feat(sidecar): meta 增写 STRUCT/VTABLE，REPL 布局恢复改读 meta

- WriteModuleFuncMeta 更名 WriteModuleMeta：追加 STRUCT <name> FIELDS n PACKED 0|1 （opaque → STRUCT <name> OPAQUE）与 VTABLE <class> SLOTS n；数据源改为 Module.NamedTypeHead / GlobalHead，不再有任何 IR 文本反推 - VpiReader REPL 布局改读 {sha1}.meta：新增 _meta_atoi / _repl_meta_scan_layout 按前缀精确匹配解析 STRUCT/VTABLE 行 - 删除旧 .stub.ll 文本解析辅助 _repl_stub_take_name / _repl_stub_body_count 及其 scratch 缓冲 ReplStubNameBuf / _REPL_STUB_NAME_CAP 去 IR 化步骤③；未编译验证。
2026-10-03 08:18:15 +08:00
1 parent 78d3941f7d
commit 35f2070d3b
@@ -29,7 +29,6 @@ import lib.Projectrans.Config as Config
 _REPL_STUB_CLS_MAX: t.CDefine = 1024
 _REPL_STUB_FIELD_MAX: int = 128
 _REPL_STUB_VT_MAX: int = 128
-_REPL_STUB_NAME_CAP: sizet = 256
 _REPL_STUB_TXT_CAP: sizet = 4194304
 _REPL_STUB_WARN_MAX: int = 10
 _REPL_STUB_FINISH_PASSES: int = 8
@@ -3686,7 +3685,6 @@ ReplStubWarnLeft: int = 0
 ReplStubTmpCount: int = -1
 ReplStubTmpPacked: int = 0
 ReplStubTmpArity: int = 0
-ReplStubNameBuf: str = None


 def _repl_stub_dupstr(pool: memhub.MemBuddy*, src: str) -> str:
@@ -3761,105 +3759,31 @@ def _repl_stub_readtext(path: str) -> str:
    return buf


-# 从 txt[pos] 起取一个标识符（跳过 '%' / '@' 前缀与可选引号）写进 ReplStubNameBuf；
-# 返回名字之后的位置（0 = 失败/超长；合法调用至少前进 2 字符，故 0 可作失败哨兵）
-def _repl_stub_take_name(txt: str, pos: sizet) -> sizet:
-    global ReplStubNameBuf
-    if txt is None or ReplStubNameBuf is None:
-        return 0
-    n: sizet = string.strlen(txt)
-    i: sizet = pos
-    if i < n and (txt[i] == '%' or txt[i] == '@'):
+# 从 txt[start, end) 解析十进制整数（跳过前导空格；无数字返回 0）
+def _meta_atoi(txt: str, start: sizet, end: sizet) -> int:
+    i: sizet = start
+    while i < end and (txt[i] == ' ' or txt[i] == '\t'):
        i += 1
-    if i >= n:
-        return 0
-    if txt[i] == '\"':
+    neg: int = 0
+    if i < end and txt[i] == '-':
+        neg = 1
        i += 1
-        k: sizet = 0
-        while i < n and txt[i] != '\"':
-            if k + 1 >= _REPL_STUB_NAME_CAP:
-                return 0
-            ReplStubNameBuf[k] = txt[i]
-            k += 1
-            i += 1
-        if i >= n:
-            return 0
-        ReplStubNameBuf[k] = 0
-        return i + 1
-    k2: sizet = 0
-    while i < n:
-        ch: char = txt[i]
-        if ch == ' ' or ch == '\t' or ch == '=' or ch == ',' or ch == '{' or ch == '}':
-            break
-        if ch == '[' or ch == ']' or ch == '(' or ch == ')' or ch == '*' or ch == '<' or ch == '>':
-            break
-        if k2 + 1 >= _REPL_STUB_NAME_CAP:
-            return 0
-        ReplStubNameBuf[k2] = ch
-        k2 += 1
+    v: int = 0
+    while i < end and txt[i] >= '0' and txt[i] <= '9':
+        v = v * 10 + (int(txt[i]) - 48)
        i += 1
-    if k2 == 0:
-        return 0
-    ReplStubNameBuf[k2] = 0
-    return i
+    if neg != 0:
+        return -v
+    return v


-# 从 txt[pos]（应指向 '{'）起数顶层字段数：{} / [] / () / <> / "..." 内的逗号不计；
-# 返回字段数（-1 = 未闭合/格式异常）
-def _repl_stub_body_count(txt: str, pos: sizet) -> int:
-    if txt is None:
-        return -1
-    n: sizet = string.strlen(txt)
-    i: sizet = pos
-    if i >= n or txt[i] != '{':
-        return -1
-    i += 1
-    depth: int = 0
-    in_str: int = 0
-    cnt: int = 0
-    has: int = 0
-    while i < n:
-        ch: char = txt[i]
-        if in_str != 0:
-            if ch == '\"':
-                in_str = 0
-            i += 1
-            continue
-        if ch == '\"':
-            in_str = 1
-            has = 1
-            i += 1
-            continue
-        if ch == 0:
-            return -1
-        if ch == ',':
-            if depth == 0:
-                cnt += 1
-            i += 1
-            continue
-        if ch == '{' or ch == '[' or ch == '(' or ch == '<':
-            depth += 1
-            has = 1
-            i += 1
-            continue
-        if ch == '}' or ch == ']' or ch == ')' or ch == '>':
-            if ch == '}' and depth == 0:
-                if has == 0:
-                    return 0
-                return cnt + 1
-            depth -= 1
-            i += 1
-            continue
-        if ch != ' ' and ch != '\t' and ch != '\n' and ch != '\r':
-            has = 1
-        i += 1
-    return -1
-
-
-# 扫描存根文本，取 `%"<key>" = type ...` 与 `@...<key>.vtable = ...` 两条元数据，
-# 结果写全局 scratch：ReplStubTmpCount（字段数 / -1 无此类 / -2 opaque）、
-# ReplStubTmpPacked、ReplStubTmpArity（.vtable 槽数，0 = 无该全局）
-def _repl_stub_scan_meta(txt: str, key: str) -> None:
+# 扫描 {sha1}.meta 文本，取 `STRUCT <key> FIELDS <n> PACKED <0|1>`（或 `STRUCT <key> OPAQUE`）
+# 与 `VTABLE <key> SLOTS <n>` 两条元数据，结果写全局 scratch：
+# ReplStubTmpCount（字段数 / -1 无此类 / -2 opaque）、ReplStubTmpPacked、
+# ReplStubTmpArity（虚表槽数，0 = 无该虚表）。key = "{sha1}.{类名}"。
+# 逐行、按前缀 strncmp 精确匹配，且 name 之后必须紧跟空格（避免前缀误命中）；
+# 不走 take_name —— 类名可能含泛型下标 [ ] 等字符。
+def _repl_meta_scan_layout(txt: str, key: str) -> None:
    global ReplStubTmpCount, ReplStubTmpPacked, ReplStubTmpArity
    ReplStubTmpCount = -1
    ReplStubTmpPacked = 0
@@ -3868,6 +3792,8 @@ def _repl_stub_scan_meta(txt: str, key: str) -> None:
        return
    n: sizet = string.strlen(txt)
    key_len: sizet = string.strlen(key)
+    if key_len == 0:
+        return
    i: sizet = 0
    while i < n:
        ls: sizet = i
@@ -3876,61 +3802,37 @@ def _repl_stub_scan_meta(txt: str, key: str) -> None:
        le: sizet = i
        if i < n:
            i += 1
-        p: sizet = ls
-        while p < le and (txt[p] == ' ' or txt[p] == '\t' or txt[p] == '\r'):
-            p += 1
-        if p >= le:
-            continue
-        c0: char = txt[p]
-        if c0 == '%':
-            # 类型定义行
-            q: sizet = _repl_stub_take_name(txt, p)
-            if q == 0 or q > le:
+        if le > ls and txt[le - 1] == '\r':
+            le -= 1
+        if ls + 7 <= le and string.strncmp(txt + ls, "STRUCT ", 7) == 0:
+            ns: sizet = ls + 7
+            ne: sizet = ns + key_len
+            if ne > le or string.strncmp(txt + ns, key, key_len) != 0:
                continue
-            if string.strcmp(ReplStubNameBuf, key) != 0:
+            if ne >= le or txt[ne] != ' ':
                continue
-            # 找 "= type"
-            j: sizet = q
-            while j + 6 < le and string.strncmp(txt + j, "= type", 6) != 0:
-                j += 1
-            if j + 6 >= le:
-                continue
-            j += 6
-            while j < le and (txt[j] == ' ' or txt[j] == '\t'):
-                j += 1
-            if j + 6 <= le and string.strncmp(txt + j, "opaque", 6) == 0:
+            # OPAQUE
+            if ne + 7 <= le and string.strncmp(txt + ne + 1, "OPAQUE", 6) == 0:
                ReplStubTmpCount = -2
                continue
-            if j < le and txt[j] == '<':
-                ReplStubTmpPacked = 1
-                j += 1
-            if j < le and txt[j] == '{':
-                ReplStubTmpCount = _repl_stub_body_count(txt, j)
+            # FIELDS <n> PACKED <0|1>
+            if ne + 8 <= le and string.strncmp(txt + ne + 1, "FIELDS ", 7) == 0:
+                ReplStubTmpCount = _meta_atoi(txt, ne + 8, le)
+                ps: sizet = ne + 8
+                while ps < le and txt[ps] >= '0' and txt[ps] <= '9':
+                    ps += 1
+                if ps + 8 <= le and string.strncmp(txt + ps + 1, "PACKED ", 7) == 0:
+                    ReplStubTmpPacked = _meta_atoi(txt, ps + 8, le)
            continue
-        if c0 == '@':
-            # 全局变量行：只关心 <key>.vtable
-            q2: sizet = _repl_stub_take_name(txt, p)
-            if q2 == 0 or q2 > le:
+        if ls + 7 <= le and string.strncmp(txt + ls, "VTABLE ", 7) == 0:
+            vs: sizet = ls + 7
+            ve: sizet = vs + key_len
+            if ve > le or string.strncmp(txt + vs, key, key_len) != 0:
                continue
-            nm_len: sizet = string.strlen(ReplStubNameBuf)
-            if nm_len != key_len + 7:
+            if ve >= le or txt[ve] != ' ':
                continue
-            if string.strncmp(ReplStubNameBuf, key, key_len) != 0:
-                continue
-            if string.strncmp(ReplStubNameBuf + key_len, ".vtable", 7) != 0:
-                continue
-            j2: sizet = q2
-            while j2 < le and txt[j2] != '=':
-                j2 += 1
-            if j2 >= le:
-                continue
-            j2 += 1
-            while j2 < le and txt[j2] != '{':
-                j2 += 1
-            if j2 < le and txt[j2] == '{':
-                ar: int = _repl_stub_body_count(txt, j2)
-                if ar > 0:
-                    ReplStubTmpArity = ar
+            if ve + 7 <= le and string.strncmp(txt + ve + 1, "SLOTS ", 6) == 0:
+                ReplStubTmpArity = _meta_atoi(txt, ve + 7, le)
            continue
    return

@@ -3939,7 +3841,7 @@ def ReplStubLayoutBegin(temp_dir: str) -> int:
    """开启会话级"存根布局恢复"（Repl.vp 在 includes 类注册前调用一次）；0=成功"""
    global ReplStubLayoutMode, ReplStubPool, ReplStubTempDir, ReplStubTable
    global ReplStubCount, ReplStubOk, ReplStubVtOk, ReplStubRefused, ReplStubNoStub
-    global ReplStubWarnLeft, ReplStubNameBuf
+    global ReplStubWarnLeft
    global ReplStubRefusedNoStub, ReplStubRefusedNoClass, ReplStubRefusedOpaque, ReplStubRefusedParent
    pool: memhub.MemBuddy* = CompilerFactory.factory.PermPool
    if pool is None or temp_dir is None:
@@ -3952,10 +3854,6 @@ def ReplStubLayoutBegin(temp_dir: str) -> int:
    if ReplStubTable is None:
        return 1
    string.memset(ReplStubTable, 0, 8 * _REPL_STUB_CLS_MAX)
-    ReplStubNameBuf = pool.alloc(_REPL_STUB_NAME_CAP)
-    if ReplStubNameBuf is None:
-        return 1
-    ReplStubNameBuf[0] = 0
    ReplStubCount = 0
    ReplStubOk = 0
    ReplStubVtOk = 0
@@ -4508,7 +4406,7 @@ def _repl_stub_add_class(pool: memhub.MemBuddy*,
    # 存根元数据（缺存根时 LayoutCount 保持"未找到"哨兵；两种成因由 have_stub 区分）
    e.LayoutCount = _REPL_STUB_NOCOUNT_NOCLASS
    if txt is not None:
-        _repl_stub_scan_meta(txt, key)
+        _repl_meta_scan_layout(txt, key)
        e.LayoutCount = ReplStubTmpCount
        e.VtArity = ReplStubTmpArity
        if ReplStubTmpPacked != 0:
@@ -4564,9 +4462,9 @@ def _repl_stub_collect_module(pool: memhub.MemBuddy*,
        return 0
    if src_sha1 is None or src_sha1[0] == '\0' or ReplStubTable is None:
        return 0
-    # 存根文本（缺失：带父类/虚表的类将因"数据缺失"被明确拒绝，逐个可见）
+    # 布局元数据（{sha1}.meta；缺失：带父类/虚表的类将因"数据缺失"被明确拒绝，逐个可见）
    txt: str = None
-    sp: str = _repl_stub_path(src_sha1, "stub.ll")
+    sp: str = _repl_stub_path(src_sha1, "meta")
    if sp is not None:
        txt = _repl_stub_readtext(sp)
        stdlib.free(sp)
@@ -4577,8 +4475,8 @@ def _repl_stub_collect_module(pool: memhub.MemBuddy*,
        ReplStubNoStub += 1
        if ReplStubWarnLeft > 0:
            ReplStubWarnLeft -= 1
-            stdio.printf("includes 类布局: 模块 %s 无 .stub.ll 存根（数据缺失）—— 该模块"
-                         "带父类/虚表的类将逐个按『数据缺失』明确拒绝（读不到存根 ⇒ 无"
+            stdio.printf("includes 类布局: 模块 %s 无 .meta 存根（数据缺失）—— 该模块"
+                         "带父类/虚表的类将逐个按『数据缺失』明确拒绝（读不到布局元数据 ⇒ 无"
                         "字段数可比，不是字段数不符）。用同一工程先跑一次批处理生成存根，"
                         "再 `--project .\\project.vpj --repl`（TransPyV 目录）即可复用\n",
                         src_sha1)
@@ -31,21 +31,25 @@ import lib.Stub.StubMerger as StubMerger
 # ============================================================

 # ============================================================
-# WriteModuleFuncMeta - 落盘 {sha1}.meta 的函数定义清单（结构化 sidecar）
+# WriteModuleMeta - 落盘 {sha1}.meta（模块结构化 sidecar：函数清单 + 结构体布局 + 虚表）
 #
-# 每行 "FUNC <mangled> DEF|DECL"：DEF = 本模块定义了该符号（有函数体，等同 .text.ll 的 define 行），
-# DECL = 仅声明（等同 .stub.ll 的 declare）。数据源为翻译期模型（Module.FuncHead 链表 +
-# Function.IsDeclared），与 dump_ir 的 stub/text 判据同源（STUB=IsDeclared==1、TEXT==0），
-# 取代「从 IR 文本反推 declare-vs-define」。
+# 行格式（每行一条，纯文本、无嵌套括号）：
+#   FUNC <mangled> DEF|DECL                       DEF=本模块定义了该符号（有函数体，等同 .text.ll 的 define 行）
+#   STRUCT <sha1>.<类名> FIELDS <n> PACKED <0|1>  命名结构体字段数/是否 packed
+#   STRUCT <sha1>.<类名> OPAQUE                   前向声明（opaque，无 body）
+#   VTABLE <sha1>.<类名> SLOTS <n>                虚表槽数（全局 @<类名>.vtable 的元素数）
+# 数据源均为翻译期模型（Module.FuncHead / NamedTypeHead / GlobalHead），与 dump_ir 同源，
+# 取代「从 IR 文本反推 declare-vs-define / 结构体布局 / 虚表槽数」。
 #
-# 写不出（路径/打开失败）即 fail-fast：缺 meta 会使 Phase B+/C 的「声明文件判定」失效。
+# 写不出（路径/打开失败）即 fail-fast：缺 meta 会使 Phase B+/C 的声明文件判定与
+# REPL 的布局恢复失效。
 #
 # Args: mod 已翻译完成的 LLVM 模块；temp_dir 基础目录；td_len temp_dir 长度；sha1 模块 SHA1
 # Returns: 0=成功（失败经 VLogger.error 终止进程）
 # ============================================================
-def WriteModuleFuncMeta(mod: llvmlite.LLVMModule*, temp_dir: str,
-                        td_len: sizet, sha1: str) -> int:
-    """把本模块的函数定义清单写入 {sha1}.meta"""
+def WriteModuleMeta(mod: llvmlite.LLVMModule*, temp_dir: str,
+                    td_len: sizet, sha1: str) -> int:
+    """把本模块的函数定义/结构体布局/虚表清单写入 {sha1}.meta"""
    if mod is None or temp_dir is None or sha1 is None:
        VLogger.error("meta 写出：参数为空", "Meta")
        return 1
@@ -78,6 +82,58 @@ def WriteModuleFuncMeta(mod: llvmlite.LLVMModule*, temp_dir: str,
            mf.write(line, string.strlen(line))
            stdlib.free(line)
        cur = llvmlite.function_get_next(cur)
+    # 2) 结构体布局：STRUCT <name> FIELDS <n> PACKED <0|1>（opaque → STRUCT <name> OPAQUE）
+    nt: llvmlite.NamedTypeNode* = mod.NamedTypeHead
+    nt_steps: sizet = 0
+    while nt is not None:
+        nt_steps += 1
+        if nt_steps > 100000:
+            VLogger.error("meta 写出：命名类型链表成环/超限", "Meta")
+            break
+        tn: str = llvmlite.get_struct_name(nt.Ty)
+        if tn is not None and tn[0] != '\0':
+            tl: sizet = string.strlen(tn)
+            lb: bytes = stdlib.malloc(tl + 48)
+            if lb is None:
+                VLogger.error("meta 写出：行缓冲分配失败", "Meta")
+                break
+            fc: int = llvmlite.get_struct_fcount(nt.Ty)
+            if fc < 0:
+                viperlib.snprintf(lb, tl + 48, "STRUCT %s OPAQUE\n", tn)
+            else:
+                viperlib.snprintf(lb, tl + 48, "STRUCT %s FIELDS %d PACKED %d\n", tn, fc, nt.IsPacked)
+            mf.write(lb, string.strlen(lb))
+            stdlib.free(lb)
+        nt = nt.Next
+    # 3) 虚表槽数：VTABLE <class> SLOTS <n>（全局名 <class>.vtable，Ty 匿名结构体 fcount=槽数）
+    gv: llvmlite.GlobalVariable* = mod.GlobalHead
+    gv_steps: sizet = 0
+    while gv is not None:
+        gv_steps += 1
+        if gv_steps > 100000:
+            VLogger.error("meta 写出：全局变量链表成环/超限", "Meta")
+            break
+        gn: str = gv.Name
+        if gn is not None:
+            gnl: sizet = string.strlen(gn)
+            if gnl > 7 and string.strcmp(gn + gnl - 7, ".vtable") == 0:
+                base_len: sizet = gnl - 7
+                base: bytes = stdlib.malloc(base_len + 1)
+                if base is not None:
+                    string.strncpy(base, gn, base_len)
+                    base[base_len] = '\0'
+                    ar: int = llvmlite.get_struct_fcount(gv.Ty)
+                    if ar < 0:
+                        ar = 0
+                    lb2: bytes = stdlib.malloc(base_len + 40)
+                    if lb2 is None:
+                        VLogger.error("meta 写出：行缓冲分配失败", "Meta")
+                    else:
+                        viperlib.snprintf(lb2, base_len + 40, "VTABLE %s SLOTS %d\n", base, ar)
+                        mf.write(lb2, string.strlen(lb2))
+                        stdlib.free(lb2)
+                    stdlib.free(base)
+        gv = gv.Next
    mf.close()
    stdlib.free(mpath)
    return 0