }
+    if is(text, "fn") { ret .Yes(.Fn); }
+    if is(text, "for") { ret .Yes(.For); }
+    if is(text, "if") { ret .Yes(.If); }
+    if is(text, "loop") { ret .Yes(.Loop); }
+    if is(text, "match") { ret .Yes(.Match); }
+    if is(text, "null") { ret .Yes(.Null); }
+    if is(text, "ret") { ret .Yes(.Ret); }
+    if is(text, "shape") { ret .Yes(.Shape); }
+    if is(text, "struct") { ret .Yes(.Struct); }
+    if is(text, "true") { ret .Yes(.True); }
+    if is(text, "try") { ret .Yes(.Try); }
+    if is(text, "type") { ret .Yes(.Type); }
+    if is(text, "undefined") { ret .Yes(.Undefined); }
+    if is(text, "union") { ret .Yes(.Union); }
+    if is(text, "var") { ret .Yes(.Var); }
+    if is(text, "while") { ret .Yes(.While); }
+    if is(text, "_") { ret .Yes(.Underscore); }
+    ret .No;
+}
+
+// ---- token 类型名（与 stage1 spl_tok_type_name 逐字一致，供 dump/diff 使用）----
+
+fn keyword_name(kw: Keyword): []u8 {
+    match kw {
+        .Any => { ret "KW_ANY"; },
+        .As => { ret "KW_AS"; },
+        .Bool => { ret "KW_BOOL"; },
+        .Break => { ret "KW_BREAK"; },
+        .Catch => { ret "KW_CATCH"; },
+        .Comptime => { ret "KW_COMPTIME"; },
+        .Const => { ret "KW_CONST"; },
+        .Continue => { ret "KW_CONTINUE"; },
+        .Defer => { ret "KW_DEFER"; },
+        .Else => { ret "KW_ELSE"; },
+        .Enum => { ret "KW_ENUM"; },
+        .Errdefer => { ret "KW_ERRDEFER"; },
+        .False => { ret "KW_FALSE"; },
+        .Fn => { ret "KW_FN"; },
+        .For => { ret "KW_FOR"; },
+        .If => { ret "KW_IF"; },
+        .Loop => { ret "KW_LOOP"; },
+        .Match => { ret "KW_MATCH"; },
+        .Null => { ret "KW_NULL"; },
+        .Ret => { ret "KW_RET"; },
+        .Shape => { ret "KW_SHAPE"; },
+        .Struct => { ret "KW_STRUCT"; },
+        .True => { ret "KW_TRUE"; },
+        .Try => { ret "KW_TRY"; },
+        .Type => { ret "KW_TYPE"; },
+        .Undefined => { ret "KW_UNDEFINED"; },
+        .Union => { ret "KW_UNION"; },
+        .Var => { ret "KW_VAR"; },
+        .While => { ret "KW_WHILE"; },
+        .Underscore => { ret "KW_UNDERSCORE"; },
+    }
+    ret "?";
+}
+
+fn punct_name(punct: Punct): []u8 {
+    match punct {
+        .Eq => { ret "TOK_EQ"; },
+        .Assign => { ret "TOK_ASSIGN"; },
+        .AddAssign => { ret "TOK_ASSIGN_ADD"; },
+        .Add => { ret "TOK_ADD"; },
+        .SubAssign => { ret "TOK_ASSIGN_SUB"; },
+        .Sub => { ret "TOK_SUB"; },
+        .MulAssign => { ret "TOK_ASSIGN_MUL"; },
+        .Mul => { ret "TOK_MUL"; },
+        .DivAssign => { ret "TOK_ASSIGN_DIV"; },
+        .Div => { ret "TOK_DIV"; },
+        .ModAssign => { ret "TOK_ASSIGN_MOD"; },
+        .Mod => { ret "TOK_MOD"; },
+        .RightArrow => { ret "TOK_RIGHT_ARROW"; },
+        .LeftArrow => { ret "TOK_LEFT_ARROW"; },
+        .ArrowBoth => { ret "TOK_ARROW_BOTH"; },
+        .FatArrowBoth => { ret "TOK_FAT_ARROW_BOTH"; },
+        .RightFatArrow => { ret "TOK_RIGHT_FAT_ARROW"; },
+        .AndAnd => { ret "TOK_AND_AND"; },
+        .AndAssign => { ret "TOK_ASSIGN_AND"; },
+        .And => { ret "TOK_AND"; },
+        .OrOr => { ret "TOK_OR_OR"; },
+        .OrAssign => { ret "TOK_ASSIGN_OR"; },
+        .Or => { ret "TOK_OR"; },
+        .XorAssign => { ret "TOK_ASSIGN_XOR"; },
+        .Xor => { ret "TOK_XOR"; },
+        .LShAssign => { ret "TOK_ASSIGN_L_SH"; },
+        .LSh => { ret "TOK_L_SH"; },
+        .Le => { ret "TOK_LE"; },
+        .Lt => { ret "TOK_LT"; },
+        .RShAssign => { ret "TOK_ASSIGN_R_SH"; },
+        .RSh => { ret "TOK_R_SH"; },
+        .Ge => { ret "TOK_GE"; },
+        .Gt => { ret "TOK_GT"; },
+        .Not => { ret "TOK_NOT"; },
+        .Neq => { ret "TOK_NEQ"; },
+        .BitNot => { ret "TOK_BIT_NOT"; },
+        .LBracket => { ret "TOK_L_BRACKET"; },
+        .RBracket => { ret "TOK_R_BRACKET"; },
+        .LParen => { ret "TOK_L_PAREN"; },
+        .RParen => { ret "TOK_R_PAREN"; },
+        .LBrace => { ret "TOK_L_BRACE"; },
+        .RBrace => { ret "TOK_R_BRACE"; },
+        .Semicolon => { ret "TOK_SEMICOLON"; },
+        .Comma => { ret "TOK_COMMA"; },
+        .Colon => { ret "TOK_COLON"; },
+        .ColonAssign => { ret "TOK_COLON_ASSIGN"; },
+        .Dot => { ret "TOK_DOT"; },
+        .Range => { ret "TOK_RANGE"; },
+        .Ellipsis => { ret "TOK_ELLIPSIS"; },
+        .Cond => { ret "TOK_COND"; },
+        .Sharp => { ret "TOK_SHARP"; },
+        .At => { ret "TOK_AT"; },
+        .Dollar => { ret "TOK_DOLLAR"; },
+    }
+    ret "?";
+}
+
+fn kind_name(kind: TokenKind): []u8 {
+    match kind {
+        .Keyword(kw) => { ret keyword_name(kw); },
+        .Punct(punct) => { ret punct_name(punct); },
+        .Ident => { ret "TOK_IDENT"; },
+        .Int(value) => { ret "TOK_INT_LITERAL"; },
+        .Float => { ret "TOK_FLOAT_LITERAL"; },
+        .Char(value) => { ret "TOK_CHAR_LITERAL"; },
+        .Str => { ret "TOK_STRING_LITERAL"; },
+        .LineComment => { ret "TOK_LINE_COMMENT"; },
+        .Endline => { ret "TOK_ENDLINE"; },
+        .Eof => { ret "TOK_EOF"; },
+        .Unknown(value) => { ret "TOK_UNKNOWN"; },
+    }
+    ret "?";
+}
+
+// ---- 测试辅助 ----
+
+fn view(text: []u8): ByteView {
+    ret ByteView.from_slice(text);
+}
+
+fn is(text: ByteView, literal: []u8): bool {
+    ret ByteView.eq(text, view(literal));
+}
+
+fn expect_keyword(result: KeywordResult, expected: Keyword) {
+    match result {
+        .Yes(kw) => { @assert(@enum2int(kw) == @enum2int(expected), "keyword"); },
+        .No => { @assert(false, "expected keyword"); },
+    }
+}
+
+fn test_keywords() {
+    expect_keyword(keyword_of(view("fn")), .Fn);
+    expect_keyword(keyword_of(view("while")), .While);
+    expect_keyword(keyword_of(view("_")), .Underscore);
+    match keyword_of(view("foo")) {
+        .Yes(kw) => { @assert(false, "foo is not a keyword"); },
+        .No => {},
+    }
+    match keyword_of(view("")) {
+        .Yes(kw) => { @assert(false, "empty is not a keyword"); },
+        .No => {},
+    }
+}
+
+fn test_adt_payloads() {
+    var t = make(.Int(.Hex), view("0xFF"), 3, 5, 0);
+    @assert(t.line == 3 && t.col == 5, "position");
+    match t.kind {
+        .Int(radix) => {
+            var hex: Radix = .Hex;
+            @assert(@enum2int(radix) == @enum2int(hex), "int radix");
+        },
+        _ => { @assert(false, "expected int"); },
+    }
+
+    var c = make(.Char('a'), view("'a'"), 1, 1, 0);
+    match c.kind {
+        .Char(value) => { @assert(value == 'a', "char value"); },
+        _ => { @assert(false, "expected char"); },
+    }
+
+    var p = make(.Punct(.ArrowBoth), view("<->"), 1, 1, 0);
+    match p.kind {
+        .Punct(punct) => {
+            var both: Punct = .ArrowBoth;
+            @assert(@enum2int(punct) == @enum2int(both), "punct");
+        },
+        _ => { @assert(false, "expected punct"); },
+    }
+
+    var k = make(.Keyword(.Ret), view("ret"), 1, 1, 0);
+    match k.kind {
+        .Keyword(kw) => {
+            var ret_kw: Keyword = .Ret;
+            @assert(@enum2int(kw) == @enum2int(ret_kw), "keyword kind");
+        },
+        _ => { @assert(false, "expected keyword"); },
+    }
+}
+
+fn test_trivia() {
+    @assert(is_trivia(.Endline), "endline trivia");
+    @assert(is_trivia(.LineComment), "comment trivia");
+    @assert(!is_trivia(.Ident), "ident not trivia");
+    @assert(!is_trivia(.Eof), "eof not trivia");
+}
+
+fn expect_kind_name(kind: TokenKind, expected: []u8) {
+    @assert(ByteView.eq(ByteView.from_slice(kind_name(kind)),
+                        ByteView.from_slice(expected)),
+            "token kind name");
+}
+
+fn test_kind_names() {
+    expect_kind_name(.Keyword(.Any), "KW_ANY");
+    expect_kind_name(.Keyword(.As), "KW_AS");
+    expect_kind_name(.Keyword(.Bool), "KW_BOOL");
+    expect_kind_name(.Keyword(.Break), "KW_BREAK");
+    expect_kind_name(.Keyword(.Catch), "KW_CATCH");
+    expect_kind_name(.Keyword(.Comptime), "KW_COMPTIME");
+    expect_kind_name(.Keyword(.Const), "KW_CONST");
+    expect_kind_name(.Keyword(.Continue), "KW_CONTINUE");
+    expect_kind_name(.Keyword(.Defer), "KW_DEFER");
+    expect_kind_name(.Keyword(.Else), "KW_ELSE");
+    expect_kind_name(.Keyword(.Enum), "KW_ENUM");
+    expect_kind_name(.Keyword(.Errdefer), "KW_ERRDEFER");
+    expect_kind_name(.Keyword(.False), "KW_FALSE");
+    expect_kind_name(.Keyword(.Fn), "KW_FN");
+    expect_kind_name(.Keyword(.For), "KW_FOR");
+    expect_kind_name(.Keyword(.If), "KW_IF");
+    expect_kind_name(.Keyword(.Loop), "KW_LOOP");
+    expect_kind_name(.Keyword(.Match), "KW_MATCH");
+    expect_kind_name(.Keyword(.Null), "KW_NULL");
+    expect_kind_name(.Keyword(.Ret), "KW_RET");
+    expect_kind_name(.Keyword(.Shape), "KW_SHAPE");
+    expect_kind_name(.Keyword(.Struct), "KW_STRUCT");
+    expect_kind_name(.Keyword(.True), "KW_TRUE");
+    expect_kind_name(.Keyword(.Try), "KW_TRY");
+    expect_kind_name(.Keyword(.Type), "KW_TYPE");
+    expect_kind_name(.Keyword(.Undefined), "KW_UNDEFINED");
+    expect_kind_name(.Keyword(.Union), "KW_UNION");
+    expect_kind_name(.Keyword(.Var), "KW_VAR");
+    expect_kind_name(.Keyword(.While), "KW_WHILE");
+    expect_kind_name(.Keyword(.Underscore), "KW_UNDERSCORE");
+
+    expect_kind_name(.Punct(.Eq), "TOK_EQ");
+    expect_kind_name(.Punct(.Assign), "TOK_ASSIGN");
+    expect_kind_name(.Punct(.AddAssign), "TOK_ASSIGN_ADD");
+    expect_kind_name(.Punct(.Add), "TOK_ADD");
+    expect_kind_name(.Punct(.SubAssign), "TOK_ASSIGN_SUB");
+    expect_kind_name(.Punct(.Sub), "TOK_SUB");
+    expect_kind_name(.Punct(.MulAssign), "TOK_ASSIGN_MUL");
+    expect_kind_name(.Punct(.Mul), "TOK_MUL");
+    expect_kind_name(.Punct(.DivAssign), "TOK_ASSIGN_DIV");
+    expect_kind_name(.Punct(.Div), "TOK_DIV");
+    expect_kind_name(.Punct(.ModAssign), "TOK_ASSIGN_MOD");
+    expect_kind_name(.Punct(.Mod), "TOK_MOD");
+    expect_kind_name(.Punct(.RightArrow), "TOK_RIGHT_ARROW");
+    expect_kind_name(.Punct(.LeftArrow), "TOK_LEFT_ARROW");
+    expect_kind_name(.Punct(.ArrowBoth), "TOK_ARROW_BOTH");
+    expect_kind_name(.Punct(.FatArrowBoth), "TOK_FAT_ARROW_BOTH");
+    expect_kind_name(.Punct(.RightFatArrow), "TOK_RIGHT_FAT_ARROW");
+    expect_kind_name(.Punct(.AndAnd), "TOK_AND_AND");
+    expect_kind_name(.Punct(.AndAssign), "TOK_ASSIGN_AND");
+    expect_kind_name(.Punct(.And), "TOK_AND");
+    expect_kind_name(.Punct(.OrOr), "TOK_OR_OR");
+    expect_kind_name(.Punct(.OrAssign), "TOK_ASSIGN_OR");
+    expect_kind_name(.Punct(.Or), "TOK_OR");
+    expect_kind_name(.Punct(.XorAssign), "TOK_ASSIGN_XOR");
+    expect_kind_name(.Punct(.Xor), "TOK_XOR");
+    expect_kind_name(.Punct(.LShAssign), "TOK_ASSIGN_L_SH");
+    expect_kind_name(.Punct(.LSh), "TOK_L_SH");
+    expect_kind_name(.Punct(.Le), "TOK_LE");
+    expect_kind_name(.Punct(.Lt), "TOK_LT");
+    expect_kind_name(.Punct(.RShAssign), "TOK_ASSIGN_R_SH");
+    expect_kind_name(.Punct(.RSh), "TOK_R_SH");
+    expect_kind_name(.Punct(.Ge), "TOK_GE");
+    expect_kind_name(.Punct(.Gt), "TOK_GT");
+    expect_kind_name(.Punct(.Not), "TOK_NOT");
+    expect_kind_name(.Punct(.Neq), "TOK_NEQ");
+    expect_kind_name(.Punct(.BitNot), "TOK_BIT_NOT");
+    expect_kind_name(.Punct(.LBracket), "TOK_L_BRACKET");
+    expect_kind_name(.Punct(.RBracket), "TOK_R_BRACKET");
+    expect_kind_name(.Punct(.LParen), "TOK_L_PAREN");
+    expect_kind_name(.Punct(.RParen), "TOK_R_PAREN");
+    expect_kind_name(.Punct(.LBrace), "TOK_L_BRACE");
+    expect_kind_name(.Punct(.RBrace), "TOK_R_BRACE");
+    expect_kind_name(.Punct(.Semicolon), "TOK_SEMICOLON");
+    expect_kind_name(.Punct(.Comma), "TOK_COMMA");
+    expect_kind_name(.Punct(.Colon), "TOK_COLON");
+    expect_kind_name(.Punct(.ColonAssign), "TOK_COLON_ASSIGN");
+    expect_kind_name(.Punct(.Dot), "TOK_DOT");
+    expect_kind_name(.Punct(.Range), "TOK_RANGE");
+    expect_kind_name(.Punct(.Ellipsis), "TOK_ELLIPSIS");
+    expect_kind_name(.Punct(.Cond), "TOK_COND");
+    expect_kind_name(.Punct(.Sharp), "TOK_SHARP");
+    expect_kind_name(.Punct(.At), "TOK_AT");
+    expect_kind_name(.Punct(.Dollar), "TOK_DOLLAR");
+
+    expect_kind_name(.Ident, "TOK_IDENT");
+    expect_kind_name(.Int(.Dec), "TOK_INT_LITERAL");
+    expect_kind_name(.Float, "TOK_FLOAT_LITERAL");
+    expect_kind_name(.Char('a'), "TOK_CHAR_LITERAL");
+    expect_kind_name(.Str, "TOK_STRING_LITERAL");
+    expect_kind_name(.LineComment, "TOK_LINE_COMMENT");
+    expect_kind_name(.Endline, "TOK_ENDLINE");
+    expect_kind_name(.Eof, "TOK_EOF");
+    expect_kind_name(.Unknown(0), "TOK_UNKNOWN");
+}
+
+fn main() {
+    test_keywords();
+    test_adt_payloads();
+    test_trivia();
+    test_kind_names();
+}

stage2/utils/algorithm.spl

@@ -0,0 +1,326 @@
+// Expect retcode = 0
+
+// 类型擦除的搜索与稳定归并排序。元素由定长字节表示，比较回调负责解释元素和键；
+// context 只借用，模块不持有元素、键或上下文。所有尺寸错误都在写入前返回。
+type .{ .Error, .Status, .make } = @import("error.spl");
+type .{ .checked_mul } = @import("memory.spl");
+type .{ .alloc, .copy, .free } = @import("../sys/mem.spl");
+
+type SearchResult = enum {
+    Found: usize,
+    NotFound,
+    Err: Error,
+}
+
+fn validate_range(data: *u8, len: usize, elem_size: usize): Status {
+    if elem_size == 0 { ret .Err(make(.InvalidArgument)); }
+    match checked_mul(len, elem_size) {
+        .Ok(bytes) => {},
+        .Err(error) => { ret .Err(error); },
+    }
+    if len != 0 && data == null { ret .Err(make(.InvalidArgument)); }
+    ret .Ok;
+}
+
+// 返回首个相等元素。空范围不读取 data、key 或调用 compare。
+fn linear_search(data: *u8, len: usize, elem_size: usize, key: *u8,
+                 context: *u8, compare: fn(*u8, *u8, *u8): i32): SearchResult {
+    match validate_range(data, len, elem_size) {
+        .Ok => {},
+        .Err(error) => { ret .Err(error); },
+    }
+    if len == 0 { ret .NotFound; }
+    if key == null { ret .Err(make(.InvalidArgument)); }
+
+    var i: usize = 0;
+    while i < len {
+        if compare(data + i * elem_size, key, context) == 0 {
+            ret .Found(i);
+        }
+        i = i + 1;
+    }
+    ret .NotFound;
+}
+
+// data 必须按 compare 升序排列。存在重复值时返回首个相等元素。
+fn binary_search(data: *u8, len: usize, elem_size: usize, key: *u8,
+                 context: *u8, compare: fn(*u8, *u8, *u8): i32): SearchResult {
+    match validate_range(data, len, elem_size) {
+        .Ok => {},
+        .Err(error) => { ret .Err(error); },
+    }
+    if len == 0 { ret .NotFound; }
+    if key == null { ret .Err(make(.InvalidArgument)); }
+
+    var low: usize = 0;
+    var high = len;
+    while low < high {
+        var mid = low + (high - low) / 2;
+        if compare(data + mid * elem_size, key, context) < 0 {
+            low = mid + 1;
+        } else {
+            high = mid;
+        }
+    }
+    if low < len && compare(data + low * elem_size, key, context) == 0 {
+        ret .Found(low);
+    }
+    ret .NotFound;
+}
+
+fn min_size(a: usize, b: usize): usize {
+    if a < b { ret a; }
+    ret b;
+}
+
+fn merge_run(dst: *u8, src: *u8, elem_size: usize,
+             begin: usize, middle: usize, end: usize,
+             context: *u8, compare: fn(*u8, *u8, *u8): i32) {
+    var left = begin;
+    var right = middle;
+    var output = begin;
+    while left < middle && right < end {
+        var left_ptr = src + left * elem_size;
+        var right_ptr = src + right * elem_size;
+        // 相等时先取左侧元素，由此保证稳定性。
+        if compare(left_ptr, right_ptr, context) <= 0 {
+            copy((dst + output * elem_size) as *any, left_ptr as *any, elem_size);
+            left = left + 1;
+        } else {
+            copy((dst + output * elem_size) as *any, right_ptr as *any, elem_size);
+            right = right + 1;
+        }
+        output = output + 1;
+    }
+    while left < middle {
+        copy((dst + output * elem_size) as *any,
+             (src + left * elem_size) as *any, elem_size);
+        left = left + 1;
+        output = output + 1;
+    }
+    while right < end {
+        copy((dst + output * elem_size) as *any,
+             (src + right * elem_size) as *any, elem_size);
+        right = right + 1;
+        output = output + 1;
+    }
+}
+
+// 自底向上的稳定归并排序。分配失败及参数错误不改变输入。
+fn stable_sort(data: *u8, len: usize, elem_size: usize,
+               context: *u8, compare: fn(*u8, *u8, *u8): i32): Status {
+    var bytes: usize = 0;
+    match validate_range(data, len, elem_size) {
+        .Ok => {},
+        .Err(error) => { ret .Err(error); },
+    }
+    match checked_mul(len, elem_size) {
+        .Ok(value) => { bytes = value; },
+        .Err(error) => { ret .Err(error); },
+    }
+    if len < 2 { ret .Ok; }
+
+    var scratch = alloc(bytes) as *u8;
+    if scratch == null { ret .Err(make(.Overflow)); }
+    defer { free(scratch as *any); }
+
+    var source = data;
+    var destination = scratch;
+    var width: usize = 1;
+    while width < len {
+        var begin: usize = 0;
+        while begin < len {
+            var left_width = min_size(width, len - begin);
+            var middle = begin + left_width;
+            var remaining = len - middle;
+            var right_width = min_size(width, remaining);
+            var end = middle + right_width;
+            merge_run(destination, source, elem_size, begin, middle, end,
+                      context, compare);
+            begin = end;
+        }
+        var swap = source;
+        source = destination;
+        destination = swap;
+        if width > len / 2 {
+            width = len;
+        } else {
+            width = width * 2;
+        }
+    }
+
+    if source != data {
+        copy(data as *any, source as *any, bytes);
+    }
+    ret .Ok;
+}
+
+// ---- 测试 ----
+
+type Item = struct {
+    key: i32,
+    sequence: i32,
+}
+
+fn compare_i32(a: *u8, b: *u8, context: *u8): i32 {
+    var av = (a as *i32).*;
+    var bv = (b as *i32).*;
+    if av < bv { ret -1; }
+    if av > bv { ret 1; }
+    ret 0;
+}
+
+fn compare_item(a: *u8, b: *u8, context: *u8): i32 {
+    var av = (a as *Item).*.key;
+    var bv = (b as *Item).*.key;
+    var direction: i32 = 1;
+    if context != null { direction = (context as *i32).*; }
+    if av < bv { ret -direction; }
+    if av > bv { ret direction; }
+    ret 0;
+}
+
+fn expect_ok(result: Status) {
+    match result {
+        .Ok => {},
+        .Err(error) => { @assert(false, "unexpected algorithm error"); },
+    }
+}
+
+fn expect_found(result: SearchResult, expected: usize) {
+    match result {
+        .Found(index) => { @assert(index == expected, "search index"); },
+        .NotFound => { @assert(false, "expected search match"); },
+        .Err(error) => { @assert(false, "unexpected search error"); },
+    }
+}
+
+fn expect_missing(result: SearchResult) {
+    match result {
+        .Found(index) => { @assert(false, "unexpected search match"); },
+        .NotFound => {},
+        .Err(error) => { @assert(false, "unexpected search error"); },
+    }
+}
+
+fn expect_search_error(result: SearchResult, overflow: bool) {
+    match result {
+        .Found(index) => { @assert(false, "expected search error"); },
+        .NotFound => { @assert(false, "expected search error"); },
+        .Err(error) => {
+            match error.kind {
+                .Overflow => { @assert(overflow); },
+                .InvalidArgument => { @assert(!overflow); },
+                _ => { @assert(false, "unexpected search error kind"); },
+            }
+        },
+    }
+}
+
+fn test_search() {
+    var values: [8]i32 = .[1, 2, 2, 2, 5, 8, 13, 21];
+    var two: i32 = 2;
+    var seven: i32 = 7;
+    expect_found(linear_search(&values[0] as *u8, 8, @sizeof(i32),
+                               &two as *u8, null, compare_i32), 1);
+    expect_found(binary_search(&values[0] as *u8, 8, @sizeof(i32),
+                               &two as *u8, null, compare_i32), 1);
+    expect_missing(linear_search(&values[0] as *u8, 8, @sizeof(i32),
+                                 &seven as *u8, null, compare_i32));
+    expect_missing(binary_search(&values[0] as *u8, 8, @sizeof(i32),
+                                 &seven as *u8, null, compare_i32));
+
+    expect_missing(linear_search(null, 0, @sizeof(i32), null, null, compare_i32));
+    expect_missing(binary_search(null, 0, @sizeof(i32), null, null, compare_i32));
+    expect_search_error(linear_search(&values[0] as *u8, 8, 0,
+                                      &two as *u8, null, compare_i32), false);
+    expect_search_error(binary_search(null, 1, @sizeof(i32),
+                                      &two as *u8, null, compare_i32), false);
+    var huge = (~(0 as usize)) / @sizeof(i32) + 1;
+    expect_search_error(linear_search(&values[0] as *u8, huge, @sizeof(i32),
+                                      &two as *u8, null, compare_i32), true);
+}
+
+fn test_stable_sort() {
+    var items: [8]Item = .[
+        .{ .key = 3, .sequence = 0 },
+        .{ .key = 1, .sequence = 1 },
+        .{ .key = 2, .sequence = 2 },
+        .{ .key = 1, .sequence = 3 },
+        .{ .key = 3, .sequence = 4 },
+        .{ .key = 2, .sequence = 5 },
+        .{ .key = 1, .sequence = 6 },
+        .{ .key = 2, .sequence = 7 },
+    ];
+    var ascending: i32 = 1;
+    expect_ok(stable_sort(&items[0] as *u8, 8, @sizeof(Item),
+                          &ascending as *u8, compare_item));
+    var expected_keys: [8]i32 = .[1, 1, 1, 2, 2, 2, 3, 3];
+    var expected_sequence: [8]i32 = .[1, 3, 6, 2, 5, 7, 0, 4];
+    var i: usize = 0;
+    while i < 8 {
+        @assert(items[i].key == expected_keys[i], "sorted key");
+        @assert(items[i].sequence == expected_sequence[i], "stable equal order");
+        i = i + 1;
+    }
+
+    var descending: i32 = -1;
+    expect_ok(stable_sort(&items[0] as *u8, 8, @sizeof(Item),
+                          &descending as *u8, compare_item));
+    i = 1;
+    while i < 8 {
+        @assert(items[i - 1].key >= items[i].key, "context controls order");
+        if items[i - 1].key == items[i].key {
+            @assert(items[i - 1].sequence < items[i].sequence,
+                    "second stable sort preserves group order");
+        }
+        i = i + 1;
+    }
+
+    expect_ok(stable_sort(null, 0, @sizeof(Item), null, compare_item));
+    expect_ok(stable_sort(&items[0] as *u8, 1, @sizeof(Item), null, compare_item));
+
+    // 非二次幂长度覆盖末尾不足一个完整归并段的路径。
+    var odd: [9]i32 = .[9, -1, 4, 4, 0, 8, 2, 7, 3];
+    var odd_expected: [9]i32 = .[-1, 0, 2, 3, 4, 4, 7, 8, 9];
+    expect_ok(stable_sort(&odd[0] as *u8, 9, @sizeof(i32), null, compare_i32));
+    i = 0;
+    while i < 9 {
+        @assert(odd[i] == odd_expected[i], "non-power-of-two length");
+        i = i + 1;
+    }
+}
+
+fn test_sort_failures_preserve_input() {
+    var values: [3]i32 = .[3, 1, 2];
+    match stable_sort(&values[0] as *u8, 3, 0, null, compare_i32) {
+        .Ok => { @assert(false, "zero element size"); },
+        .Err(error) => {
+            match error.kind {
+                .InvalidArgument => {},
+                _ => { @assert(false); },
+            }
+        },
+    }
+    @assert(values[0] == 3 && values[1] == 1 && values[2] == 2,
+            "validation failure preserves input");
+
+    var huge = (~(0 as usize)) / @sizeof(i32) + 1;
+    match stable_sort(&values[0] as *u8, huge, @sizeof(i32), null, compare_i32) {
+        .Ok => { @assert(false, "size overflow"); },
+        .Err(error) => {
+            match error.kind {
+                .Overflow => {},
+                _ => { @assert(false); },
+            }
+        },
+    }
+    @assert(values[0] == 3 && values[1] == 1 && values[2] == 2,
+            "overflow preserves input");
+}
+
+fn main() {
+    test_search();
+    test_stable_sort();
+    test_sort_failures_preserve_input();
+}

stage2/utils/arena.spl

@@ -1,9 +1,13 @@
 // 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。
+// 分配失败通过 ArenaResult 区分原因：size==0/非法对齐/内存不足/地址溢出。
+// 失败的 alloc 不改变已有分配。
 
 type .{ .Vec } = @import("vec.spl");
+type .{ .Error, .Kind, .Status } = @import("error.spl");
+type .{ .checked_add, .align_up } = @import("memory.spl");
 type Mem = @import("../sys/mem.spl");
 
 type Chunk = struct {
@@ -12,6 +16,11 @@ type Chunk = struct {
     used: usize,
 }
 
+type ArenaResult = enum {
+    Ok: *u8,
+    Err: Error,
+}
+
 type Arena = @self;
 
 chunk_size: usize,
@@ -28,61 +37,68 @@ fn new(chunk_size: usize): Arena {
     };
 }
 
-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; },
+        // 越界/无效 Arena 统一返回 null；调用方负责判空，无资源需要释放。
         .Err(error) => { ret null; },
     }
     ret null;
 }
 
-fn push_chunk(self: *Arena, size: usize): bool {
+fn push_chunk(self: *Arena, size: usize): Status {
     var base = Mem.alloc(size) as *u8;
-    if base == null { ret false; }
+    if base == null { ret .Err(Error.make(.Overflow)); }
     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; },
+        .Err(error) => { Mem.free(base as *any); ret .Err(error); },
     }
     self.current = self.chunks.len - 1;
-    ret true;
+    ret .Ok;
 }
 
-// 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; }
+// 分配 size 字节，地址按 align（非零二次幂）对齐。失败返回 ArenaResult.Err。
+fn alloc(self: *Arena, size: usize, align: usize): ArenaResult {
+    if size == 0 { ret .Err(Error.make(.InvalidArgument)); }
     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;
+            var need: usize = 0;
+            match checked_add(size, align) {
+                .Ok(value) => { need = value; },
+                .Err(error) => { ret .Err(error); },
+            }
             if need > want { want = need; }
-            if !push_chunk(self, want) { ret null; }
+            match push_chunk(self, want) {
+                .Ok => {},
+                .Err(error) => { ret .Err(error); },
+            }
         }
 
         var c = chunk_at(self, self.current);
-        if c == null { ret null; }
+        if c == null { ret .Err(Error.make(.InvalidArgument)); }
         var addr = c.base as usize;
-        var aligned = align_up(addr + c.used, align);
+        var used_addr: usize = 0;
+        match checked_add(addr, c.used) {
+            .Ok(value) => { used_addr = value; },
+            .Err(error) => { ret .Err(error); },
+        }
+        var aligned: usize = 0;
+        match align_up(used_addr, align) {
+            .Ok(value) => { aligned = value; },
+            .Err(error) => { ret .Err(error); }, // 非二次幂对齐或地址溢出
+        }
         var offset = aligned - addr;
         if offset <= c.size && size <= c.size - offset {
             c.used = offset + size;
-            ret aligned as *u8;
+            ret .Ok(aligned as *u8);
         }
         self.current = self.current + 1;
     }
-    ret null;
+    ret .Err(Error.make(.OutOfRange)); // 不可达：循环内所有路径均已返回
 }
 
 fn chunk_count(self: Arena): usize {
@@ -117,13 +133,31 @@ fn is_aligned(p: *u8, align: usize): bool {
     ret ((p as usize) & (align - 1)) == 0;
 }
 
+fn alloc_ok(a: *Arena, size: usize, align: usize): *u8 {
+    match alloc(a, size, align) {
+        .Ok(ptr) => { ret ptr; },
+        .Err(error) => { @assert(false, "unexpected arena alloc error"); },
+    }
+    ret null;
+}
+
+fn expect_alloc_error(result: ArenaResult, expected: Kind) {
+    match result {
+        .Ok(ptr) => { @assert(false, "expected arena alloc error"); },
+        .Err(error) => {
+            var want: Kind = expected;
+            @assert(@enum2int(error.kind) == @enum2int(want), "arena alloc error kind");
+        },
+    }
+}
+
 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);
+    var p1 = alloc_ok(&a, 1, 1);
+    var p8 = alloc_ok(&a, 16, 8);
+    var p16 = alloc_ok(&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");
@@ -136,18 +170,20 @@ fn test_alloc_and_alignment() {
     @assert(chunk_count(a) == 1, "one chunk");
     reset(&a);
     @assert(chunk_count(a) == 1, "reset keeps chunks");
-    var q = alloc(&a, 1, 1);
+    var q = alloc_ok(&a, 1, 1);
     @assert(q == p1, "reset reuses first chunk");
 
+    // drop 后还有断言，故保持手动释放，不改为 defer。
     drop(&a);
     @assert(chunk_count(a) == 0, "drop empties");
 }
 
 fn test_growth() {
     var a = new(64);
+    defer { drop(&a); }
     var i: i32 = 0;
     while i < 100 {
-        var p = alloc(&a, 16, 8);
+        var p = alloc_ok(&a, 16, 8);
         @assert(p != null, "grow alloc");
         @assert(is_aligned(p, 8), "grow align");
         p[0] = i as u8;
@@ -156,18 +192,17 @@ fn test_growth() {
     @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);
+    var big = alloc_ok(&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);
+    defer { drop(&a); }
+    expect_alloc_error(alloc(&a, 0, 1), .InvalidArgument);
+    var d = alloc_ok(&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);
+    expect_alloc_error(alloc(&a, 8, 3), .InvalidAlignment);
 }
 
 fn main() {

stage2/utils/ascii.spl

@@ -1,4 +1,5 @@
 // Expect retcode = 0
+
 // ASCII 字符工具。输入和输出都是单个字节，不解释 UTF-8 或本地化规则。
 
 type .{ .Error, .make } = @import("error.spl");

stage2/utils/big_int.spl

@@ -1,871 +0,0 @@
-// Expect retcode = 0
-// 任意精度有符号整数。limbs 为小端 u32；零规范为 len=0 且 negative=false。
-// 普通赋值只借用同一存储；深复制使用 clone，所有权转移使用 take。
-type .{ .Error, .Status } = @import("error.spl");
-type .{ .checked_mul } = @import("memory.spl");
-type .{ .digit_char, .digit_value } = @import("ascii.spl");
-type .{ .ByteBuffer } = @import("byte_buffer.spl");
-type .{ .alloc, .copy, .free } = @import("../sys/mem.spl");
-
-type BigInt = @self;
-
-limbs: *u32,
-len: usize,
-cap: usize,
-negative: bool,
-
-fn zero(): BigInt {
-    ret BigInt { .limbs = null, .len = 0, .cap = 0, .negative = false };
-}
-
-fn valid(self: BigInt): bool {
-    if self.len > self.cap { ret false; }
-    if self.cap == 0 {
-        ret self.limbs == null && self.len == 0 && !self.negative;
-    }
-    if self.limbs == null { ret false; }
-    if self.len == 0 { ret !self.negative; }
-    ret self.limbs[self.len - 1] != 0;
-}
-
-fn is_zero(self: BigInt): bool {
-    ret self.len == 0;
-}
-
-fn is_negative(self: BigInt): bool {
-    ret self.negative;
-}
-
-fn normalize(self: *BigInt) {
-    while self.len > 0 {
-        if self.limbs[self.len - 1] != 0 { break; }
-        self.len = self.len - 1;
-    }
-    if self.len == 0 { self.negative = false; }
-}
-
-fn reserve(self: *BigInt, need: usize): Status {
-    if !valid(self.*) { ret .Err(Error.make(.InvalidArgument)); }
-    if need <= self.cap { ret .Ok; }
-
-    var new_cap = self.cap;
-    if new_cap < 2 { new_cap = 2; }
-    while new_cap < need {
-        if new_cap > (~(0 as usize)) / 2 {
-            new_cap = need;
-            break;
-        }
-        new_cap = new_cap * 2;
-    }
-
-    var byte_count: usize = 0;
-    match checked_mul(new_cap, @sizeof(u32)) {
-        .Ok(value) => { byte_count = value; },
-        .Err(error) => { ret .Err(error); },
-    }
-    var fresh = alloc(byte_count) as *u32;
-    if self.len != 0 {
-        copy(fresh as *any, self.limbs as *any, self.len * @sizeof(u32));
-    }
-    free(self.limbs as *any);
-    self.limbs = fresh;
-    self.cap = new_cap;
-    ret .Ok;
-}
-
-fn drop(self: *BigInt) {
-    free(self.limbs as *any);
-    self.* = zero();
-}
-
-fn take(source: *BigInt): BigInt {
-    var result = source.*;
-    source.* = zero();
-    ret result;
-}
-
-fn replace(out: *BigInt, fresh: *BigInt) {
-    drop(out);
-    out.* = take(fresh);
-}
-
-fn clone(out: *BigInt, source: BigInt): Status {
-    if !valid(out.*) || !valid(source) { ret .Err(Error.make(.InvalidArgument)); }
-    var fresh = zero();
-    match reserve(&fresh, source.len) {
-        .Ok => {},
-        .Err(error) => { ret .Err(error); },
-    }
-    if source.len != 0 {
-        copy(fresh.limbs as *any, source.limbs as *any, source.len * @sizeof(u32));
-    }
-    fresh.len = source.len;
-    fresh.negative = source.negative;
-    replace(out, &fresh);
-    ret .Ok;
-}
-
-fn from_u64(out: *BigInt, value: u64): Status {
-    if !valid(out.*) { ret .Err(Error.make(.InvalidArgument)); }
-    var fresh = zero();
-    if value != 0 {
-        match reserve(&fresh, 2) {
-            .Ok => {},
-            .Err(error) => { ret .Err(error); },
-        }
-        fresh.limbs[0] = value as u32;
-        fresh.limbs[1] = (value >> 32) as u32;
-        fresh.len = 1;
-        if fresh.limbs[1] != 0 { fresh.len = 2; }
-    }
-    replace(out, &fresh);
-    ret .Ok;
-}
-
-fn from_i64(out: *BigInt, value: i64): Status {
-    var magnitude = value as u64;
-    var negative = value < 0;
-    if negative { magnitude = 0 - magnitude; }
-    match from_u64(out, magnitude) {
-        .Ok => {},
-        .Err(error) => { ret .Err(error); },
-    }
-    if out.len != 0 { out.negative = negative; }
-    ret .Ok;
-}
-
-fn cmp_abs(a: BigInt, b: BigInt): i32 {
-    if a.len < b.len { ret -1; }
-    if a.len > b.len { ret 1; }
-    var i = a.len;
-    while i > 0 {
-        i = i - 1;
-        if a.limbs[i] < b.limbs[i] { ret -1; }
-        if a.limbs[i] > b.limbs[i] { ret 1; }
-    }
-    ret 0;
-}
-
-fn cmp(a: BigInt, b: BigInt): i32 {
-    if a.negative != b.negative {
-        if a.negative { ret -1; }
-        ret 1;
-    }
-    var order = cmp_abs(a, b);
-    if a.negative { ret 0 - order; }
-    ret order;
-}
-
-fn eq(a: BigInt, b: BigInt): bool {
-    if a.negative != b.negative || a.len != b.len { ret false; }
-    var i: usize = 0;
-    while i < a.len {
-        if a.limbs[i] != b.limbs[i] { ret false; }
-        i = i + 1;
-    }
-    ret true;
-}
-
-fn add_abs_into(out: *BigInt, a: BigInt, b: BigInt): Status {
-    var count = a.len;
-    if b.len > count { count = b.len; }
-    if count == ~(0 as usize) { ret .Err(Error.make(.Overflow)); }
-    match reserve(out, count + 1) {
-        .Ok => {},
-        .Err(error) => { ret .Err(error); },
-    }
-    var carry: u64 = 0;
-    var i: usize = 0;
-    while i < count {
-        var av: u64 = 0;
-        var bv: u64 = 0;
-        if i < a.len { av = a.limbs[i] as u64; }
-        if i < b.len { bv = b.limbs[i] as u64; }
-        var sum = av + bv + carry;
-        out.limbs[i] = sum as u32;
-        carry = sum >> 32;
-        i = i + 1;
-    }
-    if carry != 0 {
-        out.limbs[count] = carry as u32;
-        count = count + 1;
-    }
-    out.len = count;
-    out.negative = false;
-    normalize(out);
-    ret .Ok;
-}
-
-// 要求 |a| >= |b|。
-fn sub_abs_into(out: *BigInt, a: BigInt, b: BigInt): Status {
-    match reserve(out, a.len) {
-        .Ok => {},
-        .Err(error) => { ret .Err(error); },
-    }
-    var borrow: u64 = 0;
-    var i: usize = 0;
-    while i < a.len {
-        var av = a.limbs[i] as u64;
-        var bv: u64 = 0;
-        if i < b.len { bv = b.limbs[i] as u64; }
-        var subtrahend = bv + borrow;
-        if av < subtrahend {
-            out.limbs[i] = (4294967296 + av - subtrahend) as u32;
-            borrow = 1;
-        } else {
-            out.limbs[i] = (av - subtrahend) as u32;
-            borrow = 0;
-        }
-        i = i + 1;
-    }
-    out.len = a.len;
-    out.negative = false;
-    normalize(out);
-    ret .Ok;
-}
-
-fn add(out: *BigInt, a: BigInt, b: BigInt): Status {
-    if !valid(out.*) || !valid(a) || !valid(b) {
-        ret .Err(Error.make(.InvalidArgument));
-    }
-    var fresh = zero();
-    if a.negative == b.negative {
-        match add_abs_into(&fresh, a, b) {
-            .Ok => {}, .Err(error) => { drop(&fresh); ret .Err(error); },
-        }
-        if fresh.len != 0 { fresh.negative = a.negative; }
-    } else {
-        var order = cmp_abs(a, b);
-        if order >= 0 {
-            match sub_abs_into(&fresh, a, b) {
-                .Ok => {}, .Err(error) => { drop(&fresh); ret .Err(error); },
-            }
-            if fresh.len != 0 { fresh.negative = a.negative; }
-        } else {
-            match sub_abs_into(&fresh, b, a) {
-                .Ok => {}, .Err(error) => { drop(&fresh); ret .Err(error); },
-            }
-            if fresh.len != 0 { fresh.negative = b.negative; }
-        }
-    }
-    replace(out, &fresh);
-    ret .Ok;
-}
-
-fn sub(out: *BigInt, a: BigInt, b: BigInt): Status {
-    var neg_b = b;
-    if neg_b.len != 0 { neg_b.negative = !neg_b.negative; }
-    ret add(out, a, neg_b);
-}
-
-fn neg(out: *BigInt, source: BigInt): Status {
-    match clone(out, source) {
-        .Ok => {}, .Err(error) => { ret .Err(error); },
-    }
-    if out.len != 0 { out.negative = !out.negative; }
-    ret .Ok;
-}
-
-fn abs(out: *BigInt, source: BigInt): Status {
-    match clone(out, source) {
-        .Ok => {}, .Err(error) => { ret .Err(error); },
-    }
-    out.negative = false;
-    ret .Ok;
-}
-
-fn mul(out: *BigInt, a: BigInt, b: BigInt): Status {
-    if !valid(out.*) || !valid(a) || !valid(b) {
-        ret .Err(Error.make(.InvalidArgument));
-    }
-    var fresh = zero();
-    if a.len == 0 || b.len == 0 {
-        replace(out, &fresh);
-        ret .Ok;
-    }
-    if b.len > (~(0 as usize)) - a.len {
-        ret .Err(Error.make(.Overflow));
-    }
-    var count = a.len + b.len;
-    match reserve(&fresh, count) {
-        .Ok => {}, .Err(error) => { ret .Err(error); },
-    }
-    var z: usize = 0;
-    while z < count { fresh.limbs[z] = 0; z = z + 1; }
-
-    var i: usize = 0;
-    while i < a.len {
-        var carry: u64 = 0;
-        var j: usize = 0;
-        while j < b.len {
-            var at = i + j;
-            var product = (a.limbs[i] as u64) * (b.limbs[j] as u64) +
-                          (fresh.limbs[at] as u64) + carry;
-            fresh.limbs[at] = product as u32;
-            carry = product >> 32;
-            j = j + 1;
-        }
-        fresh.limbs[i + b.len] = carry as u32;
-        i = i + 1;
-    }
-    fresh.len = count;
-    fresh.negative = a.negative != b.negative;
-    normalize(&fresh);
-    replace(out, &fresh);
-    ret .Ok;
-}
-
-fn bit_length(out: *usize, self: BigInt): Status {
-    if out == null || !valid(self) { ret .Err(Error.make(.InvalidArgument)); }
-    if self.len == 0 { out.* = 0; ret .Ok; }
-    if self.len - 1 > (~(0 as usize)) / 32 { ret .Err(Error.make(.Overflow)); }
-    var top = self.limbs[self.len - 1];
-    var bits = (self.len - 1) * 32;
-    while top != 0 {
-        bits = bits + 1;
-        top = top >> 1;
-    }
-    out.* = bits;
-    ret .Ok;
-}
-
-fn mul_pow2(out: *BigInt, source: BigInt, amount: usize): Status {
-    if !valid(out.*) || !valid(source) { ret .Err(Error.make(.InvalidArgument)); }
-    if source.len == 0 { ret from_u64(out, 0); }
-    var words = amount / 32;
-    var shift = amount % 32;
-    if source.len == ~(0 as usize) || words > (~(0 as usize)) - source.len - 1 {
-        ret .Err(Error.make(.Overflow));
-    }
-    var fresh = zero();
-    var count = source.len + words + 1;
-    match reserve(&fresh, count) {
-        .Ok => {}, .Err(error) => { ret .Err(error); },
-    }
-    var i: usize = 0;
-    while i < words { fresh.limbs[i] = 0; i = i + 1; }
-    var carry: u64 = 0;
-    i = 0;
-    while i < source.len {
-        var value = ((source.limbs[i] as u64) << shift as u32) | carry;
-        fresh.limbs[i + words] = value as u32;
-        carry = value >> 32;
-        i = i + 1;
-    }
-    fresh.len = source.len + words;
-    if carry != 0 {
-        fresh.limbs[fresh.len] = carry as u32;
-        fresh.len = fresh.len + 1;
-    }
-    fresh.negative = source.negative;
-    replace(out, &fresh);
-    ret .Ok;
-}
-
-// 按绝对值右移，等价于向零除以 2^amount。
-fn div_pow2(out: *BigInt, source: BigInt, amount: usize): Status {
-    if !valid(out.*) || !valid(source) { ret .Err(Error.make(.InvalidArgument)); }
-    var words = amount / 32;
-    var shift = amount % 32;
-    if words >= source.len { ret from_u64(out, 0); }
-    var count = source.len - words;
-    var fresh = zero();
-    match reserve(&fresh, count) {
-        .Ok => {}, .Err(error) => { ret .Err(error); },
-    }
-    var i: usize = 0;
-    while i < count {
-        var src = i + words;
-        var value = (source.limbs[src] as u64) >> shift as u32;
-        if shift != 0 && src + 1 < source.len {
-            value = value | ((source.limbs[src + 1] << (32 - shift) as u32) as u64);
-        }
-        fresh.limbs[i] = value as u32;
-        i = i + 1;
-    }
-    fresh.len = count;
-    fresh.negative = source.negative;
-    normalize(&fresh);
-    replace(out, &fresh);
-    ret .Ok;
-}
-
-fn shift_one(self: *BigInt) {
-    var carry: u64 = 0;
-    var i: usize = 0;
-    while i < self.len {
-        var value = ((self.limbs[i] as u64) << 1) | carry;
-        self.limbs[i] = value as u32;
-        carry = value >> 32;
-        i = i + 1;
-    }
-    if carry != 0 {
-        self.limbs[self.len] = carry as u32;
-        self.len = self.len + 1;
-    }
-}
-
-fn sub_abs_in_place(a: *BigInt, b: BigInt) {
-    var borrow: u64 = 0;
-    var i: usize = 0;
-    while i < a.len {
-        var av = a.limbs[i] as u64;
-        var bv: u64 = 0;
-        if i < b.len { bv = b.limbs[i] as u64; }
-        var subtrahend = bv + borrow;
-        if av < subtrahend {
-            a.limbs[i] = (4294967296 + av - subtrahend) as u32;
-            borrow = 1;
-        } else {
-            a.limbs[i] = (av - subtrahend) as u32;
-            borrow = 0;
-        }
-        i = i + 1;
-    }
-    normalize(a);
-}
-
-fn div_rem(quotient: *BigInt, remainder: *BigInt, dividend: BigInt,
-           divisor: BigInt): Status {
-    if (quotient as usize) == (remainder as usize) {
-        ret .Err(Error.make(.InvalidArgument));
-    }
-    if !valid(quotient.*) || !valid(remainder.*) || !valid(dividend) || !valid(divisor) {
-        ret .Err(Error.make(.InvalidArgument));
-    }
-    if divisor.len == 0 { ret .Err(Error.make(.DivisionByZero)); }
-
-    var q = zero();
-    var r = zero();
-    match reserve(&q, dividend.len) {
-        .Ok => {}, .Err(error) => { ret .Err(error); },
-    }
-    if divisor.len == ~(0 as usize) { drop(&q); ret .Err(Error.make(.Overflow)); }
-    match reserve(&r, divisor.len + 1) {
-        .Ok => {}, .Err(error) => { drop(&q); ret .Err(error); },
-    }
-    var i: usize = 0;
-    while i < dividend.len { q.limbs[i] = 0; i = i + 1; }
-    q.len = dividend.len;
-
-    var bit: usize = 0;
-    match bit_length(&bit, dividend) {
-        .Ok => {},
-        .Err(error) => { drop(&q); drop(&r); ret .Err(error); },
-    }
-    while bit > 0 {
-        bit = bit - 1;
-        shift_one(&r);
-        var word = bit / 32;
-        var offset = bit % 32;
-        var incoming = (dividend.limbs[word] >> offset as u32) & 1;
-        if incoming != 0 {
-            if r.len == 0 { r.len = 1; r.limbs[0] = 0; }
-            r.limbs[0] = r.limbs[0] | 1;
-        }
-        if cmp_abs(r, divisor) >= 0 {
-            sub_abs_in_place(&r, divisor);
-            q.limbs[word] = q.limbs[word] | ((1 as u32) << offset as u32);
-        }
-    }
-    q.negative = q.len != 0 && dividend.negative != divisor.negative;
-    r.negative = r.len != 0 && dividend.negative;
-    normalize(&q);
-    normalize(&r);
-    replace(quotient, &q);
-    replace(remainder, &r);
-    ret .Ok;
-}
-
-fn mul_small(self: *BigInt, value: u32): Status {
-    if self.len == 0 || value == 1 { ret .Ok; }
-    if value == 0 { self.len = 0; self.negative = false; ret .Ok; }
-    if self.len == ~(0 as usize) { ret .Err(Error.make(.Overflow)); }
-    match reserve(self, self.len + 1) {
-        .Ok => {}, .Err(error) => { ret .Err(error); },
-    }
-    var carry: u64 = 0;
-    var i: usize = 0;
-    while i < self.len {
-        var product = (self.limbs[i] as u64) * (value as u64) + carry;
-        self.limbs[i] = product as u32;
-        carry = product >> 32;
-        i = i + 1;
-    }
-    if carry != 0 {
-        self.limbs[self.len] = carry as u32;
-        self.len = self.len + 1;
-    }
-    ret .Ok;
-}
-
-fn add_small(self: *BigInt, value: u32): Status {
-    if value == 0 { ret .Ok; }
-    if self.len == ~(0 as usize) { ret .Err(Error.make(.Overflow)); }
-    match reserve(self, self.len + 1) {
-        .Ok => {}, .Err(error) => { ret .Err(error); },
-    }
-    var carry = value as u64;
-    var i: usize = 0;
-    while carry != 0 && i < self.len {
-        var sum = (self.limbs[i] as u64) + carry;
-        self.limbs[i] = sum as u32;
-        carry = sum >> 32;
-        i = i + 1;
-    }
-    if carry != 0 {
-        self.limbs[self.len] = carry as u32;
-        self.len = self.len + 1;
-    }
-    ret .Ok;
-}
-
-fn parse(out: *BigInt, bytes: []u8, radix: u8): Status {
-    if !valid(out.*) { ret .Err(Error.make(.InvalidArgument)); }
-    if radix < 2 || radix > 36 { ret .Err(Error.make(.InvalidArgument)); }
-    if bytes.len == 0 { ret .Err(Error.at(.UnexpectedEnd, 0)); }
-    if bytes.ptr == null { ret .Err(Error.make(.InvalidArgument)); }
-
-    var i: usize = 0;
-    var negative = false;
-    if bytes.ptr[0] == '-' || bytes.ptr[0] == '+' {
-        negative = bytes.ptr[0] == '-';
-        i = 1;
-    }
-    if i == bytes.len { ret .Err(Error.at(.UnexpectedEnd, i)); }
-
-    var fresh = zero();
-    while i < bytes.len {
-        var digit: u8 = 0;
-        match digit_value(bytes.ptr[i], radix) {
-            .Ok(value) => { digit = value; },
-            .Err(error) => { drop(&fresh); ret .Err(Error.at(.InvalidDigit, i)); },
-        }
-        match mul_small(&fresh, radix as u32) {
-            .Ok => {}, .Err(error) => { drop(&fresh); ret .Err(error); },
-        }
-        match add_small(&fresh, digit as u32) {
-            .Ok => {}, .Err(error) => { drop(&fresh); ret .Err(error); },
-        }
-        i = i + 1;
-    }
-    fresh.negative = fresh.len != 0 && negative;
-    replace(out, &fresh);
-    ret .Ok;
-}
-
-fn div_small(self: *BigInt, radix: u32): u32 {
-    var remainder: u64 = 0;
-    var i = self.len;
-    while i > 0 {
-        i = i - 1;
-        var current = (remainder << 32) | (self.limbs[i] as u64);
-        self.limbs[i] = (current / radix as u64) as u32;
-        remainder = current % radix as u64;
-    }
-    normalize(self);
-    ret remainder as u32;
-}
-
-fn format(out: *ByteBuffer, source: BigInt, radix: u8, uppercase: bool): Status {
-    if !valid(source) { ret .Err(Error.make(.InvalidArgument)); }
-    if radix < 2 || radix > 36 { ret .Err(Error.make(.InvalidArgument)); }
-    var reverse = ByteBuffer.new();
-    var fresh = ByteBuffer.new();
-    var magnitude = zero();
-    match clone(&magnitude, source) {
-        .Ok => {}, .Err(error) => { ret .Err(error); },
-    }
-    magnitude.negative = false;
-    if magnitude.len == 0 {
-        match ByteBuffer.append_byte(&reverse, '0') {
-            .Ok => {}, .Err(error) => { drop(&magnitude); ret .Err(error); },
-        }
-    }
-    while magnitude.len != 0 {
-        var remainder = div_small(&magnitude, radix as u32);
-        var byte: u8 = 0;
-        match digit_char(remainder as u8, uppercase) {
-            .Ok(value) => { byte = value; },
-            .Err(error) => { drop(&magnitude); ByteBuffer.drop(&reverse); ret .Err(error); },
-        }
-        match ByteBuffer.append_byte(&reverse, byte) {
-            .Ok => {},
-            .Err(error) => {
-                drop(&magnitude); ByteBuffer.drop(&reverse); ret .Err(error);
-            },
-        }
-    }
-    if source.negative {
-        match ByteBuffer.append_byte(&fresh, '-') {
-            .Ok => {},
-            .Err(error) => {
-                drop(&magnitude); ByteBuffer.drop(&reverse); ret .Err(error);
-            },
-        }
-    }
-    var i = ByteBuffer.size(reverse);
-    var digits = ByteBuffer.as_slice(reverse);
-    while i > 0 {
-        i = i - 1;
-        match ByteBuffer.append_byte(&fresh, digits.ptr[i]) {
-            .Ok => {},
-            .Err(error) => {
-                drop(&magnitude); ByteBuffer.drop(&reverse); ByteBuffer.drop(&fresh);
-                ret .Err(error);
-            },
-        }
-    }
-    drop(&magnitude);
-    ByteBuffer.drop(&reverse);
-    ByteBuffer.drop(out);
-    out.* = ByteBuffer.take(&fresh);
-    ret .Ok;
-}
-
-fn to_u64(out: *u64, source: BigInt): Status {
-    if out == null || !valid(source) { ret .Err(Error.make(.InvalidArgument)); }
-    if source.negative || source.len > 2 { ret .Err(Error.make(.Overflow)); }
-    var value: u64 = 0;
-    if source.len > 0 { value = source.limbs[0] as u64; }
-    if source.len > 1 { value = value | ((source.limbs[1] as u64) << 32); }
-    out.* = value;
-    ret .Ok;
-}
-
-fn to_i64(out: *i64, source: BigInt): Status {
-    if out == null || !valid(source) { ret .Err(Error.make(.InvalidArgument)); }
-    if source.len > 2 { ret .Err(Error.make(.Overflow)); }
-    var magnitude: u64 = 0;
-    if source.len > 0 { magnitude = source.limbs[0] as u64; }
-    if source.len > 1 { magnitude = magnitude | ((source.limbs[1] as u64) << 32); }
-    if !source.negative {
-        if magnitude > 9223372036854775807 { ret .Err(Error.make(.Overflow)); }
-        out.* = magnitude as i64;
-        ret .Ok;
-    }
-    if magnitude > 9223372036854775808 { ret .Err(Error.make(.Overflow)); }
-    if magnitude == 9223372036854775808 {
-        out.* = magnitude as i64;
-    } else {
-        out.* = 0 - magnitude as i64;
-    }
-    ret .Ok;
-}
-
-fn to_usize(out: *usize, source: BigInt): Status {
-    if out == null { ret .Err(Error.make(.InvalidArgument)); }
-    var value: u64 = 0;
-    match to_u64(&value, source) {
-        .Ok => {}, .Err(error) => { ret .Err(error); },
-    }
-    if value > (~(0 as usize)) as u64 { ret .Err(Error.make(.Overflow)); }
-    out.* = value as usize;
-    ret .Ok;
-}
-
-fn expect_ok(result: Status) {
-    match result { .Ok => {}, .Err(error) => { @assert(false, "unexpected BigInt error"); }, }
-}
-
-fn expect_text_radix(value: BigInt, radix: u8, uppercase: bool, expected: []u8) {
-    var buffer = ByteBuffer.new();
-    expect_ok(format(&buffer, value, radix, uppercase));
-    var actual = ByteBuffer.as_slice(buffer);
-    @assert(actual.len == expected.len, "formatted length");
-    var i: usize = 0;
-    while i < expected.len {
-        @assert(actual.ptr[i] == expected.ptr[i], "formatted byte");
-        i = i + 1;
-    }
-    ByteBuffer.drop(&buffer);
-}
-
-fn expect_text(value: BigInt, expected: []u8) {
-    expect_text_radix(value, 10, false, expected);
-}
-
-fn parsed(text: []u8): BigInt {
-    var value = zero();
-    expect_ok(parse(&value, text, 10));
-    ret value;
-}
-
-fn test_arithmetic() {
-    var a = parsed("4294967295");
-    var one = zero();
-    expect_ok(from_u64(&one, 1));
-    expect_ok(add(&a, a, one));
-    expect_text(a, "4294967296");
-
-    var all_ones = zero();
-    expect_ok(parse(&all_ones, "ffffffffffffffffffffffffffffffff", 16));
-    expect_ok(add(&all_ones, all_ones, one));
-    expect_text(all_ones, "340282366920938463463374607431768211456");
-    expect_ok(sub(&all_ones, all_ones, one));
-    expect_text_radix(all_ones, 16, false, "ffffffffffffffffffffffffffffffff");
-
-    var two_64 = parsed("18446744073709551616");
-    var product = zero();
-    expect_ok(mul(&product, two_64, two_64));
-    expect_text(product, "340282366920938463463374607431768211456");
-
-    var difference = zero();
-    expect_ok(sub(&difference, product, two_64));
-    expect_text(difference, "340282366920938463444927863358058659840");
-
-    var shifted = zero();
-    expect_ok(mul_pow2(&shifted, one, 257));
-    var bits: usize = 0;
-    expect_ok(bit_length(&bits, shifted));
-    @assert(bits == 258);
-    expect_ok(div_pow2(&shifted, shifted, 257));
-    @assert(eq(shifted, one));
-
-    drop(&shifted); drop(&difference); drop(&product); drop(&two_64);
-    drop(&all_ones); drop(&one); drop(&a);
-}
-
-fn test_division_and_sign() {
-    var dividend = parsed("340282366920938463463374607431768211456");
-    var divisor = parsed("18446744073709551616");
-    var quotient = zero();
-    var remainder = zero();
-    expect_ok(div_rem(&quotient, &remainder, dividend, divisor));
-    @assert(eq(quotient, divisor));
-    @assert(is_zero(remainder));
-
-    var long_dividend = parsed("123456789012345678901234567890");
-    var long_divisor = parsed("9876543210");
-    expect_ok(div_rem(&quotient, &remainder, long_dividend, long_divisor));
-    expect_text(quotient, "12499999887343749990");
-    expect_text(remainder, "1562499990");
-
-    var negative = zero();
-    var seven = zero();
-    expect_ok(from_i64(&negative, -100));
-    expect_ok(from_u64(&seven, 7));
-    expect_ok(div_rem(&negative, &remainder, negative, seven));
-    expect_text(negative, "-14");
-    expect_text(remainder, "-2");
-
-    var reconstructed = zero();
-    expect_ok(mul(&reconstructed, negative, seven));
-    expect_ok(add(&reconstructed, reconstructed, remainder));
-    expect_text(reconstructed, "-100");
-
-    var saved = parsed("123");
-    var zero_divisor = zero();
-    match div_rem(&saved, &remainder, dividend, zero_divisor) {
-        .Ok => { @assert(false, "division by zero"); },
-        .Err(error) => { match error.kind { .DivisionByZero => {}, _ => { @assert(false); }, } },
-    }
-    expect_text(saved, "123");
-
-    drop(&zero_divisor); drop(&saved); drop(&reconstructed); drop(&seven); drop(&negative);
-    drop(&long_divisor); drop(&long_dividend);
-    drop(&remainder); drop(&quotient); drop(&divisor); drop(&dividend);
-}
-
-fn test_parse_format_and_conversion() {
-    var value = zero();
-    expect_ok(parse(&value, "-deadBEEF", 16));
-    var output = ByteBuffer.new();
-    expect_ok(format(&output, value, 16, false));
-    var text = ByteBuffer.as_slice(output);
-    @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();
-    expect_ok(parse(&base36, "zzzzzzzzzz", 36));
-    expect_text_radix(base36, 36, false, "zzzzzzzzzz");
-
-    var preserved = parsed("77");
-    match parse(&preserved, "12x", 10) {
-        .Ok => { @assert(false, "invalid digit"); },
-        .Err(error) => {
-            @assert(error.has_position && error.position == 2);
-            match error.kind { .InvalidDigit => {}, _ => { @assert(false); }, }
-        },
-    }
-    expect_text(preserved, "77");
-
-    var max = parsed("18446744073709551615");
-    var converted: u64 = 0;
-    expect_ok(to_u64(&converted, max));
-    @assert(converted == 18446744073709551615);
-    var over = parsed("18446744073709551616");
-    converted = 17;
-    match to_u64(&converted, over) {
-        .Ok => { @assert(false); },
-        .Err(error) => { match error.kind { .Overflow => {}, _ => { @assert(false); }, } },
-    }
-    @assert(converted == 17, "failed conversion preserves output");
-
-    var minimum = parsed("-9223372036854775808");
-    var signed_value: i64 = 0;
-    expect_ok(to_i64(&signed_value, minimum));
-    @assert(signed_value == (9223372036854775808 as u64) as i64);
-    var signed_over = parsed("9223372036854775808");
-    signed_value = 23;
-    match to_i64(&signed_value, signed_over) {
-        .Ok => { @assert(false); },
-        .Err(error) => { match error.kind { .Overflow => {}, _ => { @assert(false); }, } },
-    }
-    @assert(signed_value == 23, "failed signed conversion preserves output");
-
-    var copied = zero();
-    expect_ok(clone(&copied, max));
-    var one_more = parsed("1");
-    expect_ok(add(&max, max, one_more));
-    expect_text(copied, "18446744073709551615");
-    var moved = take(&copied);
-    @assert(copied.limbs == null && copied.len == 0 && copied.cap == 0);
-    expect_text(moved, "18446744073709551615");
-
-    drop(&moved); drop(&copied); drop(&one_more); drop(&signed_over); drop(&minimum);
-    drop(&over); drop(&max); drop(&preserved); drop(&base36);
-    ByteBuffer.drop(&output); drop(&value);
-}
-
-fn expect_i64_value(value: BigInt, expected: i64) {
-    var actual: i64 = 0;
-    expect_ok(to_i64(&actual, value));
-    @assert(actual == expected, "small integer reference result");
-}
-
-fn test_small_reference() {
-    var left = zero();
-    var right = zero();
-    var result = zero();
-    var remainder = zero();
-    var a: i64 = 0;
-    while a <= 0 {
-        var b: i64 = 2;
-        while b <= 2 {
-            expect_ok(from_i64(&left, a));
-            expect_ok(from_i64(&right, b));
-            expect_ok(add(&result, left, right));
-            expect_i64_value(result, a + b);
-            expect_ok(sub(&result, left, right));
-            expect_i64_value(result, a - b);
-            expect_ok(mul(&result, left, right));
-            expect_i64_value(result, a * b);
-            if b != 0 {
-                expect_ok(div_rem(&result, &remainder, left, right));
-                expect_i64_value(result, a / b);
-                expect_i64_value(remainder, a % b);
-            }
-            b = b + 1;
-        }
-        a = a + 1;
-    }
-    drop(&remainder); drop(&result); drop(&right); drop(&left);
-}
-
-fn main() {
-    test_arithmetic();
-    test_division_and_sign();
-    test_parse_format_and_conversion();
-    test_small_reference();
-}

stage2/utils/byte_buffer.spl

@@ -1,4 +1,5 @@
 // Expect retcode = 0
+
 // 拥有字节存储的动态缓冲区。普通赋值会复制所有权，转移必须使用 take。
 // 借用切片和 z_ptr 返回的指针在下一次可能扩容的操作或 drop 后失效。
 
@@ -205,6 +206,7 @@ fn expect_z(self: *ByteBuffer) {
 
 fn test_append_and_resize() {
     var buffer = new();
+    defer { drop(&buffer); }
     @assert(is_empty(buffer));
     @assert(size(buffer) == 0 && capacity(buffer) == 0);
     expect_z(&buffer);
@@ -232,11 +234,11 @@ fn test_append_and_resize() {
     clear(&buffer);
     @assert(is_empty(buffer) && capacity(buffer) != 0, "clear retains allocation");
     expect_z(&buffer);
-    drop(&buffer);
 }
 
 fn test_growth_and_self_append() {
     var buffer = new();
+    defer { drop(&buffer); }
     var i: usize = 0;
     while i < 8 {
         expect_ok(append_byte(&buffer, (65 + i) as u8));
@@ -253,14 +255,15 @@ fn test_growth_and_self_append() {
         i = i + 1;
     }
     expect_z(&buffer);
-    drop(&buffer);
 }
 
 fn test_ownership_and_failures() {
     var source = new();
+    defer { drop(&source); }
     expect_ok(append(&source, "owned"));
 
     var copy = new();
+    defer { drop(&copy); }
     expect_ok(clone(&copy, source));
     @assert(copy.storage.len == 5, "clone logical length");
     source.storage.data[0] = 79;
@@ -293,6 +296,7 @@ fn test_ownership_and_failures() {
     expect_bytes(copy, "wne");
 
     var moved = take(&copy);
+    defer { drop(&moved); }
     @assert(copy.storage.raw == null && copy.storage.len == 0 && copy.storage.cap == 0);
     expect_bytes(moved, "wne");
 
@@ -304,10 +308,6 @@ fn test_ownership_and_failures() {
             moved.storage.cap == old_cap,
             "overflow preserves byte buffer");
     expect_bytes(moved, "wne");
-
-    drop(&moved);
-    drop(&copy);
-    drop(&source);
 }
 
 fn main() {

stage2/utils/byte_view.spl

@@ -1,4 +1,5 @@
 // Expect retcode = 0
+
 // 借用的字节字符串视图；不拥有、不分配也不释放 ptr 指向的存储。
 // 非空视图要求 ptr 在整个借用期内可读。空视图统一表示为 null/0。
 
@@ -34,6 +35,14 @@ fn from_slice(bytes: []u8): ByteView {
     ret ByteView { .ptr = bytes.ptr, .len = bytes.len };
 }
 
+// 以 NUL 结尾的 C 字符串转为视图；不复制、不拥有。null 规范化为空视图。
+fn from_cstr(text: *u8): ByteView {
+    if text == null { ret empty(); }
+    var len: usize = 0;
+    while text[len] != 0 { len = len + 1; }
+    ret from_slice(.{ .ptr = text, .len = len });
+}
+
 fn as_slice(self: ByteView): []u8 {
     ret .{ .ptr = self.ptr, .len = self.len };
 }
@@ -212,6 +221,14 @@ fn test_construction_and_access() {
 
     var bytes = as_slice(text);
     @assert(bytes.ptr == text.ptr && bytes.len == text.len, "slice conversion borrows");
+
+    var cstr_data: [4]u8 = .[104, 105, 0, 88];
+    var cstr = from_cstr(&cstr_data[0]);
+    @assert(cstr.len == 2 && cstr.ptr[0] == 104 && cstr.ptr[1] == 105,
+            "from_cstr stops at NUL");
+    @assert(from_cstr(&cstr_data[2]).len == 0, "from_cstr at NUL is empty");
+    var none = from_cstr(null);
+    @assert(none.ptr == null && none.len == 0, "from_cstr null normalized");
 }
 
 fn test_compare_and_slice() {

stage2/utils/diag.spl

@@ -1,243 +0,0 @@
-// 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();
-}

stage2/utils/error.spl

@@ -1,4 +1,5 @@
 // Expect retcode = 0
+
 // 公共错误值不持有资源，可按值复制；不分配错误消息字符串。
 // position 是调用方输入中的从零开始的位置；仅在 has_position 时有效。
 // 未定位错误的 position 规范为 0，不能用 position == 0 判断是否有位置。

stage2/utils/fmt.spl

@@ -1,10 +1,13 @@
 // Expect retcode = 0
+
 // 数值/布尔格式化，写入 ByteBuffer。radix 取值 2..36；uppercase 只影响 10..35 的字母形式。
-// 不分配额外内存（除目标 ByteBuffer 自身扩容），失败不产生部分输出。
+// 整数转换统一复用 MPA。先在临时缓冲区完成转换，再一次性追加，失败不产生部分输出。
 
 type .{ .Error, .Status, .make } = @import("error.spl");
 type .{ .ByteBuffer } = @import("byte_buffer.spl");
-type .{ .digit_char } = @import("ascii.spl");
+type .{ .ByteView } = @import("byte_view.spl");
+type .{ .Value, .int, .uint, .char, .str, .addr } = @import("value.spl");
+type .{ .BigInt, .Bits } = @import("mp.spl");
 
 // 注意：radix 边界直接用字面量。stage1 刻意不做常量求值/折叠，
 // 顶层 const 不应用于运行时表达式。
@@ -12,53 +15,47 @@ fn valid_radix(radix: u8): bool {
     ret radix >= 2 && radix <= 36;
 }
 
-fn append_u64(out: *ByteBuffer, value: u64, radix: u8, uppercase: bool): Status {
+fn append_big_int(out: *ByteBuffer, value: BigInt,
+                  radix: u8, uppercase: bool): Status {
     if !valid_radix(radix) { ret .Err(make(.InvalidArgument)); }
-
-    var digits: [64]u8 = undefined;
-    var count: usize = 0;
-
-    if value == 0 {
-        digits[0] = '0';
-        count = 1;
-    } else {
-        var v = value;
-        var base = radix as u64;
-        while v != 0 {
-            var d = (v % base) as u8;
-            match digit_char(d, uppercase) {
-                .Ok(c) => { digits[count] = c; },
-                .Err(error) => { ret .Err(error); },
-            }
-            count = count + 1;
-            v = v / base;
-        }
-    }
-
-    // digits 是低位在前，反向写入得到正常顺序。
-    var i = count;
-    while i > 0 {
-        i = i - 1;
-        match ByteBuffer.append_byte(out, digits[i]) {
+    var text = ByteBuffer.new();
+    defer { ByteBuffer.drop(&text); }
+    match BigInt.format(&text, value, radix, uppercase) {
         .Ok => {},
         .Err(error) => { ret .Err(error); },
     }
+    ret ByteBuffer.append(out, ByteBuffer.as_slice(text));
+}
+
+fn append_bits(out: *ByteBuffer, value: Bits, signed: bool,
+               radix: u8, uppercase: bool): Status {
+    var integer = BigInt.zero();
+    defer { BigInt.drop(&integer); }
+    match Bits.to_big(&integer, value, signed) {
+        .Ok => {},
+        .Err(error) => { ret .Err(error); },
     }
-    ret .Ok;
+    ret append_big_int(out, integer, radix, uppercase);
+}
+
+fn append_u64(out: *ByteBuffer, value: u64, radix: u8, uppercase: bool): Status {
+    var integer = BigInt.zero();
+    defer { BigInt.drop(&integer); }
+    match BigInt.from_u64(&integer, value) {
+        .Ok => {},
+        .Err(error) => { ret .Err(error); },
+    }
+    ret append_big_int(out, integer, radix, uppercase);
 }
 
 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, '-') {
+    var integer = BigInt.zero();
+    defer { BigInt.drop(&integer); }
+    match BigInt.from_i64(&integer, value) {
         .Ok => {},
         .Err(error) => { ret .Err(error); },
     }
-        // value 为补码，(0 - (value as u64)) 得到绝对值（含 i64 最小值）。
-        var magnitude = 0 - (value as u64);
-        ret append_u64(out, magnitude, radix, uppercase);
-    }
-    ret append_u64(out, value as u64, radix, uppercase);
+    ret append_big_int(out, integer, radix, uppercase);
 }
 
 fn append_usize(out: *ByteBuffer, value: usize, radix: u8, uppercase: bool): Status {
@@ -74,6 +71,158 @@ fn append_bool(out: *ByteBuffer, value: bool): Status {
     ret ByteBuffer.append(out, "false");
 }
 
+// ---- 显式装箱值格式化 ----
+
+// Int/Uint 落到整数；signed 仅对 Int 的十进制生效，其余按无符号位模式输出。
+fn append_value_int(out: *ByteBuffer, value: Value, radix: u8, uppercase: bool,
+                    signed: bool): Status {
+    match value {
+        .Int(v) => {
+            if signed && radix == 10 { ret append_i64(out, v, radix, uppercase); }
+            ret append_u64(out, v as u64, radix, uppercase);
+        },
+        .Uint(v) => { ret append_u64(out, v, radix, uppercase); },
+        _ => { ret .Err(make(.InvalidFormat)); },
+    }
+}
+
+fn append_value_char(out: *ByteBuffer, value: Value): Status {
+    match value {
+        .Char(c) => { ret ByteBuffer.append_byte(out, c); },
+        _ => { ret .Err(make(.InvalidFormat)); },
+    }
+}
+
+fn append_value_str(out: *ByteBuffer, value: Value): Status {
+    match value {
+        .Str(view) => { ret ByteBuffer.append(out, ByteView.as_slice(view)); },
+        _ => { ret .Err(make(.InvalidFormat)); },
+    }
+}
+
+fn append_value_ptr(out: *ByteBuffer, value: Value): Status {
+    match value {
+        .Ptr(p) => {
+            match ByteBuffer.append(out, "0x") {
+                .Ok => {},
+                .Err(error) => { ret .Err(error); },
+            }
+            ret append_u64(out, p as usize, 16, false);
+        },
+        _ => { ret .Err(make(.InvalidFormat)); },
+    }
+}
+
+// {} 自动格式化：按 Value 变体选择默认表示。
+fn append_value_auto(out: *ByteBuffer, value: Value): Status {
+    match value {
+        .Int(v) => { ret append_i64(out, v, 10, false); },
+        .Uint(v) => { ret append_u64(out, v, 10, false); },
+        .Bool(v) => { ret append_bool(out, v); },
+        .Char(c) => { ret ByteBuffer.append_byte(out, c); },
+        .Str(view) => { ret ByteBuffer.append(out, ByteView.as_slice(view)); },
+        .Ptr(p) => {
+            match ByteBuffer.append(out, "0x") {
+                .Ok => {},
+                .Err(error) => { ret .Err(error); },
+            }
+            ret append_u64(out, p as usize, 16, false);
+        },
+    }
+    ret .Ok;
+}
+
+// {} 风格：{} 自动；{:d} {:i} {:u} {:x} {:X} {:o} {:b} {:c} {:s} {:p} 指定。
+// {{ 与 }} 输出字面量括号。参数按序从 args 消费；数量不足、类型不符、未知转换
+// 或括号不配返回 Err(.InvalidFormat)。失败不产生部分输出：先在临时缓冲完成。
+fn format(out: *ByteBuffer, fmt: []u8, args: []Value): Status {
+    var temp = ByteBuffer.new();
+    defer { ByteBuffer.drop(&temp); }
+
+    var argi: usize = 0;
+    var i: usize = 0;
+    while i < fmt.len {
+        var c = fmt.ptr[i];
+
+        if c == '{' {
+            if i + 1 < fmt.len && fmt.ptr[i + 1] == '{' {
+                match ByteBuffer.append_byte(&temp, '{') {
+                    .Ok => {},
+                    .Err(error) => { ret .Err(error); },
+                }
+                i = i + 2;
+                continue;
+            }
+
+            var j = i + 1;
+            if j < fmt.len && fmt.ptr[j] == ':' { j = j + 1; }
+            var has_spec = false;
+            var spec: u8 = 0;
+            if j < fmt.len && fmt.ptr[j] != '}' {
+                has_spec = true;
+                spec = fmt.ptr[j];
+                j = j + 1;
+            }
+            if j >= fmt.len || fmt.ptr[j] != '}' { ret .Err(make(.InvalidFormat)); }
+            i = j + 1;
+
+            if argi >= args.len { ret .Err(make(.InvalidFormat)); }
+            var value = args.ptr[argi];
+            argi = argi + 1;
+
+            var step: Status = .Ok;
+            if !has_spec {
+                step = append_value_auto(&temp, value);
+            } else if spec == 'd' || spec == 'i' {
+                step = append_value_int(&temp, value, 10, false, true);
+            } else if spec == 'u' {
+                step = append_value_int(&temp, value, 10, false, false);
+            } else if spec == 'x' {
+                step = append_value_int(&temp, value, 16, false, false);
+            } else if spec == 'X' {
+                step = append_value_int(&temp, value, 16, true, false);
+            } else if spec == 'o' {
+                step = append_value_int(&temp, value, 8, false, false);
+            } else if spec == 'b' {
+                step = append_value_int(&temp, value, 2, false, false);
+            } else if spec == 'c' {
+                step = append_value_char(&temp, value);
+            } else if spec == 's' {
+                step = append_value_str(&temp, value);
+            } else if spec == 'p' {
+                step = append_value_ptr(&temp, value);
+            } else {
+                ret .Err(make(.InvalidFormat));
+            }
+            match step {
+                .Ok => {},
+                .Err(error) => { ret .Err(error); },
+            }
+            continue;
+        }
+
+        if c == '}' {
+            if i + 1 < fmt.len && fmt.ptr[i + 1] == '}' {
+                match ByteBuffer.append_byte(&temp, '}') {
+                    .Ok => {},
+                    .Err(error) => { ret .Err(error); },
+                }
+                i = i + 2;
+                continue;
+            }
+            ret .Err(make(.InvalidFormat));
+        }
+
+        match ByteBuffer.append_byte(&temp, c) {
+            .Ok => {},
+            .Err(error) => { ret .Err(error); },
+        }
+        i = i + 1;
+    }
+
+    ret ByteBuffer.append(out, ByteBuffer.as_slice(temp));
+}
+
 // ---- 测试辅助 ----
 
 fn expect_ok(result: Status) {
@@ -107,6 +256,7 @@ fn expect_text(buffer: *ByteBuffer, expected: []u8) {
 
 fn test_unsigned() {
     var b = ByteBuffer.new();
+    defer { ByteBuffer.drop(&b); }
 
     expect_ok(append_u64(&b, 0, 10, false));
     expect_text(&b, "0");
@@ -147,12 +297,11 @@ fn test_unsigned() {
     expect_ok(append_u64(&b, 18446744073709551615, 2, false));
     expect_text(&b, "1111111111111111111111111111111111111111111111111111111111111111");
     ByteBuffer.clear(&b);
-
-    ByteBuffer.drop(&b);
 }
 
 fn test_signed() {
     var b = ByteBuffer.new();
+    defer { ByteBuffer.drop(&b); }
 
     expect_ok(append_i64(&b, 0, 10, false));
     expect_text(&b, "0");
@@ -177,12 +326,11 @@ fn test_signed() {
     expect_ok(append_u64(&b, 48879, 16, true));
     expect_text(&b, "BEEF");
     ByteBuffer.clear(&b);
-
-    ByteBuffer.drop(&b);
 }
 
 fn test_usize_bool_and_errors() {
     var b = ByteBuffer.new();
+    defer { ByteBuffer.drop(&b); }
 
     expect_ok(append_usize(&b, 1000 as usize, 10, false));
     expect_text(&b, "1000");
@@ -202,11 +350,48 @@ fn test_usize_bool_and_errors() {
     expect_invalid(append_i64(&b, -1, 0, false));
     @assert(ByteBuffer.size(b) == 0, "invalid radix writes nothing");
 
-    ByteBuffer.drop(&b);
+    expect_ok(ByteBuffer.append(&b, "prefix"));
+    expect_invalid(append_i64(&b, -123, 1, false));
+    expect_text(&b, "prefix");
+}
+
+fn test_mpa_integer_formatting() {
+    var b = ByteBuffer.new();
+    defer { ByteBuffer.drop(&b); }
+
+    var huge = BigInt.zero();
+    defer { BigInt.drop(&huge); }
+    expect_ok(BigInt.parse(&huge, "1234567890123456789012345678901234567890", 10));
+    expect_ok(append_big_int(&b, huge, 10, false));
+    expect_text(&b, "1234567890123456789012345678901234567890");
+    ByteBuffer.clear(&b);
+
+    expect_ok(BigInt.parse(&huge, "-ffffffffffffffffffffffffffffffff", 16));
+    expect_ok(append_big_int(&b, huge, 16, true));
+    expect_text(&b, "-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF");
+    ByteBuffer.clear(&b);
+
+    var raw = BigInt.zero();
+    defer { BigInt.drop(&raw); }
+    expect_ok(BigInt.from_u64(&raw, 255));
+    var bits: Bits = undefined;
+    match Bits.new(8) {
+        .Ok(value) => { bits = value; },
+        .Err(error) => { @assert(false, "Bits construction"); },
+    }
+    defer { Bits.drop(&bits); }
+    expect_ok(Bits.from_big_checked(&bits, raw, 8, false));
+
+    expect_ok(append_bits(&b, bits, false, 10, false));
+    expect_text(&b, "255");
+    ByteBuffer.clear(&b);
+    expect_ok(append_bits(&b, bits, true, 10, false));
+    expect_text(&b, "-1");
 }
 
 fn test_composition_and_growth() {
     var b = ByteBuffer.new();
+    defer { ByteBuffer.drop(&b); }
 
     expect_ok(ByteBuffer.append(&b, "x="));
     expect_ok(append_i64(&b, -42, 10, false));
@@ -229,13 +414,98 @@ fn test_composition_and_growth() {
     var s = ByteBuffer.as_slice(b);
     @assert(s.ptr[0] == '0', "starts with 0");
     @assert(s.ptr[1] == ',', "0 then comma");
+}
 
-    ByteBuffer.drop(&b);
+fn expect_format(fmt: []u8, args: []Value, expected: []u8) {
+    var b = ByteBuffer.new();
+    defer { ByteBuffer.drop(&b); }
+    expect_ok(format(&b, fmt, args));
+    expect_text(&b, expected);
+}
+
+fn expect_invalid_format(fmt: []u8, args: []Value) {
+    var b = ByteBuffer.new();
+    defer { ByteBuffer.drop(&b); }
+    match format(&b, fmt, args) {
+        .Ok => { @assert(false, "expected invalid format"); },
+        .Err(error) => {
+            match error.kind {
+                .InvalidFormat => {},
+                _ => { @assert(false, "expected InvalidFormat kind"); },
+            }
+        },
+    }
+}
+
+fn test_format_specs() {
+    var one: [1]Value = .[int(-5)];
+    expect_format("{}", one[0..1], "-5");
+    expect_format("{d}", one[0..1], "-5");
+    expect_format("{i}", one[0..1], "-5");
+
+    var u: [1]Value = .[uint(7)];
+    expect_format("{}", u[0..1], "7");
+    expect_format("{u}", u[0..1], "7");
+    expect_format("{x}", u[0..1], "7");
+    expect_format("{o}", u[0..1], "7");
+    expect_format("{b}", u[0..1], "111");
+
+    var hex: [1]Value = .[uint(255)];
+    expect_format("{x}", hex[0..1], "ff");
+    expect_format("{X}", hex[0..1], "FF");
+
+    var oct: [1]Value = .[uint(8)];
+    expect_format("{o}", oct[0..1], "10");
+
+    var bits: [1]Value = .[uint(5)];
+    expect_format("{b}", bits[0..1], "101");
+
+    var ch: [1]Value = .[char('A')];
+    expect_format("{}", ch[0..1], "A");
+    expect_format("{c}", ch[0..1], "A");
+
+    var text: [1]Value = .[str("hi")];
+    expect_format("{}", text[0..1], "hi");
+    expect_format("{s}", text[0..1], "hi");
+
+    var nul: [1]Value = .[addr(null)];
+    expect_format("{p}", nul[0..1], "0x0");
+
+    var none: [1]Value = .[int(0)];
+    expect_format("100%", none[0..0], "100%");
+    expect_format("{{}}", none[0..0], "{}");
+    expect_format("plain", none[0..0], "plain");
+}
+
+fn test_format_mixed_and_failure() {
+    var args: [3]Value = .[int(42), str("x"), char('!')];
+    expect_format("n={} s={} c={}", args[0..3], "n=42 s=x c=!");
+    expect_format("n={} s={s} c={c}", args[0..3], "n=42 s=x c=!");
+
+    // 失败不改动输出，且不产生部分结果：前两个转换成功、第三个参数不足。
+    var b = ByteBuffer.new();
+    defer { ByteBuffer.drop(&b); }
+    expect_ok(ByteBuffer.append(&b, "prefix"));
+    var two: [2]Value = .[int(1), int(2)];
+    expect_invalid_format("a={} b={} c={}", two[0..2]);
+    expect_text(&b, "prefix");
+
+    var only_int: [1]Value = .[int(1)];
+    expect_invalid_format("{} {}", only_int[0..1]);
+    expect_invalid_format("{}", only_int[0..0]);
+    expect_invalid_format("trailing {", only_int[0..1]);
+    expect_invalid_format("dangling }", only_int[0..1]);
+    expect_invalid_format("{q}", only_int[0..1]);
+    expect_invalid_format("{s}", only_int[0..1]);
+    expect_invalid_format("{c}", only_int[0..1]);
 }
 
 fn main() {
     test_unsigned();
     test_signed();
     test_usize_bool_and_errors();
+    test_mpa_integer_formatting();
     test_composition_and_growth();
+    test_format_specs();
+    test_format_mixed_and_failure();
 }

stage2/utils/hash.spl

@@ -1,4 +1,5 @@
 // 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。
@@ -7,10 +8,11 @@ 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 h: usize = 14695981039346656037; // FNV-1a 64 位 offset basis
+    var prime: usize = 1099511628211;    // FNV 质数
     var i: usize = 0;
     while i < len {
-        h = (h ^ (data[i] as usize)) * 1099511628211;
+        h = (h ^ (data[i] as usize)) * prime;
         i = i + 1;
     }
     ret h;
@@ -22,12 +24,13 @@ fn hash_byte_view(view: ByteView): usize {
 
 // 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 prime: usize = 1099511628211; // splitmix 风格的混合乘子
     var h = value;
-    h = (h ^ (h >> 30)) * 1099511628211;
+    h = (h ^ (h >> 30)) * prime;
     h = h ^ (h >> 27);
-    h = h * 1099511628211;
+    h = h * prime;
     h = h ^ (h >> 31);
-    if h == 0 { ret 1099511628211; } // 0 映射到非零
+    if h == 0 { ret prime; } // 0 映射到非零
     ret h;
 }
 

stage2/utils/io.spl

@@ -1,14 +1,17 @@
 // 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）
+//   - 精确读取、完整写入和普通文件整读
 //   - 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。
+// 短读、短写后的零进展及关闭失败返回 Err(.Io)；文件写入不承诺回滚。
 
 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");
+type Mem = @import("../sys/mem.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 {
@@ -31,59 +34,99 @@ fn open_cstr(path: ByteView, mode: ByteView): File {
     var mbuf = ByteBuffer.new();
     var p = cstr(&pbuf, path);
     var m = cstr(&mbuf, mode);
-    var f = File.open(p, m);
+    var f = File { .handle = null };
+    if p != null && m != null { 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;
+// 反复写入直到完成。发生错误时文件可能已经包含前缀，不承诺回滚。
+fn write_all(f: *File, bytes: []u8): Status {
+    if f == null || !File.valid(f) { ret .Err(make(.Io)); }
+    if bytes.len != 0 && bytes.ptr == null { ret .Err(make(.InvalidArgument)); }
+    var total: usize = 0;
+    while total < bytes.len {
+        var written = File.write(f, bytes.ptr + total, (bytes.len - total) as u64);
+        if written == 0 || written > (bytes.len - total) as u64 {
+            ret .Err(make(.Io));
         }
-    ret total;
+        total = total + written as usize;
+    }
+    ret .Ok;
 }
 
-fn print(bytes: []u8) {
+fn print(bytes: []u8): Status {
     var out = File.stdout();
-    write_all(&out, bytes);
+    ret write_all(&out, bytes);
 }
 
-fn println(bytes: []u8) {
-    print(bytes);
-    print("\n");
+fn println(bytes: []u8): Status {
+    match print(bytes) {
+        .Ok => {},
+        .Err(error) => { ret .Err(error); },
+    }
+    ret print("\n");
 }
 
-fn eprint(bytes: []u8) {
+fn eprint(bytes: []u8): Status {
     var err = File.stderr();
-    write_all(&err, bytes);
+    ret 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)。
+// 仅供已经隔离的临时存储使用；失败时可能写入目标前缀。
+fn read_fully(f: *File, data: *u8, len: usize): Status {
+    var total: usize = 0;
+    while total < len {
+        var got = File.read(f, data + total, len - total);
+        if got == 0 || got > len - total { ret .Err(make(.Io)); }
+        total = total + got;
+    }
+    ret .Ok;
+}
+
+// 精确读取 bytes.len 字节。短读返回 Err(.Io)，失败保持 bytes 原内容不变。
+fn read_exact(f: *File, bytes: []u8): Status {
+    if f == null || !File.valid(f) { ret .Err(make(.Io)); }
+    if bytes.len == 0 { ret .Ok; }
+    if bytes.ptr == null { ret .Err(make(.InvalidArgument)); }
+
+    var temporary = ByteBuffer.new();
+    defer { ByteBuffer.drop(&temporary); }
+    match ByteBuffer.reserve(&temporary, bytes.len) {
+        .Ok => {},
+        .Err(error) => { ret .Err(error); },
+    }
+    match read_fully(f, temporary.storage.data, bytes.len) {
+        .Ok => {},
+        .Err(error) => { ret .Err(error); },
+    }
+    Mem.copy(bytes.ptr as *any, temporary.storage.data as *any, bytes.len);
+    ret .Ok;
+}
+
+// 读取整个普通文件到 out。打开、短读或关闭失败时保持 out 不变。
 fn read_all(path: ByteView, out: *ByteBuffer): Status {
+    if out == null { ret .Err(make(.InvalidArgument)); }
     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);
+    var fresh = ByteBuffer.new();
     if sz != 0 {
-        match ByteBuffer.reserve(out, sz as usize) {
+        match ByteBuffer.reserve(&fresh, sz) {
             .Ok => {},
-            .Err(error) => { File.close(&f); ret .Err(error); },
+            .Err(error) => { File.close(&f); ByteBuffer.drop(&fresh); 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;
+        match read_fully(&f, fresh.storage.data, sz) {
+            .Ok => {},
+            .Err(error) => { File.close(&f); ByteBuffer.drop(&fresh); ret .Err(error); },
         }
-    File.close(&f);
+        fresh.storage.len = sz;
+    }
+    if File.close(&f) != 0 { ByteBuffer.drop(&fresh); ret .Err(make(.Io)); }
+    ByteBuffer.drop(out);
+    out.* = ByteBuffer.take(&fresh);
     ret .Ok;
 }
 
@@ -93,14 +136,32 @@ fn view(text: []u8): ByteView {
     ret ByteView.from_slice(text);
 }
 
+fn expect_ok(result: Status) {
+    match result {
+        .Ok => {},
+        .Err(error) => { @assert(false, "unexpected io error"); },
+    }
+}
+
+fn expect_io(result: Status) {
+    match result {
+        .Ok => { @assert(false, "expected io error"); },
+        .Err(error) => {
+            match error.kind {
+                .Io => {},
+                _ => { @assert(false, "expected Io kind"); },
+            }
+        },
+    }
+}
+
 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);
+    expect_ok(write_all(&f, "hello\nworld"));
+    @assert(File.close(&f) == 0, "close writer");
 
     var out = ByteBuffer.new();
     match read_all(path, &out) {
@@ -118,14 +179,13 @@ fn test_append_and_missing() {
 
     var f = open_cstr(path, view("wb"));
     @assert(File.valid(&f), "open wb");
-    write_all(&f, "ab");
-    File.close(&f);
+    expect_ok(write_all(&f, "ab"));
+    @assert(File.close(&f) == 0, "close initial writer");
 
     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");
+    expect_ok(write_all(&f, "cd"));
+    @assert(File.close(&f) == 0, "close append writer");
 
     var out = ByteBuffer.new();
     match read_all(path, &out) {
@@ -142,11 +202,43 @@ fn test_append_and_missing() {
             match error.kind { .Io => {}, _ => { @assert(false, "error kind"); } }
         },
     }
+    got = ByteBuffer.as_slice(out);
+    @assert(got.len == 4 && got.ptr[0] == 'a' && got.ptr[3] == 'd',
+            "failed read_all preserves output");
     ByteBuffer.drop(&out);
 }
 
+fn test_read_exact_and_errors() {
+    var path = view("build/_io_exact.tmp");
+    var f = open_cstr(path, view("wb"));
+    @assert(File.valid(&f));
+    expect_ok(write_all(&f, "xyz"));
+    @assert(File.close(&f) == 0);
+
+    f = open_cstr(path, view("rb"));
+    @assert(File.valid(&f));
+    var too_long: [4]u8 = .[9, 8, 7, 6];
+    expect_io(read_exact(&f, too_long[0..4]));
+    @assert(too_long[0] == 9 && too_long[1] == 8 &&
+            too_long[2] == 7 && too_long[3] == 6,
+            "short read preserves destination");
+    @assert(File.close(&f) == 0);
+
+    f = open_cstr(path, view("rb"));
+    @assert(File.valid(&f));
+    var exact: [3]u8 = .[0, 0, 0];
+    expect_ok(read_exact(&f, exact[0..3]));
+    @assert(exact[0] == 'x' && exact[1] == 'y' && exact[2] == 'z');
+    @assert(File.close(&f) == 0);
+
+    var invalid = File { .handle = null };
+    expect_io(write_all(&invalid, "x"));
+    expect_io(read_exact(&invalid, exact[0..3]));
+}
+
 fn main() {
     test_roundtrip();
     test_append_and_missing();
-    println("io: ok");
+    test_read_exact_and_errors();
+    expect_ok(println("io: ok"));
 }

stage2/utils/json.spl

@@ -4,13 +4,14 @@
 //   - 输入 ByteView，字符串/数字以“源切片”借用，不做解码，round-trip 保真；
 //   - 节点池 Vec[JsonNode]；数组/对象子项存 Vec[JsonChild]，保留插入顺序；
 //   - 错误带字节位置（error.spl）；输出写 ByteBuffer；缩进/数字用 ByteBuffer/fmt。
-// 限制：不做字符串转义解码、无递归深度保护、数字不做数值化（原样保存）。
+// 限制：不做字符串转义解码、数字不做数值化（原样保存）；递归/输出深度上限见 max_depth()。
 
 type .{ .Error, .Status, .make, .at } = @import("error.spl");
 type .{ .ByteView, .ViewResult, .FindResult } = @import("byte_view.spl");
 type .{ .Vec } = @import("vec.spl");
 type .{ .ByteBuffer } = @import("byte_buffer.spl");
 type .{ .is_digit, .is_hex_digit, .is_space } = @import("ascii.spl");
+type .{ .checked_mul } = @import("memory.spl");
 type .{ .append_i64, .append_u64 } = @import("fmt.spl");
 
 type JsonKind = enum {
@@ -29,6 +30,10 @@ type JsonChild = struct {
     val: usize,
 }
 
+// 节点：kind 决定有效载荷字段，其余字段保持空。访问方必须按 kind 读取：
+//   Null/True/False -> text 空、children 空；
+//   Number/String   -> text 为源切片、children 空；
+//   Array/Object    -> children 保存子项、text 空。
 type JsonNode = struct {
     kind: JsonKind,
     text: ByteView,
@@ -57,6 +62,11 @@ type Parser = struct {
     nodes: Vec,
 }
 
+// 递归/输出深度上限：防止病态输入耗尽调用栈。256 层足够覆盖正常 JSON。
+fn max_depth(): usize {
+    ret 256;
+}
+
 fn blank_node(kind: JsonKind): JsonNode {
     ret JsonNode {
         .kind = kind,
@@ -78,7 +88,7 @@ fn intern_node(p: *Parser, node: JsonNode): NodeResult {
 fn node_at(doc: *Json, index: usize): *JsonNode {
     match Vec.get(&doc.nodes, index) {
         .Ok(ptr) => { ret ptr as *JsonNode; },
-        .Err(error) => { ret null; },
+        .Err(error) => { ret null; }, // 越界/无效池统一返回 null，调用方判空
     }
     ret null;
 }
@@ -122,6 +132,7 @@ fn parse_string_raw(p: *Parser): ViewResult {
                 }
             }
         } else if c < 32 {
+            // JSON 只禁止未转义的 U+0000..U+001F；0x7F 合法，不能用 ascii.is_control。
             ret .Err(at(.InvalidFormat, p.pos));
         }
         p.pos = p.pos + 1;
@@ -192,7 +203,8 @@ fn parse_number(p: *Parser): NodeResult {
     }
 }
 
-fn parse_array(p: *Parser): NodeResult {
+fn parse_array(p: *Parser, depth: usize): NodeResult {
+    if depth >= max_depth() { ret .Err(at(.OutOfRange, p.pos)); }
     p.pos = p.pos + 1; // '['
     var kids = Vec.blank(@sizeof(JsonChild), 8);
 
@@ -201,11 +213,14 @@ fn parse_array(p: *Parser): NodeResult {
         p.pos = p.pos + 1;
         var empty = blank_node(.Array);
         empty.children = kids;
-        ret intern_node(p, empty);
+        match intern_node(p, empty) {
+            .Ok(index) => { ret .Ok(index); },
+            .Err(error) => { Vec.drop(&empty.children); ret .Err(error); },
+        }
     }
 
     while true {
-        var element = parse_value(p);
+        var element = parse_value(p, depth + 1);
         var index: usize = 0;
         match element {
             .Ok(v) => { index = v; },
@@ -232,10 +247,14 @@ fn parse_array(p: *Parser): NodeResult {
 
     var node = blank_node(.Array);
     node.children = kids;
-    ret intern_node(p, node);
+    match intern_node(p, node) {
+        .Ok(index) => { ret .Ok(index); },
+        .Err(error) => { Vec.drop(&node.children); ret .Err(error); },
+    }
 }
 
-fn parse_object(p: *Parser): NodeResult {
+fn parse_object(p: *Parser, depth: usize): NodeResult {
+    if depth >= max_depth() { ret .Err(at(.OutOfRange, p.pos)); }
     p.pos = p.pos + 1; // '{'
     var kids = Vec.blank(@sizeof(JsonChild), 8);
 
@@ -244,7 +263,10 @@ fn parse_object(p: *Parser): NodeResult {
         p.pos = p.pos + 1;
         var empty = blank_node(.Object);
         empty.children = kids;
-        ret intern_node(p, empty);
+        match intern_node(p, empty) {
+            .Ok(index) => { ret .Ok(index); },
+            .Err(error) => { Vec.drop(&empty.children); ret .Err(error); },
+        }
     }
 
     while true {
@@ -262,7 +284,7 @@ fn parse_object(p: *Parser): NodeResult {
         }
         p.pos = p.pos + 1;
 
-        var value = parse_value(p);
+        var value = parse_value(p, depth + 1);
         var index: usize = 0;
         match value {
             .Ok(v) => { index = v; },
@@ -290,10 +312,13 @@ fn parse_object(p: *Parser): NodeResult {
 
     var node = blank_node(.Object);
     node.children = kids;
-    ret intern_node(p, node);
+    match intern_node(p, node) {
+        .Ok(index) => { ret .Ok(index); },
+        .Err(error) => { Vec.drop(&node.children); ret .Err(error); },
+    }
 }
 
-fn parse_value(p: *Parser): NodeResult {
+fn parse_value(p: *Parser, depth: usize): NodeResult {
     skip_ws(p);
     var c = peek(p);
     if c < 0 { ret .Err(at(.UnexpectedEnd, p.pos)); }
@@ -309,8 +334,8 @@ fn parse_value(p: *Parser): NodeResult {
         node.text = view;
         ret intern_node(p, node);
     }
-    if ch == '{' { ret parse_object(p); } // '{'
-    if ch == '[' { ret parse_array(p); }  // '['
+    if ch == '{' { ret parse_object(p, depth); } // '{'
+    if ch == '[' { ret parse_array(p, depth); }  // '['
     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); }
@@ -325,6 +350,7 @@ fn drop_node_pool(pool: *Vec) {
                 var n = ptr as *JsonNode;
                 Vec.drop(&n.children);
             },
+            // i < pool.len 且池有效，get 不会失败；失败时无额外资源可释放。
             .Err(error) => {},
         }
         i = i + 1;
@@ -341,7 +367,7 @@ fn parse(src: ByteView): JsonResult {
 
     skip_ws(&p);
     var root: usize = 0;
-    var result = parse_value(&p);
+    var result = parse_value(&p, 0);
     match result {
         .Ok(index) => { root = index; },
         .Err(error) => {
@@ -367,12 +393,6 @@ fn root_index(doc: *Json): usize {
     ret doc.root;
 }
 
-fn kind_of(doc: *Json, index: usize): JsonKind {
-    var n = node_at(doc, index);
-    if n == null { ret .Null; }
-    ret n.kind;
-}
-
 // 枚举 tag 取整（stage1 枚举 == 有缺陷，用 @enum2int 比较更稳妥）。
 fn kind_code(kind: JsonKind): usize {
     ret @enum2int(kind) as usize;
@@ -424,6 +444,7 @@ fn object_find(doc: *Json, index: usize, key: ByteView): FindResult {
                         var child = ptr as *JsonChild;
                         if ByteView.eq(child.key, key) { ret .Found(child.val); }
                     },
+                    // i < count == children.len，get 不会失败。
                     .Err(error) => {},
                 }
                 i = i + 1;
@@ -452,10 +473,16 @@ fn emit_indent(out: *ByteBuffer, indent: usize, depth: usize): Status {
         .Ok => {},
         .Err(error) => { ret .Err(error); },
     }
-    ret ByteBuffer.append_repeat(out, 32, depth * indent);
+    var spaces: usize = 0;
+    match checked_mul(depth, indent) {
+        .Ok(value) => { spaces = value; },
+        .Err(error) => { ret .Err(error); },
+    }
+    ret ByteBuffer.append_repeat(out, 32, spaces);
 }
 
 fn emit(out: *ByteBuffer, doc: *Json, index: usize, indent: usize, depth: usize): Status {
+    if depth >= max_depth() { ret .Err(make(.OutOfRange)); }
     var n = node_at(doc, index);
     if n == null { ret .Err(make(.OutOfRange)); }
 
@@ -738,19 +765,44 @@ fn test_errors() {
     expect_parse_error(parse(ByteView.from_slice("01")), 1);
 }
 
+fn test_depth_limit() {
+    var out = ByteBuffer.new();
+    defer { ByteBuffer.drop(&out); }
+    var i: usize = 0;
+    while i < max_depth() + 8 {
+        expect_ok(ByteBuffer.append_byte(&out, '['));
+        i = i + 1;
+    }
+    i = 0;
+    while i < max_depth() + 8 {
+        expect_ok(ByteBuffer.append_byte(&out, ']'));
+        i = i + 1;
+    }
+    match parse(ByteView.from_slice(ByteBuffer.as_slice(out))) {
+        .Ok(doc) => { drop(&doc); @assert(false, "deep input should be rejected"); },
+        .Err(error) => {
+            match error.kind {
+                .OutOfRange => {},
+                _ => { @assert(false, "depth limit error kind"); },
+            }
+        },
+    }
+}
+
 fn test_construction_helpers() {
     var out = ByteBuffer.new();
+    defer { ByteBuffer.drop(&out); }
     expect_ok(ByteBuffer.append(&out, "{\"id\":"));
     expect_ok(write_u64(&out, 18446744073709551615));
     expect_ok(ByteBuffer.append(&out, ",\"n\":"));
     expect_ok(write_int(&out, -7));
     expect_ok(ByteBuffer.append(&out, "}"));
     expect_bytes(&out, "{\"id\":18446744073709551615,\"n\":-7}");
-    ByteBuffer.drop(&out);
 }
 
 fn test_large_array() {
     var out = ByteBuffer.new();
+    defer { ByteBuffer.drop(&out); }
     expect_ok(ByteBuffer.append_byte(&out, '[')); // '['
     var i: usize = 0;
     while i < 1000 {
@@ -771,7 +823,6 @@ fn test_large_array() {
         },
         .Err(error) => { @assert(false, "parse large array"); },
     }
-    ByteBuffer.drop(&out);
 }
 
 fn main() {
@@ -780,6 +831,7 @@ fn main() {
     test_roundtrip_compact();
     test_pretty();
     test_errors();
+    test_depth_limit();
     test_construction_helpers();
     test_large_array();
 }

stage2/utils/log.spl

@@ -0,0 +1,205 @@
+// Expect retcode = 0
+
+// 编译器自身日志（区别于 diag：diag 负责带源码位置的诊断）。
+// 输出到 stderr，格式稳定：[LEVEL] name: message\n。
+// 消息为借用；Logger 不拥有 name。fatal 打印后终止进程（退出码 1）。
+// write 返回 Status；OOM 归类为 .Overflow（无独立内存耗尽错误码）。
+
+type .{ .Status, .make } = @import("error.spl");
+type .{ .ByteView } = @import("byte_view.spl");
+type .{ .ByteBuffer } = @import("byte_buffer.spl");
+type .{ .format } = @import("fmt.spl");
+type .{ .Value, .int, .str } = @import("value.spl");
+type .{ .eprint } = @import("io.spl");
+type Exit = @import("../sys/exit.spl");
+
+type Level = enum { Debug, Info, Warn, Error, Fatal }
+
+fn level_name(level: Level): []u8 {
+    match level {
+        .Debug => { ret "DEBUG"; },
+        .Info => { ret "INFO "; },
+        .Warn => { ret "WARN "; },
+        .Error => { ret "ERROR"; },
+        .Fatal => { ret "FATAL"; },
+    }
+    ret "NOTSET";
+}
+
+type Logger = @self;
+
+name: ByteView,
+min_level: Level,
+
+fn new(name: ByteView): Logger {
+    ret Logger { .name = name, .min_level = .Debug };
+}
+
+fn set_level(self: *Logger, level: Level) {
+    self.min_level = level;
+}
+
+fn enabled(self: Logger, level: Level): bool {
+    ret @enum2int(level) >= @enum2int(self.min_level);
+}
+
+// 追加成功返回 true；失败（仅 OOM）返回 false。
+fn push(line: *ByteBuffer, text: []u8): bool {
+    match ByteBuffer.append(line, text) {
+        .Ok => { ret true; },
+        .Err(error) => { ret false; },
+    }
+    ret false;
+}
+
+// 低于 min_level 的日志不写出，直接成功。
+fn write(self: *Logger, level: Level, message: []u8): Status {
+    if !enabled(self.*, level) { ret .Ok; }
+    var line = ByteBuffer.new();
+    defer { ByteBuffer.drop(&line); }
+    if !push(&line, "[") { ret .Err(make(.Overflow)); }
+    if !push(&line, level_name(level)) { ret .Err(make(.Overflow)); }
+    if !push(&line, "] ") { ret .Err(make(.Overflow)); }
+    if !push(&line, ByteView.as_slice(self.name)) { ret .Err(make(.Overflow)); }
+    if !push(&line, ": ") { ret .Err(make(.Overflow)); }
+    if !push(&line, message) { ret .Err(make(.Overflow)); }
+    if !push(&line, "\n") { ret .Err(make(.Overflow)); }
+    ret eprint(ByteBuffer.as_slice(line));
+}
+
+fn debug(self: *Logger, message: []u8): Status {
+    ret write(self, .Debug, message);
+}
+
+fn info(self: *Logger, message: []u8): Status {
+    ret write(self, .Info, message);
+}
+
+fn warn(self: *Logger, message: []u8): Status {
+    ret write(self, .Warn, message);
+}
+
+fn error(self: *Logger, message: []u8): Status {
+    ret write(self, .Error, message);
+}
+
+// 尽力写出后终止；写失败无法挽回，直接退出。
+fn fatal(self: *Logger, message: []u8) {
+    write(self, .Fatal, message);
+    Exit.fail();
+}
+
+// ---- 格式化入口（显式装箱，见 value.spl / fmt.format）----
+
+// 先格式化到临时缓冲；格式错误在写出前返回 Err(.InvalidFormat)。
+fn logf(self: *Logger, level: Level, fmt: []u8, args: []Value): Status {
+    if !enabled(self.*, level) { ret .Ok; }
+    var message = ByteBuffer.new();
+    defer { ByteBuffer.drop(&message); }
+    match format(&message, fmt, args) {
+        .Ok => {},
+        .Err(error) => { ret .Err(error); },
+    }
+    ret write(self, level, ByteBuffer.as_slice(message));
+}
+
+fn debugf(self: *Logger, fmt: []u8, args: []Value): Status {
+    ret logf(self, .Debug, fmt, args);
+}
+
+fn infof(self: *Logger, fmt: []u8, args: []Value): Status {
+    ret logf(self, .Info, fmt, args);
+}
+
+fn warnf(self: *Logger, fmt: []u8, args: []Value): Status {
+    ret logf(self, .Warn, fmt, args);
+}
+
+fn errorf(self: *Logger, fmt: []u8, args: []Value): Status {
+    ret logf(self, .Error, fmt, args);
+}
+
+// 格式化后终止；格式化失败也要退出（避免吞掉致命错误）。
+fn fatalf(self: *Logger, fmt: []u8, args: []Value) {
+    logf(self, .Fatal, fmt, args);
+    Exit.fail();
+}
+
+// ---- 测试 ----
+
+fn expect_name(level: Level, expected: []u8) {
+    @assert(ByteView.eq(ByteView.from_slice(level_name(level)),
+                        ByteView.from_slice(expected)),
+            "level name");
+}
+
+fn expect_ok(result: Status) {
+    match result {
+        .Ok => {},
+        .Err(error) => { @assert(false, "unexpected log error"); },
+    }
+}
+
+fn test_level_names() {
+    expect_name(.Debug, "DEBUG");
+    expect_name(.Info, "INFO ");
+    expect_name(.Warn, "WARN ");
+    expect_name(.Error, "ERROR");
+    expect_name(.Fatal, "FATAL");
+
+    var debug: Level = .Debug;
+    var info: Level = .Info;
+    var warn: Level = .Warn;
+    var err: Level = .Error;
+    var fatal: Level = .Fatal;
+    @assert(@enum2int(fatal) > @enum2int(err), "fatal above error");
+    @assert(@enum2int(err) > @enum2int(warn), "error above warn");
+    @assert(@enum2int(warn) > @enum2int(info), "warn above info");
+    @assert(@enum2int(info) > @enum2int(debug), "info above debug");
+}
+
+fn test_filter_and_write() {
+    var log = new(ByteView.from_slice("splc1"));
+    @assert(enabled(log, .Debug), "default shows debug");
+    @assert(enabled(log, .Fatal), "default shows fatal");
+
+    set_level(&log, .Warn);
+    @assert(!enabled(log, .Debug), "filtered debug");
+    @assert(!enabled(log, .Info), "filtered info");
+    @assert(enabled(log, .Warn), "warn passes");
+    @assert(enabled(log, .Error), "error passes");
+
+    // 被过滤的写入不产生输出，且返回成功。
+    expect_ok(debug(&log, "hidden"));
+    expect_ok(write(&log, .Info, "hidden too"));
+    expect_ok(error(&log, "visible"));
+}
+
+fn test_logf() {
+    var log = new(ByteView.from_slice("splc1"));
+    set_level(&log, .Error);
+
+    var args: [2]Value = .[int(3), str("x")];
+
+    // 被过滤的级别不写出，直接成功。
+    expect_ok(infof(&log, "n={} s={}", args[0..2]));
+    // 通过级别正常写出。
+    expect_ok(errorf(&log, "n={} s={}", args[0..2]));
+
+    // 格式化失败向上传播，且不产出消息。
+    match errorf(&log, "{} {}", args[0..1]) {
+        .Ok => { @assert(false, "expected invalid format"); },
+        .Err(error) => {
+            match error.kind {
+                .InvalidFormat => {},
+                _ => { @assert(false, "expected InvalidFormat"); },
+            }
+        },
+    }
+}
+
+fn main() {
+    test_level_names();
+    test_filter_and_write();
+    test_logf();
+}

stage2/utils/lsp_frame.spl

@@ -1,4 +1,5 @@
 // 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。
 
@@ -6,6 +7,8 @@ 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 .{ .checked_add, .checked_mul } = @import("memory.spl");
+type .{ .to_lower } = @import("ascii.spl");
 
 type FrameResult = enum {
     Ok: usize, // 消耗的字节数
@@ -13,17 +16,12 @@ type FrameResult = enum {
     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; }
+        if to_lower(data[j + i]) != to_lower(word.ptr[i]) { ret false; }
         i = i + 1;
     }
     ret true;
@@ -64,7 +62,16 @@ fn parse_frame(input: ByteView, body_out: *ByteBuffer): FrameResult {
             var n: usize = 0;
             var digits = 0;
             while k < term && data[k] >= '0' && data[k] <= '9' {
-                n = n * 10 + (data[k] - '0') as usize;
+                var digit = (data[k] - '0') as usize;
+                var scaled: usize = 0;
+                match checked_mul(n, 10) {
+                    .Ok(value) => { scaled = value; },
+                    .Err(error) => { ret .Err(error); },
+                }
+                match checked_add(scaled, digit) {
+                    .Ok(value) => { n = value; },
+                    .Err(error) => { ret .Err(error); },
+                }
                 digits = digits + 1;
                 k = k + 1;
             }
@@ -80,9 +87,15 @@ fn parse_frame(input: ByteView, body_out: *ByteBuffer): FrameResult {
     var body_start = term + 4;
     if len < body_start || content_len > len - body_start { ret .Incomplete; }
 
-    ByteBuffer.clear(body_out);
+    // 先写入临时缓冲区，成功后再替换 body_out，失败不产生部分输出。
+    var fresh = ByteBuffer.new();
     var body = data[body_start..body_start + content_len];
-    match ByteBuffer.append(body_out, body) { .Ok => {}, .Err(e) => { ret .Err(e); } }
+    match ByteBuffer.append(&fresh, body) {
+        .Ok => {},
+        .Err(e) => { ByteBuffer.drop(&fresh); ret .Err(e); },
+    }
+    ByteBuffer.drop(body_out);
+    body_out.* = ByteBuffer.take(&fresh);
     ret .Ok(body_start + content_len);
 }
 
@@ -107,11 +120,13 @@ fn expect_body(out: *ByteBuffer, expected: []u8) {
 
 fn test_roundtrip() {
     var wire = ByteBuffer.new();
+    defer { ByteBuffer.drop(&wire); }
     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();
+    defer { ByteBuffer.drop(&body); }
     match parse_frame(ByteView.from_slice(bytes), &body) {
         .Ok(used) => {
             @assert(used == bytes.len, "consumed all");
@@ -120,17 +135,17 @@ fn test_roundtrip() {
         .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();
+    defer { ByteBuffer.drop(&wire); }
     expect_ok(write_frame(&wire, "abcd"));
     expect_ok(write_frame(&wire, "ef"));
     var bytes = ByteBuffer.as_slice(wire);
 
     var body = ByteBuffer.new();
+    defer { ByteBuffer.drop(&body); }
     match parse_frame(ByteView.from_slice(bytes), &body) {
         .Ok(used) => {
             expect_body(&body, "abcd");
@@ -142,12 +157,11 @@ fn test_multiple_and_leftover() {
         },
         _ => { @assert(false, "first frame"); },
     }
-    ByteBuffer.drop(&body);
-    ByteBuffer.drop(&wire);
 }
 
 fn test_incomplete_and_errors() {
     var body = ByteBuffer.new();
+    defer { ByteBuffer.drop(&body); }
 
     // 头未结束
     var h1 = "Content-Length: 3\r\n";
@@ -174,7 +188,6 @@ fn test_incomplete_and_errors() {
         .Ok(used) => { expect_body(&body, "hi"); },
         _ => { @assert(false, "h4"); },
     }
-    ByteBuffer.drop(&body);
 }
 
 fn main() {

stage2/utils/map.spl

@@ -1,4 +1,5 @@
 // Expect retcode = 0
+
 // 开放寻址 + tombstone 哈希表。键和值都按字节复制到表自己的存储。
 // 键类型由 hash_fn / eq_fn 解释：回调收到指向键字节的指针（如 *usize / *ByteView）。
 // 普通赋值只复制句柄（与 Vec 一样），所有权转移必须使用 take。
@@ -7,11 +8,29 @@ 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 .{ .checked_add, .checked_mul, .grow_capacity } = @import("memory.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）。
+// 槽位状态：Empty/Occupied/Deleted（墓碑，仅在扩容重哈希时清除）。
 // 负载因子 70/128。
 
+type SlotState = enum {
+    Empty,
+    Occupied,
+    Deleted,
+}
+
+fn state_code(state: SlotState): u8 {
+    var s: SlotState = state;
+    ret @enum2int(s) as u8;
+}
+
+// 查找/删除结果：显式区分命中与缺失，替代 bool 哨兵。
+type LookupResult = enum {
+    Found,
+    NotFound,
+}
+
 type Map = @self;
 
 keys: Vec,
@@ -64,12 +83,22 @@ fn slot_count(self: Map): usize {
     ret self.states.len;
 }
 
-fn slot_state(self: Map, index: usize): u8 {
-    ret self.states.data[index];
+fn slot_state(self: Map, index: usize): SlotState {
+    var code = self.states.data[index];
+    var occupied: SlotState = .Occupied;
+    var deleted_state: SlotState = .Deleted;
+    if code == state_code(occupied) { ret .Occupied; }
+    if code == state_code(deleted_state) { ret .Deleted; }
+    ret .Empty;
 }
 
-fn slot_occupied(state: u8): bool {
-    ret state == 1;
+fn slot_occupied(state: SlotState): bool {
+    match state {
+        .Occupied => { ret true; },
+        .Empty => { ret false; },
+        .Deleted => { ret false; },
+    }
+    ret false;
 }
 
 fn key_at(self: *Map, index: usize): *u8 {
@@ -85,25 +114,42 @@ fn find_slot(self: *Map, key: *u8): usize {
     if self.states.cap == 0 { ret ~(0 as usize); }
     var mask = self.states.cap - 1;
     var i = self.hash_fn(key) & mask;
-    while self.states.data[i] != 0 {
-        if self.states.data[i] == 1 && self.eq_fn(self.keys.data + i * self.key_size, key) {
+    var empty: SlotState = .Empty;
+    var occupied: SlotState = .Occupied;
+    var probes: usize = 0;
+    while probes < self.states.cap && self.states.data[i] != state_code(empty) {
+        if self.states.data[i] == state_code(occupied) &&
+           self.eq_fn(self.keys.data + i * self.key_size, key) {
             ret i;
         }
         i = (i + 1) & mask;
+        probes = probes + 1;
     }
     ret ~(0 as usize);
 }
 
-// 要求容量非零且键不存在，直接写入第一个空槽（重哈希专用）。
+// 要求容量非零且键不存在。优先复用探测链中的第一个墓碑。
 fn place(self: *Map, key: *u8, val: *u8) {
     var mask = self.states.cap - 1;
     var i = self.hash_fn(key) & mask;
-    while self.states.data[i] != 0 {
+    var empty: SlotState = .Empty;
+    var deleted: SlotState = .Deleted;
+    var occupied: SlotState = .Occupied;
+    var first_deleted = ~(0 as usize);
+    var probes: usize = 0;
+    while probes < self.states.cap && self.states.data[i] != state_code(empty) {
+        if self.states.data[i] == state_code(deleted) && first_deleted == ~(0 as usize) {
+            first_deleted = i;
+        }
         i = (i + 1) & mask;
+        probes = probes + 1;
+    }
+    if first_deleted != ~(0 as usize) {
+        i = first_deleted;
     }
     copy((self.keys.data + i * self.key_size) as *any, key as *any, self.key_size);
     copy((self.vals.data + i * self.val_size) as *any, val as *any, self.val_size);
-    self.states.data[i] = 1;
+    self.states.data[i] = state_code(occupied);
     self.size = self.size + 1;
 }
 
@@ -112,15 +158,7 @@ fn reserve(self: *Map, need: usize): Status {
     if !valid(self.*) { ret .Err(make(.InvalidArgument)); }
     if need <= self.states.cap { ret .Ok; }
 
-    var new_cap = self.states.cap;
-    if new_cap < 8 { new_cap = 8; }
-    while new_cap < need {
-        if new_cap > (~(0 as usize)) / 2 {
-            new_cap = need;
-            break;
-        }
-        new_cap = new_cap * 2;
-    }
+    var new_cap = grow_capacity(self.states.cap, 8, need);
 
     var fresh = self.*;
     fresh.keys = Vec.blank(self.key_size, self.keys.elem_align);
@@ -145,14 +183,16 @@ fn reserve(self: *Map, need: usize): Status {
     fresh.states.len = new_cap;
 
     var i: usize = 0;
+    var empty: SlotState = .Empty;
+    var occupied: SlotState = .Occupied;
     while i < new_cap {
-        fresh.states.data[i] = 0;
+        fresh.states.data[i] = state_code(empty);
         i = i + 1;
     }
 
     i = 0;
     while i < self.states.len {
-        if self.states.data[i] == 1 {
+        if self.states.data[i] == state_code(occupied) {
             place(&fresh, self.keys.data + i * self.key_size,
                   self.vals.data + i * self.val_size);
         }
@@ -181,42 +221,64 @@ fn put(self: *Map, key: *u8, val: *u8): Status {
         ret .Ok;
     }
 
-    if self.states.cap == 0 || (self.size + 1) * 128 / self.states.cap >= 70 {
-        var need = self.states.cap * 2;
-        if self.states.cap == 0 { need = 8; }
+    if self.states.cap == 0 {
+        match reserve(self, 8) {
+            .Ok => {},
+            .Err(error) => { ret .Err(error); },
+        }
+    } else {
+        var next_size: usize = 0;
+        match checked_add(self.size, 1) {
+            .Ok(value) => { next_size = value; },
+            .Err(error) => { ret .Err(error); },
+        }
+        var scaled: usize = 0;
+        match checked_mul(next_size, 128) {
+            .Ok(value) => { scaled = value; },
+            .Err(error) => { ret .Err(error); },
+        }
+        if scaled / self.states.cap >= 70 {
+            var need: usize = 0;
+            match checked_mul(self.states.cap, 2) {
+                .Ok(value) => { need = value; },
+                .Err(error) => { ret .Err(error); },
+            }
             match reserve(self, need) {
                 .Ok => {},
                 .Err(error) => { ret .Err(error); },
             }
         }
+    }
 
     place(self, key, val);
     ret .Ok;
 }
 
-// 找到键则把值复制到 out 并返回 true；否则 out 不变并返回 false。
-fn get(self: *Map, key: *u8, out: *u8): bool {
-    if !valid(self.*) || key == null || out == null { ret false; }
+// 找到键则把值复制到 out 并返回 Found；否则 out 不变并返回 NotFound。
+fn get(self: *Map, key: *u8, out: *u8): LookupResult {
+    if !valid(self.*) || key == null || out == null { ret .NotFound; }
     var i = find_slot(self, key);
-    if i == ~(0 as usize) { ret false; }
+    if i == ~(0 as usize) { ret .NotFound; }
     copy(out as *any, (self.vals.data + i * self.val_size) as *any, self.val_size);
-    ret true;
+    ret .Found;
 }
 
-fn del(self: *Map, key: *u8): bool {
-    if !valid(self.*) || key == null { ret false; }
+fn del(self: *Map, key: *u8): LookupResult {
+    if !valid(self.*) || key == null { ret .NotFound; }
     var i = find_slot(self, key);
-    if i == ~(0 as usize) { ret false; }
-    self.states.data[i] = 2;
+    if i == ~(0 as usize) { ret .NotFound; }
+    var deleted_state: SlotState = .Deleted;
+    self.states.data[i] = state_code(deleted_state);
     self.size = self.size - 1;
-    ret true;
+    ret .Found;
 }
 
 // 清空内容但保留容量。墓碑与负载计数一并复位。
 fn clear(self: *Map) {
     var i: usize = 0;
+    var empty: SlotState = .Empty;
     while i < self.states.len {
-        self.states.data[i] = 0;
+        self.states.data[i] = state_code(empty);
         i = i + 1;
     }
     self.size = 0;
@@ -269,6 +331,22 @@ fn count_occupied(m: *Map): usize {
     ret n;
 }
 
+fn found(m: *Map, key: *u8, out: *u8): bool {
+    match get(m, key, out) {
+        .Found => { ret true; },
+        .NotFound => { ret false; },
+    }
+    ret false;
+}
+
+fn deleted(m: *Map, key: *u8): bool {
+    match del(m, key) {
+        .Found => { ret true; },
+        .NotFound => { ret false; },
+    }
+    ret false;
+}
+
 fn make_int_map(): Map {
     ret new(@sizeof(usize), @alignof(usize), @sizeof(i32), @alignof(i32), hash_usize, eq_usize);
 }
@@ -292,24 +370,26 @@ fn test_put_get_update() {
     @assert(size(m) == 1 && valid(m));
 
     out = -1;
-    @assert(get(&m, &key as *u8, &out as *u8) && out == 70, "insert then get");
+    @assert(found(&m, &key as *u8, &out as *u8) && out == 70, "insert then get");
 
     value = 71;
     expect_ok(put(&m, &key as *u8, &value as *u8));
     @assert(size(m) == 1, "update does not grow size");
     out = -1;
-    @assert(get(&m, &key as *u8, &out as *u8) && out == 71, "update overwrites");
+    @assert(found(&m, &key as *u8, &out as *u8) && out == 71, "update overwrites");
 
     key = 8;
     out = 123;
-    @assert(!get(&m, &key as *u8, &out as *u8) && out == 123, "missing key preserves out");
+    @assert(!found(&m, &key as *u8, &out as *u8) && out == 123, "missing key preserves out");
 
+    // drop 后还有断言，故保持手动释放，不改为 defer。
     drop(&m);
     @assert(size(m) == 0 && capacity(m) == 0, "drop resets map");
 }
 
 fn test_growth_and_delete() {
     var m = make_int_map();
+    defer { drop(&m); }
     var i: i32 = 0;
     while i < 1000 {
         var key: usize = i as usize;
@@ -324,7 +404,7 @@ fn test_growth_and_delete() {
     while i < 1000 {
         var key: usize = i as usize;
         var out: i32 = 0;
-        @assert(get(&m, &key as *u8, &out as *u8), "get after growth");
+        @assert(found(&m, &key as *u8, &out as *u8), "get after growth");
         @assert(out == i * 3 - 7, "value after growth");
         i = i + 1;
     }
@@ -332,7 +412,7 @@ fn test_growth_and_delete() {
     i = 0;
     while i < 500 {
         var key: usize = i as usize;
-        @assert(del(&m, &key as *u8));
+        @assert(deleted(&m, &key as *u8));
         i = i + 1;
     }
     @assert(size(m) == 500 && count_occupied(&m) == 500, "deleted slots not occupied");
@@ -341,7 +421,7 @@ fn test_growth_and_delete() {
     while i < 500 {
         var key: usize = i as usize;
         var out: i32 = 0;
-        @assert(!get(&m, &key as *u8, &out as *u8), "deleted key missing");
+        @assert(!found(&m, &key as *u8, &out as *u8), "deleted key missing");
         i = i + 1;
     }
 
@@ -351,41 +431,40 @@ fn test_growth_and_delete() {
     expect_ok(put(&m, &key as *u8, &value as *u8));
     @assert(size(m) == 501);
     var out: i32 = 0;
-    @assert(get(&m, &key as *u8, &out as *u8) && out == 99, "reuse tombstone");
+    @assert(found(&m, &key as *u8, &out as *u8) && out == 99, "reuse tombstone");
 
-    @assert(del(&m, &key as *u8), "delete key");
+    @assert(deleted(&m, &key as *u8), "delete key");
     @assert(size(m) == 500);
-    @assert(!get(&m, &key as *u8, &out as *u8));
+    @assert(!found(&m, &key as *u8, &out as *u8));
 
     clear(&m);
     @assert(size(m) == 0 && count_occupied(&m) == 0, "clear empties");
-
-    drop(&m);
 }
 
 fn test_take_and_self_put() {
     var m = make_int_map();
+    defer { drop(&m); }
     var key: usize = 5;
     var value: i32 = 55;
     expect_ok(put(&m, &key as *u8, &value as *u8));
 
     var moved = take(&m);
+    defer { drop(&moved); }
     @assert(size(m) == 0 && capacity(m) == 0, "take leaves source empty");
     var out: i32 = 0;
-    @assert(get(&moved, &key as *u8, &out as *u8) && out == 55, "moved keeps content");
+    @assert(found(&moved, &key as *u8, &out as *u8) && out == 55, "moved keeps content");
 
     // 值指针可以指向表内已有的值（按字节移动，允许重叠）。
     var alias: i32 = 0;
-    @assert(get(&moved, &key as *u8, &alias as *u8));
+    @assert(found(&moved, &key as *u8, &alias as *u8));
     expect_ok(put(&moved, &key as *u8, &alias as *u8));
-    @assert(get(&moved, &key as *u8, &out as *u8) && out == 55, "self value put");
-
-    drop(&moved);
+    @assert(found(&moved, &key as *u8, &out as *u8) && out == 55, "self value put");
 }
 
 fn test_byte_view_keys() {
     var m = new(@sizeof(ByteView), @alignof(ByteView), @sizeof(i32), @alignof(i32),
                 hash_view, eq_view);
+    defer { drop(&m); }
 
     var alpha = ByteView.from_slice("alpha");
     var beta = ByteView.from_slice("beta");
@@ -400,21 +479,19 @@ fn test_byte_view_keys() {
     @assert(size(m) == 3);
 
     var out: i32 = 0;
-    @assert(get(&m, &alpha as *u8, &out as *u8) && out == 1, "alpha");
-    @assert(get(&m, &beta as *u8, &out as *u8) && out == 2, "beta");
-    @assert(get(&m, &gamma as *u8, &out as *u8) && out == 3, "gamma");
+    @assert(found(&m, &alpha as *u8, &out as *u8) && out == 1, "alpha");
+    @assert(found(&m, &beta as *u8, &out as *u8) && out == 2, "beta");
+    @assert(found(&m, &gamma as *u8, &out as *u8) && out == 3, "gamma");
 
     // 内容相同的不同 ByteView 视为同一个键。
     var alpha2 = ByteView.from_slice("alpha");
     out = 0;
-    @assert(get(&m, &alpha2 as *u8, &out as *u8) && out == 1, "equal content key");
+    @assert(found(&m, &alpha2 as *u8, &out as *u8) && out == 1, "equal content key");
     @assert(size(m) == 3, "equal content does not insert");
 
-    @assert(del(&m, &beta as *u8), "delete beta");
+    @assert(deleted(&m, &beta as *u8), "delete beta");
     @assert(size(m) == 2);
-    @assert(!get(&m, &beta as *u8, &out as *u8));
-
-    drop(&m);
+    @assert(!found(&m, &beta as *u8, &out as *u8));
 }
 
 fn main() {

stage2/utils/memory.spl

@@ -1,4 +1,5 @@
 // Expect retcode = 0
+
 // 尺寸计算和字节移动；不分配、不释放内存，不改变任何资源的所有权。
 
 type .{ .Error, .Status, .make } = @import("error.spl");
@@ -31,6 +32,22 @@ fn align_up(size: usize, alignment: usize): SizeResult {
     ret .Ok((size + mask) & ~mask);
 }
 
+// 计算至少 need 的容量：从 current 起（不低于 min_cap）按倍增增长；
+// 无法再倍增（会溢出）时直接取 need。need 可表示时不溢出。
+fn grow_capacity(current: usize, min_cap: usize, need: usize): usize {
+    var cap = current;
+    if cap < min_cap { cap = min_cap; }
+    var limit: usize = ~(0 as usize);
+    while cap < need {
+        if cap > limit / 2 {
+            cap = need;
+            break;
+        }
+        cap = cap * 2;
+    }
+    ret cap;
+}
+
 // 将 src.len 个字节复制到 dst 的开头，允许两个切片任意重叠。
 // 零长度源不访问内存，允许空指针。所有可恢复错误均在首次写入前返回。
 // 非空源要求可读，目标前 src.len 字节要求可写，生命周期由调用方保证。
@@ -164,6 +181,13 @@ fn test_sizes() {
     expect_size(align_up(1, highest_power), highest_power);
     expect_size(align_up(highest_power, highest_power), highest_power);
     expect_overflow(align_up(highest_power + 1, highest_power));
+
+    @assert(grow_capacity(0, 8, 0) == 8, "grow respects minimum");
+    @assert(grow_capacity(0, 8, 5) == 8, "grow minimum covers need");
+    @assert(grow_capacity(8, 8, 9) == 16, "grow doubles");
+    @assert(grow_capacity(8, 8, 100) == 128, "grow repeats");
+    @assert(grow_capacity(limit / 2, 8, limit) == limit, "grow exact to limit");
+    @assert(grow_capacity(limit / 2 + 1, 8, limit) == limit, "grow clamps to need");
 }
 
 // 穷举同一 8 字节数组内的所有合法源、目标和长度组合。

stage2/utils/path.spl

@@ -1,231 +1,400 @@
 // 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");
+// 纯词法路径工具，不访问文件系统。调用方必须显式选择 POSIX 或 Windows 规则；
+// 输入视图均为借用，拥有型结果写入 ByteBuffer，失败保持原输出不变。
+type .{ .Status, .make } = @import("error.spl");
 type .{ .ByteView } = @import("byte_view.spl");
 type .{ .ByteBuffer } = @import("byte_buffer.spl");
 type .{ .Vec } = @import("vec.spl");
+type .{ .is_alpha } = @import("ascii.spl");
 
-fn is_sep(c: u8): bool {
-    ret c == '/' || c == '\\'; // '/' '\\'
+type PathStyle = enum { Posix, Windows }
+
+fn is_windows(style: PathStyle): bool {
+    match style { .Posix => { ret false; }, .Windows => { ret true; }, }
+    ret false;
 }
 
-fn is_drive(path: ByteView): bool {
-    ret path.len >= 2 && path.ptr[1] == ':'; // ':'
+fn separator(style: PathStyle): u8 {
+    if is_windows(style) { ret '\\'; }
+    ret '/';
 }
 
-fn is_absolute(path: ByteView): bool {
+fn is_sep(style: PathStyle, byte: u8): bool {
+    if is_windows(style) { ret byte == '/' || byte == '\\'; }
+    ret byte == '/';
+}
+
+fn is_drive_prefix(path: ByteView): bool {
+    ret path.len >= 2 && is_alpha(path.ptr[0]) && path.ptr[1] == ':';
+}
+
+fn is_unc(path: ByteView, style: PathStyle): bool {
+    ret is_windows(style) && path.len >= 2 &&
+        is_sep(style, path.ptr[0]) && is_sep(style, path.ptr[1]);
+}
+
+fn is_absolute(path: ByteView, style: PathStyle): bool {
     if path.len == 0 { ret false; }
-    if is_sep(path.ptr[0]) { ret true; }
-    ret is_drive(path);
+    if !is_windows(style) { ret path.ptr[0] == '/'; }
+    if is_unc(path, style) { ret true; }
+    if is_drive_prefix(path) {
+        ret path.len >= 3 && is_sep(style, path.ptr[2]);
+    }
+    // 根相对路径以当前盘符为根，本模块将其视为绝对。
+    ret is_sep(style, path.ptr[0]);
 }
 
-// 最后一个分隔符下标；无返回 ~0。
-fn last_sep(path: ByteView): usize {
+// UNC 根长度不含共享名后的分隔符；不完整时返回 0。
+fn unc_root_len(path: ByteView, style: PathStyle): usize {
+    if !is_unc(path, style) { ret 0; }
+    var i: usize = 2;
+    while i < path.len && is_sep(style, path.ptr[i]) { i = i + 1; }
+    var server = i;
+    while i < path.len && !is_sep(style, path.ptr[i]) { i = i + 1; }
+    if i == server { ret 0; }
+    while i < path.len && is_sep(style, path.ptr[i]) { i = i + 1; }
+    var share = i;
+    while i < path.len && !is_sep(style, path.ptr[i]) { i = i + 1; }
+    if i == share { ret 0; }
+    ret i;
+}
+
+fn root_len(path: ByteView, style: PathStyle): usize {
+    if path.len == 0 { ret 0; }
+    if !is_windows(style) {
+        if path.ptr[0] == '/' { ret 1; }
+        ret 0;
+    }
+    var unc = unc_root_len(path, style);
+    if unc != 0 { ret unc; }
+    if is_drive_prefix(path) {
+        if path.len >= 3 && is_sep(style, path.ptr[2]) { ret 3; }
+        ret 2;
+    }
+    if is_sep(style, path.ptr[0]) { ret 1; }
+    ret 0;
+}
+
+fn last_sep(path: ByteView, style: PathStyle): usize {
     var i = path.len;
     while i > 0 {
         i = i - 1;
-        if is_sep(path.ptr[i]) { ret i; }
+        if is_sep(style, 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; },
+fn slice_or_empty(path: ByteView, start: usize, end: usize): ByteView {
+    match ByteView.slice(path, start, end) {
+        .Ok(value) => { ret value; },
         .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(); },
+fn dirname(path: ByteView, style: PathStyle): ByteView {
+    var root = root_len(path, style);
+    if root != 0 && path.len == root { ret path; }
+    var index = last_sep(path, style);
+    if index == ~(0 as usize) {
+        if is_windows(style) && is_drive_prefix(path) {
+            ret slice_or_empty(path, 0, 2);
         }
         ret ByteView.empty();
+    }
+    if root != 0 && index <= root { ret slice_or_empty(path, 0, root); }
+    if index == 0 { ret slice_or_empty(path, 0, 1); }
+    ret slice_or_empty(path, 0, index);
 }
 
-// 扩展名（不含 '.'）；无扩展名返回空。
-fn extension(path: ByteView): ByteView {
-    var base = basename(path);
+fn basename(path: ByteView, style: PathStyle): ByteView {
+    var root = root_len(path, style);
+    if root != 0 && path.len == root { ret ByteView.empty(); }
+    var index = last_sep(path, style);
+    if index == ~(0 as usize) {
+        if is_windows(style) && is_drive_prefix(path) {
+            ret slice_or_empty(path, 2, path.len);
+        }
+        ret path;
+    }
+    ret slice_or_empty(path, index + 1, path.len);
+}
+
+fn extension(path: ByteView, style: PathStyle): ByteView {
+    var base = basename(path, style);
     var i = base.len;
     while i > 0 {
         i = i - 1;
-        if base.ptr[i] == '.' { // '.'
+        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 slice_or_empty(base, i + 1, base.len);
         }
     }
     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));
+// Windows 盘符相对路径依赖进程的盘符工作目录，join 明确拒绝该歧义。
+fn join_into(out: *ByteBuffer, dir: ByteView,
+             rel: ByteView, style: PathStyle): Status {
+    if out == null { ret .Err(make(.InvalidArgument)); }
+    var fresh = ByteBuffer.new();
+    if is_absolute(rel, style) {
+        match ByteBuffer.append(&fresh, ByteView.as_slice(rel)) {
+            .Ok => {}, .Err(error) => { ByteBuffer.drop(&fresh); ret .Err(error); },
         }
-    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); } }
+        ByteBuffer.drop(out);
+        out.* = ByteBuffer.take(&fresh);
+        ret .Ok;
     }
-    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
+    if is_windows(style) && is_drive_prefix(rel) {
+        ByteBuffer.drop(&fresh);
+        ret .Err(make(.InvalidArgument));
     }
-
-    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); },
+    match ByteBuffer.append(&fresh, ByteView.as_slice(dir)) {
+        .Ok => {}, .Err(error) => { ByteBuffer.drop(&fresh); 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); },
+    var drive_relative_base = is_windows(style) && dir.len == 2 && is_drive_prefix(dir);
+    if dir.len != 0 && !is_sep(style, dir.ptr[dir.len - 1]) && rel.len != 0 &&
+       !drive_relative_base {
+        match ByteBuffer.append_byte(&fresh, separator(style)) {
+            .Ok => {}, .Err(error) => { ByteBuffer.drop(&fresh); ret .Err(error); },
         }
     }
-
-    Vec.drop(&starts);
+    match ByteBuffer.append(&fresh, ByteView.as_slice(rel)) {
+        .Ok => {}, .Err(error) => { ByteBuffer.drop(&fresh); ret .Err(error); },
+    }
+    ByteBuffer.drop(out);
+    out.* = ByteBuffer.take(&fresh);
     ret .Ok;
 }
 
-// ---- 测试辅助 ----
+fn segment_is_parent(buffer: ByteBuffer, rollback: usize, style: PathStyle): bool {
+    var start = rollback;
+    if start < buffer.storage.len && is_sep(style, buffer.storage.data[start]) {
+        start = start + 1;
+    }
+    ret buffer.storage.len - start == 2 &&
+        buffer.storage.data[start] == '.' && buffer.storage.data[start + 1] == '.';
+}
 
-fn expect_ok(result: Status) {
-    match result {
+fn append_segment(out: *ByteBuffer, starts: *Vec,
+                  segment: ByteView, style: PathStyle): Status {
+    var rollback = out.storage.len;
+    if rollback != 0 && !is_sep(style, out.storage.data[rollback - 1]) &&
+       !(is_windows(style) && out.storage.data[rollback - 1] == ':') {
+        match ByteBuffer.append_byte(out, separator(style)) {
+            .Ok => {}, .Err(error) => { ret .Err(error); },
+        }
+    }
+    match Vec.push(starts, &rollback as *u8) {
         .Ok => {},
-        .Err(error) => { @assert(false, "unexpected path error"); },
+        .Err(error) => { out.storage.len = rollback; ret .Err(error); },
+    }
+    match ByteBuffer.append(out, ByteView.as_slice(segment)) {
+        .Ok => {},
+        .Err(error) => {
+            starts.len = starts.len - 1;
+            out.storage.len = rollback;
+            ret .Err(error);
+        },
+    }
+    ret .Ok;
+}
+
+// 规范化分隔符并折叠 '.'/'..'。相对路径无法消去的前导 '..' 会被保留；
+// 绝对路径的 '..' 不会越过根，UNC 的 server/share 也不可被消去。
+fn normalize_into(out: *ByteBuffer, path: ByteView, style: PathStyle): Status {
+    if out == null { ret .Err(make(.InvalidArgument)); }
+    var fresh = ByteBuffer.new();
+    var starts = Vec.blank(@sizeof(usize), @alignof(usize));
+    var i: usize = 0;
+    var absolute = false;
+    var protected: usize = 0;
+
+    if path.len != 0 {
+        if !is_windows(style) && path.ptr[0] == '/' {
+            absolute = true;
+            match ByteBuffer.append_byte(&fresh, '/') {
+                .Ok => {}, .Err(error) => { Vec.drop(&starts); ByteBuffer.drop(&fresh); ret .Err(error); },
+            }
+            while i < path.len && is_sep(style, path.ptr[i]) { i = i + 1; }
+        } else if is_unc(path, style) {
+            absolute = true;
+            protected = 2;
+            match ByteBuffer.append(&fresh, "\\\\") {
+                .Ok => {}, .Err(error) => { Vec.drop(&starts); ByteBuffer.drop(&fresh); ret .Err(error); },
+            }
+            i = 2;
+            while i < path.len && is_sep(style, path.ptr[i]) { i = i + 1; }
+        } else if is_windows(style) && is_drive_prefix(path) {
+            match ByteBuffer.append(&fresh, path.ptr[0..2]) {
+                .Ok => {}, .Err(error) => { Vec.drop(&starts); ByteBuffer.drop(&fresh); ret .Err(error); },
+            }
+            i = 2;
+            if i < path.len && is_sep(style, path.ptr[i]) {
+                absolute = true;
+                match ByteBuffer.append_byte(&fresh, '\\') {
+                    .Ok => {}, .Err(error) => { Vec.drop(&starts); ByteBuffer.drop(&fresh); ret .Err(error); },
+                }
+                while i < path.len && is_sep(style, path.ptr[i]) { i = i + 1; }
+            }
+        } else if is_windows(style) && is_sep(style, path.ptr[0]) {
+            absolute = true;
+            match ByteBuffer.append_byte(&fresh, '\\') {
+                .Ok => {}, .Err(error) => { Vec.drop(&starts); ByteBuffer.drop(&fresh); ret .Err(error); },
+            }
+            while i < path.len && is_sep(style, path.ptr[i]) { i = i + 1; }
+        }
+    }
+
+    while i < path.len {
+        while i < path.len && is_sep(style, path.ptr[i]) { i = i + 1; }
+        if i >= path.len { break; }
+        var begin = i;
+        while i < path.len && !is_sep(style, path.ptr[i]) { i = i + 1; }
+        var segment = slice_or_empty(path, begin, i);
+
+        if ByteView.eq(segment, ByteView.from_slice(".")) {
+            if protected != 0 && starts.len < protected {
+                Vec.drop(&starts); ByteBuffer.drop(&fresh);
+                ret .Err(make(.InvalidArgument));
+            }
+            continue;
+        }
+        if ByteView.eq(segment, ByteView.from_slice("..")) {
+            if protected != 0 && starts.len < protected {
+                Vec.drop(&starts); ByteBuffer.drop(&fresh);
+                ret .Err(make(.InvalidArgument));
+            }
+            if starts.len > protected {
+                var last: usize = 0;
+                match Vec.get(&starts, starts.len - 1) {
+                    .Ok(ptr) => { last = (ptr as *usize).*; },
+                    .Err(error) => { Vec.drop(&starts); ByteBuffer.drop(&fresh); ret .Err(error); },
+                }
+                if !segment_is_parent(fresh, last, style) {
+                    fresh.storage.len = last;
+                    starts.len = starts.len - 1;
+                    continue;
+                }
+            }
+            if absolute { continue; }
+        }
+
+        match append_segment(&fresh, &starts, segment, style) {
+            .Ok => {}, .Err(error) => { Vec.drop(&starts); ByteBuffer.drop(&fresh); ret .Err(error); },
+        }
+    }
+
+    if protected != 0 && starts.len < protected {
+        Vec.drop(&starts); ByteBuffer.drop(&fresh);
+        ret .Err(make(.InvalidArgument));
+    }
+    Vec.drop(&starts);
+    ByteBuffer.drop(out);
+    out.* = ByteBuffer.take(&fresh);
+    ret .Ok;
+}
+
+// ---- 测试 ----
+fn expect_ok(result: Status) {
+    match result { .Ok => {}, .Err(error) => { @assert(false, "unexpected path error"); }, }
+}
+
+fn expect_invalid(result: Status) {
+    match result {
+        .Ok => { @assert(false, "expected invalid path"); },
+        .Err(error) => {
+            match error.kind { .InvalidArgument => {}, _ => { @assert(false); }, }
+        },
     }
 }
 
-fn view(text: []u8): ByteView {
-    ret ByteView.from_slice(text);
-}
+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");
+    @assert(ByteView.eq(got, view(expected)), "path view");
 }
 
 fn expect_out(out: *ByteBuffer, expected: []u8) {
     var actual = ByteBuffer.as_slice(out.*);
-    @assert(actual.len == expected.len, "out length");
+    @assert(actual.len == expected.len, "path output length");
     var i: usize = 0;
     while i < expected.len {
-        @assert(actual.ptr[i] == expected.ptr[i], "out byte");
+        @assert(actual.ptr[i] == expected.ptr[i], "path output 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_style_and_parts() {
+    @assert(is_absolute(view("/a"), .Posix));
+    @assert(!is_absolute(view("\\a"), .Posix));
+    @assert(is_absolute(view("C:\\a"), .Windows));
+    @assert(!is_absolute(view("C:a"), .Windows));
+    @assert(is_absolute(view("\\\\server\\share\\a"), .Windows));
+    @assert(!is_absolute(view("C:/a"), .Posix));
+    expect_view(dirname(view("/a/b"), .Posix), "/a");
+    expect_view(dirname(view("/a"), .Posix), "/");
+    expect_view(basename(view("/a/b"), .Posix), "b");
+    expect_view(extension(view("a/b.spl"), .Posix), "spl");
+    expect_view(extension(view("a/.hidden"), .Posix), "");
+    expect_view(dirname(view("C:\\a"), .Windows), "C:\\");
+    expect_view(basename(view("C:\\a"), .Windows), "a");
+    expect_view(dirname(view("C:relative"), .Windows), "C:");
+    expect_view(basename(view("C:relative"), .Windows), "relative");
+    expect_view(dirname(view("\\\\server\\share\\file"), .Windows),
+                "\\\\server\\share");
 }
 
 fn test_join() {
     var out = ByteBuffer.new();
-    expect_ok(join_into(&out, view("stage2/utils"), view("vec.spl")));
+    defer { ByteBuffer.drop(&out); }
+    expect_ok(join_into(&out, view("stage2/utils"), view("vec.spl"), .Posix));
     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);
+    expect_ok(join_into(&out, view("C:\\base"), view("child\\x"), .Windows));
+    expect_out(&out, "C:\\base\\child\\x");
+    expect_ok(join_into(&out, view("C:"), view("child"), .Windows));
+    expect_out(&out, "C:child");
+    expect_ok(join_into(&out, view("ignored"), view("D:/abs"), .Windows));
+    expect_out(&out, "D:/abs");
+    expect_invalid(join_into(&out, view("C:\\base"), view("D:relative"), .Windows));
+    expect_out(&out, "D:/abs");
 }
 
-fn test_normalize() {
+fn test_normalize_posix() {
     var out = ByteBuffer.new();
-    expect_ok(normalize_into(&out, view("a/b/../c")));
+    defer { ByteBuffer.drop(&out); }
+    expect_ok(normalize_into(&out, view("a/b/../c"), .Posix));
     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);
+    expect_ok(normalize_into(&out, view("../../a/./b"), .Posix));
+    expect_out(&out, "../../a/b");
+    expect_ok(normalize_into(&out, view("/../../a//b"), .Posix));
+    expect_out(&out, "/a/b");
+    expect_ok(normalize_into(&out, view("a\\b/../c"), .Posix));
+    expect_out(&out, "c");
+}
+
+fn test_normalize_windows() {
+    var out = ByteBuffer.new();
+    defer { ByteBuffer.drop(&out); }
+    expect_ok(normalize_into(&out, view("C:/a\\b/../c"), .Windows));
+    expect_out(&out, "C:\\a\\c");
+    expect_ok(normalize_into(&out, view("C:foo\\..\\bar"), .Windows));
+    expect_out(&out, "C:bar");
+    expect_ok(normalize_into(&out, view("\\root\\..\\file"), .Windows));
+    expect_out(&out, "\\file");
+    expect_ok(normalize_into(&out,
+                             view("\\\\server\\share\\a\\..\\file"), .Windows));
+    expect_out(&out, "\\\\server\\share\\file");
+    expect_ok(ByteBuffer.assign(&out, "preserved"));
+    expect_invalid(normalize_into(&out, view("\\\\server\\..\\file"), .Windows));
+    expect_out(&out, "preserved");
 }
 
 fn main() {
-    test_parts();
+    test_style_and_parts();
     test_join();
-    test_normalize();
+    test_normalize_posix();
+    test_normalize_windows();
 }

stage2/utils/set.spl

@@ -0,0 +1,235 @@
+// Expect retcode = 0
+
+// 开放寻址集合。键按字节复制到集合自己的存储中；集合只释放自身存储，
+// 不释放键值内部引用的资源。哈希与相等回调的约定和 Map 一致。
+type .{ .Status } = @import("error.spl");
+type .{ .Map, .LookupResult } = @import("map.spl");
+type .{ .ByteView } = @import("byte_view.spl");
+type .{ .hash_usize, .eq_usize, .hash_view, .eq_view } = @import("hash.spl");
+
+type Set = @self;
+
+map: Map,
+
+fn new(key_size: usize, key_align: usize,
+       hash_fn: fn(*u8): usize, eq_fn: fn(*u8, *u8): bool): Set {
+    ret Set {
+        .map = Map.new(key_size, key_align, 1, 1, hash_fn, eq_fn),
+    };
+}
+
+fn valid(self: Set): bool {
+    ret Map.valid(self.map) && self.map.val_size == 1;
+}
+
+fn size(self: Set): usize {
+    ret self.map.size;
+}
+
+fn capacity(self: Set): usize {
+    ret self.map.states.cap;
+}
+
+fn reserve(self: *Set, need: usize): Status {
+    ret Map.reserve(&self.map, need);
+}
+
+// 重复插入保持集合大小不变。
+fn insert(self: *Set, key: *u8): Status {
+    var marker: u8 = 1;
+    ret Map.put(&self.map, key, &marker);
+}
+
+fn find(self: *Set, key: *u8): LookupResult {
+    var marker: u8 = 0;
+    ret Map.get(&self.map, key, &marker);
+}
+
+fn remove(self: *Set, key: *u8): LookupResult {
+    ret Map.del(&self.map, key);
+}
+
+// 遍历 [0, slot_count)；仅 slot_occupied 为真时 key_at 有效。
+fn slot_count(self: Set): usize {
+    ret Map.slot_count(self.map);
+}
+
+fn slot_occupied(self: Set, index: usize): bool {
+    ret Map.slot_occupied(Map.slot_state(self.map, index));
+}
+
+fn key_at(self: *Set, index: usize): *u8 {
+    ret Map.key_at(&self.map, index);
+}
+
+fn clear(self: *Set) {
+    Map.clear(&self.map);
+}
+
+fn take(source: *Set): Set {
+    ret Set { .map = Map.take(&source.map) };
+}
+
+fn drop(self: *Set) {
+    Map.drop(&self.map);
+}
+
+// ---- 测试 ----
+
+fn expect_ok(result: Status) {
+    match result {
+        .Ok => {},
+        .Err(error) => { @assert(false, "unexpected Set error"); },
+    }
+}
+
+fn found(set: *Set, key: *u8): bool {
+    match find(set, key) {
+        .Found => { ret true; },
+        .NotFound => { ret false; },
+    }
+    ret false;
+}
+
+fn removed(set: *Set, key: *u8): bool {
+    match remove(set, key) {
+        .Found => { ret true; },
+        .NotFound => { ret false; },
+    }
+    ret false;
+}
+
+fn constant_hash(key: *u8): usize {
+    ret 3;
+}
+
+fn test_insert_find_remove() {
+    var set = new(@sizeof(usize), @alignof(usize), hash_usize, eq_usize);
+    defer { drop(&set); }
+    @assert(valid(set));
+    @assert(size(set) == 0 && capacity(set) == 0);
+
+    var i: usize = 0;
+    while i < 1000 {
+        expect_ok(insert(&set, &i as *u8));
+        i = i + 1;
+    }
+    @assert(size(set) == 1000, "all unique keys inserted");
+
+    i = 0;
+    while i < 1000 {
+        @assert(found(&set, &i as *u8), "key survives growth");
+        i = i + 1;
+    }
+
+    // 重复键不增加 size。
+    i = 0;
+    while i < 1000 {
+        expect_ok(insert(&set, &i as *u8));
+        i = i + 1;
+    }
+    @assert(size(set) == 1000, "duplicate insertion is idempotent");
+
+    i = 0;
+    while i < 1000 {
+        if i % 2 == 0 { @assert(removed(&set, &i as *u8)); }
+        i = i + 1;
+    }
+    @assert(size(set) == 500);
+
+    i = 0;
+    while i < 1000 {
+        @assert(found(&set, &i as *u8) == (i % 2 != 0), "removed membership");
+        i = i + 1;
+    }
+
+    i = 0;
+    while i < 1000 {
+        if i % 2 == 0 { expect_ok(insert(&set, &i as *u8)); }
+        i = i + 1;
+    }
+    @assert(size(set) == 1000, "removed keys can be reinserted");
+
+    var occupied: usize = 0;
+    i = 0;
+    while i < slot_count(set) {
+        if slot_occupied(set, i) {
+            var key = (key_at(&set, i) as *usize).*;
+            @assert(key < 1000, "iteration yields stored key");
+            occupied = occupied + 1;
+        }
+        i = i + 1;
+    }
+    @assert(occupied == size(set), "iteration visits each member once");
+
+    var missing: usize = 2000;
+    @assert(!removed(&set, &missing as *u8));
+    clear(&set);
+    @assert(size(set) == 0 && capacity(set) != 0, "clear retains allocation");
+}
+
+fn test_collisions_and_take() {
+    var set = new(@sizeof(usize), @alignof(usize), constant_hash, eq_usize);
+    defer { drop(&set); }
+
+    var i: usize = 0;
+    while i < 64 {
+        expect_ok(insert(&set, &i as *u8));
+        i = i + 1;
+    }
+    i = 0;
+    while i < 64 {
+        @assert(found(&set, &i as *u8), "collision chain lookup");
+        i = i + 1;
+    }
+
+    // 删除整条碰撞链后重新插入，验证墓碑会被复用且探测会终止。
+    i = 0;
+    while i < 64 {
+        @assert(removed(&set, &i as *u8));
+        i = i + 1;
+    }
+    @assert(size(set) == 0);
+    i = 64;
+    while i < 128 {
+        expect_ok(insert(&set, &i as *u8));
+        i = i + 1;
+    }
+    i = 64;
+    while i < 128 {
+        @assert(found(&set, &i as *u8), "tombstone collision chain reuse");
+        i = i + 1;
+    }
+
+    var moved = take(&set);
+    defer { drop(&moved); }
+    @assert(size(set) == 0 && capacity(set) == 0, "take clears source");
+    @assert(size(moved) == 64, "take preserves members");
+    i = 64;
+    while i < 128 {
+        @assert(found(&moved, &i as *u8));
+        i = i + 1;
+    }
+}
+
+fn test_byte_view_keys() {
+    var set = new(@sizeof(ByteView), @alignof(ByteView), hash_view, eq_view);
+    defer { drop(&set); }
+
+    var alpha = ByteView.from_slice("alpha");
+    var alpha_copy = ByteView.from_slice("alpha");
+    var beta = ByteView.from_slice("beta");
+    expect_ok(insert(&set, &alpha as *u8));
+    expect_ok(insert(&set, &alpha_copy as *u8));
+    expect_ok(insert(&set, &beta as *u8));
+    @assert(size(set) == 2, "ByteView keys compare by content");
+    @assert(found(&set, &alpha_copy as *u8));
+    @assert(removed(&set, &alpha_copy as *u8));
+    @assert(!found(&set, &alpha as *u8));
+}
+
+fn main() {
+    test_insert_find_remove();
+    test_collisions_and_take();
+    test_byte_view_keys();
+}

stage2/utils/source.spl

@@ -173,7 +173,10 @@ fn set_at(set: *Sources, index: usize): *Source {
 
 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); }
+    match Map.get(&set.by_name, &fname as *u8, &out as *u8) {
+        .Found => { ret .Found(out); },
+        .NotFound => {},
+    }
     ret .NotFound;
 }
 

stage2/utils/str.spl

@@ -138,6 +138,7 @@ fn expect_text(s: *Str, expected: []u8) {
 
 fn test_construction() {
     var s = new();
+    defer { drop(&s); }
     @assert(is_empty(s) && size(s) == 0, "empty new");
     expect_ok(append(&s, "hello"));
     expect_ok(append_byte(&s, ' '));
@@ -154,18 +155,19 @@ fn test_construction() {
         },
         .Err(error) => { @assert(false, "from_slice"); },
     }
-    drop(&s);
 }
 
 fn test_compare_and_search() {
     var hello_r = from_slice("hello");
     var hello: Str = new();
+    defer { drop(&hello); }
     match hello_r {
         .Ok(v) => { hello = v; },
         .Err(error) => { @assert(false, "hello"); },
     }
     var help_r = from_slice("help");
     var help: Str = new();
+    defer { drop(&help); }
     match help_r {
         .Ok(v) => { help = v; },
         .Err(error) => { @assert(false, "help"); },
@@ -181,12 +183,14 @@ fn test_compare_and_search() {
     @assert(ends_with(hello, hello), "ends_with");
     var pre_r = from_slice("he");
     var pre: Str = new();
+    defer { drop(&pre); }
     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();
+    defer { drop(&needle); }
     match needle_r { .Ok(v) => { needle = v; }, .Err(e) => { @assert(false, "needle"); } }
     match find(hello, needle) {
         .Found(index) => { @assert(index == 2, "find ll"); },
@@ -196,18 +200,15 @@ fn test_compare_and_search() {
         .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();
+    defer { drop(&source); }
     expect_ok(append(&source, "owned"));
 
     var copy = new();
+    defer { drop(&copy); }
     expect_ok(clone(&copy, source));
     expect_text(&copy, "owned");
     expect_ok(append_byte(&source, '!')); // '!'
@@ -219,16 +220,13 @@ fn test_clone_take_and_failures() {
     expect_text(&copy, "owned");
 
     var moved = take(&copy);
+    defer { drop(&moved); }
     @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() {

stage2/utils/utf8.spl

@@ -1,60 +1,70 @@
 // 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;
+// 解码结果：Ok 携带消耗字节数；明确区分越界/截断/非法，避免 0 哨兵歧义。
+type DecodeResult = enum {
+    Ok: usize,
+    Eof,       // index 已越过输入
+    Truncated, // 序列尾部字节不足
+    Invalid,   // 续字节/过长编码/代理区/超出码点范围
 }
 
-// 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; }
+// 从 index 解码一个码点到 out_cp；返回 DecodeResult。
+fn decode(data: *u8, len: usize, index: usize, out_cp: *u32): DecodeResult {
+    if index >= len { ret .Eof; }
     var b0 = data[index];
 
     if b0 < 0x80 {
         out_cp.* = b0 as u32;
-        ret 1;
+        ret .Ok(1);
     }
-    if b0 < 0xC2 { ret 0; } // 续字节或过长编码的前导
+    if b0 < 0xC2 { ret .Invalid; } // 续字节或过长编码的前导
 
     if b0 < 0xE0 {
-        if index + 1 >= len { ret 0; }
+        if index + 1 >= len { ret .Truncated; }
         var b1 = data[index + 1];
-        if (b1 & 0xC0) != 0x80 { ret 0; }
+        if (b1 & 0xC0) != 0x80 { ret .Invalid; }
         out_cp.* = (((b0 & 0x1F) as u32) << 6) | ((b1 & 0x3F) as u32);
-        ret 2;
+        ret .Ok(2);
     }
     if b0 < 0xF0 {
-        if index + 2 >= len { ret 0; }
+        if index + 2 >= len { ret .Truncated; }
         var b1 = data[index + 1];
         var b2 = data[index + 2];
-        if (b1 & 0xC0) != 0x80 || (b2 & 0xC0) != 0x80 { ret 0; }
+        if (b1 & 0xC0) != 0x80 || (b2 & 0xC0) != 0x80 { ret .Invalid; }
         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
+        if cp < 0x800 { ret .Invalid; }             // 过长
+        if cp >= 0xD800 && cp <= 0xDFFF { ret .Invalid; } // 代理区
         out_cp.* = cp;
-        ret 3;
+        ret .Ok(3);
     }
     if b0 < 0xF5 {
-        if index + 3 >= len { ret 0; }
+        if index + 3 >= len { ret .Truncated; }
         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; }
+        if (b1 & 0xC0) != 0x80 || (b2 & 0xC0) != 0x80 || (b3 & 0xC0) != 0x80 { ret .Invalid; }
         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; }
+        if cp < 0x10000 || cp > 0x10FFFF { ret .Invalid; }
         out_cp.* = cp;
-        ret 4;
+        ret .Ok(4);
+    }
+    ret .Invalid;
+}
+
+// 测试辅助：把 DecodeResult 归约为“消耗字节数”，非 Ok 视为 0。
+fn decode_used(data: *u8, len: usize, index: usize, out_cp: *u32): usize {
+    match decode(data, len, index, out_cp) {
+        .Ok(used) => { ret used; },
+        .Eof => { ret 0; },
+        .Truncated => { ret 0; },
+        .Invalid => { ret 0; },
     }
     ret 0;
 }
@@ -63,9 +73,10 @@ 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;
+        match decode(data, len, i, &cp) {
+            .Ok(used) => { i = i + used; },
+            _ => { ret false; },
+        }
     }
     ret true;
 }
@@ -107,10 +118,10 @@ fn count_codepoints(data: *u8, len: usize): usize {
     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;
+        match decode(data, len, i, &cp) {
+            .Ok(used) => { i = i + used; n = n + 1; },
+            _ => { ret n; },
+        }
     }
     ret n;
 }
@@ -121,10 +132,13 @@ fn utf16_units(data: *u8, len: usize): usize {
     var n: usize = 0;
     var cp: u32 = 0;
     while i < len {
-        var used = decode(data, len, i, &cp);
-        if used == 0 { ret n; }
+        match decode(data, len, i, &cp) {
+            .Ok(used) => {
                 if cp >= 0x10000 { n = n + 2; } else { n = n + 1; }
                 i = i + used;
+            },
+            _ => { ret n; },
+        }
     }
     ret n;
 }
@@ -135,10 +149,13 @@ fn byte_to_utf16(data: *u8, len: usize, byte_off: usize): usize {
     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; }
+        match decode(data, len, i, &cp) {
+            .Ok(used) => {
                 if cp >= 0x10000 { n = n + 2; } else { n = n + 1; }
                 i = i + used;
+            },
+            _ => { ret n; },
+        }
     }
     ret n;
 }
@@ -150,8 +167,11 @@ fn utf16_to_byte(data: *u8, len: usize, utf16_off: usize): usize {
     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 used: usize = 0;
+        match decode(data, len, i, &cp) {
+            .Ok(value) => { used = value; },
+            _ => { ret i; },
+        }
         var units: usize = 1;
         if cp >= 0x10000 { units = 2; }
         if n + units > utf16_off { ret i; }
@@ -172,6 +192,7 @@ fn expect_ok(result: Status) {
 
 fn test_decode_encode_roundtrip() {
     var out = ByteBuffer.new();
+    defer { ByteBuffer.drop(&out); }
     expect_ok(encode_into(&out, 0x41));    // 'A'
     expect_ok(encode_into(&out, 0xE9));    // é
     expect_ok(encode_into(&out, 0x20AC));  // €
@@ -188,11 +209,10 @@ fn test_decode_encode_roundtrip() {
     @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);
+    @assert(decode_used(bytes.ptr, bytes.len, 0, &cp) == 1 && cp == 0x41, "dec A");
+    @assert(decode_used(bytes.ptr, bytes.len, 1, &cp) == 2 && cp == 0xE9, "dec e");
+    @assert(decode_used(bytes.ptr, bytes.len, 3, &cp) == 3 && cp == 0x20AC, "dec euro");
+    @assert(decode_used(bytes.ptr, bytes.len, 6, &cp) == 4 && cp == 0x1F600, "dec emoji");
 }
 
 fn test_invalid() {
@@ -212,7 +232,27 @@ fn test_invalid() {
     @assert(!valid(s4.ptr, s4.len), "lone cont");
     @assert(!valid(s5.ptr, s5.len), "truncated");
 
+    // 新 ADT：越界/截断/非法可区分。
+    var cp: u32 = 0;
+    match decode(s5.ptr, s5.len, 0, &cp) {
+        .Truncated => {},
+        _ => { @assert(false, "truncated sequence should be Truncated"); },
+    }
+    match decode(s5.ptr, s5.len, s5.len, &cp) {
+        .Eof => {},
+        _ => { @assert(false, "past end should be Eof"); },
+    }
+    match decode(s1.ptr, s1.len, 0, &cp) {
+        .Invalid => {},
+        _ => { @assert(false, "overlong should be Invalid"); },
+    }
+    match decode(s4.ptr, s4.len, 0, &cp) {
+        .Invalid => {},
+        _ => { @assert(false, "lone continuation should be Invalid"); },
+    }
+
     var out = ByteBuffer.new();
+    defer { ByteBuffer.drop(&out); }
     match encode_into(&out, 0xD800) {
         .Ok => { @assert(false, "surrogate encode"); },
         .Err(error) => {},
@@ -221,11 +261,11 @@ fn test_invalid() {
         .Ok => { @assert(false, "out of range encode"); },
         .Err(error) => {},
     }
-    ByteBuffer.drop(&out);
 }
 
 fn test_offsets() {
     var out = ByteBuffer.new();
+    defer { ByteBuffer.drop(&out); }
     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
@@ -241,7 +281,6 @@ fn test_offsets() {
     @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() {

stage2/utils/value.spl

@@ -0,0 +1,100 @@
+// Expect retcode = 0
+
+// 显式装箱值：运行期格式化用的最小 ADT。调用方显式构造（如 Value.int(x)），
+// 库内用 match 分派。不做隐式装箱、不带运行期类型系统。
+// Str 借用 ByteView，不拥有存储；Ptr 是类型擦除指针。
+
+type .{ .ByteView } = @import("byte_view.spl");
+
+type Value = enum {
+    Int: i64,
+    Uint: u64,
+    Bool: bool,
+    Char: u8,
+    Str: ByteView,
+    Ptr: *any,
+}
+
+fn int(value: i64): Value {
+    ret .Int(value);
+}
+
+fn uint(value: u64): Value {
+    ret .Uint(value);
+}
+
+fn boolean(value: bool): Value {
+    ret .Bool(value);
+}
+
+fn char(value: u8): Value {
+    ret .Char(value);
+}
+
+fn str(bytes: []u8): Value {
+    ret .Str(ByteView.from_slice(bytes));
+}
+
+fn str_view(view: ByteView): Value {
+    ret .Str(view);
+}
+
+fn addr(value: *any): Value {
+    ret .Ptr(value);
+}
+
+// ---- 测试 ----
+
+fn test_int_and_uint() {
+    match int(-5) {
+        .Int(v) => { @assert(v == -5, "int payload"); },
+        _ => { @assert(false, "expected int"); },
+    }
+    match uint(5) {
+        .Uint(v) => { @assert(v == 5, "uint payload"); },
+        _ => { @assert(false, "expected uint"); },
+    }
+}
+
+fn test_scalars() {
+    match boolean(true) {
+        .Bool(v) => { @assert(v, "bool payload"); },
+        _ => { @assert(false, "expected bool"); },
+    }
+    match char('A') {
+        .Char(v) => { @assert(v == 'A', "char payload"); },
+        _ => { @assert(false, "expected char"); },
+    }
+}
+
+fn test_borrowed_str() {
+    var text: [3]u8 = .[104, 105, 0];
+    var s = str(text[0..2]);
+    match s {
+        .Str(view) => {
+            @assert(view.len == 2 && view.ptr[0] == 104 && view.ptr[1] == 105,
+                    "str payload borrows");
+        },
+        _ => { @assert(false, "expected str"); },
+    }
+    @assert(text[0] == 104, "source untouched");
+}
+
+fn test_ptr() {
+    var x: i32 = 7;
+    match addr(&x as *any) {
+        .Ptr(p) => { @assert((p as *i32).* == 7, "ptr payload"); },
+        _ => { @assert(false, "expected ptr"); },
+    }
+    match addr(null) {
+        .Ptr(p) => { @assert(p == null, "null ptr payload"); },
+        _ => { @assert(false, "expected ptr"); },
+    }
+}
+
+fn main() {
+    test_int_and_uint();
+    test_scalars();
+    test_borrowed_str();
+    test_ptr();
+}

stage2/utils/vec.spl

@@ -1,9 +1,10 @@
 // Expect retcode = 0
+
 // 类型擦除的动态数组。元素按值复制，数组只释放自身存储，不释放元素内部资源。
 // Vec 拥有 raw 指向的分配；普通赋值会复制所有权，所有权转移必须使用 take。
 
 type .{ .Error, .Kind, .Status } = @import("error.spl");
-type .{ .SizeResult, .checked_add, .checked_mul, .move } = @import("memory.spl");
+type .{ .SizeResult, .checked_add, .checked_mul, .grow_capacity, .move } = @import("memory.spl");
 type .{ .alloc, .copy, .free } = @import("../sys/mem.spl");
 
 type Vec = @self;
@@ -87,6 +88,9 @@ fn allocate(capacity: usize, elem_size: usize, elem_align: usize): VecResult {
     }
 
     var raw = alloc(total) as *u8;
+    if raw == null {
+        ret .Err(Error.make(.Overflow));
+    }
     var address = raw as usize;
     var mask = elem_align - 1;
     if address > (~(0 as usize)) - mask {
@@ -109,15 +113,7 @@ fn reserve(self: *Vec, need: usize): Status {
     if !valid(self.*) { ret status_error(.InvalidArgument); }
     if need <= self.cap { ret .Ok; }
 
-    var new_cap = self.cap;
-    if new_cap < 8 { new_cap = 8; }
-    while new_cap < need {
-        if new_cap > (~(0 as usize)) / 2 {
-            new_cap = need;
-            break;
-        }
-        new_cap = new_cap * 2;
-    }
+    var new_cap = grow_capacity(self.cap, 8, need);
 
     var fresh: Vec = undefined;
     match allocate(new_cap, self.elem_size, self.elem_align) {
@@ -337,6 +333,7 @@ fn test_construction_and_alignment() {
 
 fn test_growth_and_access() {
     var v = make_vec(@sizeof(i32), @alignof(i32));
+    defer { drop(&v); }
     var i: i32 = 0;
     while i < 1000 {
         var value = i * 3 - 7;
@@ -358,11 +355,11 @@ fn test_growth_and_access() {
     var popped: i32 = 0;
     expect_ok(pop(&v, &popped as *u8));
     @assert(popped == 2990 && v.len == 999, "pop copies then removes");
-    drop(&v);
 }
 
 fn test_self_reference() {
     var v = make_vec(@sizeof(i32), @alignof(i32));
+    defer { drop(&v); }
     var i: i32 = 0;
     while i < 8 {
         expect_ok(push(&v, &i as *u8));
@@ -391,17 +388,18 @@ fn test_self_reference() {
     }
     expect_ok(set(&v, 0, last_ptr));
     expect_i32(&v, 0, 7);
-    drop(&v);
 }
 
 fn test_clone_take_and_failures() {
     var source = make_vec(@sizeof(i32), @alignof(i32));
+    defer { drop(&source); }
     var one: i32 = 1;
     var two: i32 = 2;
     expect_ok(push(&source, &one as *u8));
     expect_ok(push(&source, &two as *u8));
 
     var copy = make_vec(@sizeof(i32), @alignof(i32));
+    defer { drop(&copy); }
     expect_ok(clone(&copy, source));
     expect_i32(&copy, 0, 1);
     one = 99;
@@ -415,6 +413,7 @@ fn test_clone_take_and_failures() {
     expect_i32(&copy, 1, 2);
 
     var moved = take(&copy);
+    defer { drop(&moved); }
     @assert(copy.raw == null && copy.data == null && copy.len == 0 && copy.cap == 0);
     expect_i32(&moved, 0, 1);
 
@@ -430,10 +429,6 @@ fn test_clone_take_and_failures() {
     var output: i32 = 55;
     expect_error(pop(&moved, &output as *u8), false);
     @assert(output == 55, "failed pop preserves output");
-
-    drop(&moved);
-    drop(&copy);
-    drop(&source);
 }
 
 fn main() {