refactor: share VPI escape and willFree scanners

Replace the two near-identical VPI annotation line scanners with one marker scanner parameterized by annotation kind. Source-only change: 91 added, 123 deleted (net -32 lines).
2026-11-02 13:29:29 +08:00
1 parent 38a521fcd2
commit 9e0729b855
@@ -121,7 +121,6 @@ define VPI_IDX_FNV_OFFSET = 2166136261
 define VPI_IDX_FNV_PRIME = 16777619
 define VPI_IDX_FNV_MASK = 4294967295

-# v3 跨模块逃逸摘要缓存（.vpi 行尾 `# esc=<mask> ret=<ret>`，由 StubGen.Converter._PyiFunction 写出）。
 # 键 = "{sha1}.{Class}.{method}"（方法）或 "{sha1}.{func}"（模块级函数）；
 # 值打包：低 32 位 = 形参逃逸掩码，高 32 位 = 返回借用位。无逃逸/无返回借用的 def 不写条目，
 # 因此"查不到"即"无摘要"（调用点保守放行），与本地表口径一致。
@@ -137,7 +136,6 @@ _g_vpi_wf_keys: bytes = None
 _g_vpi_wf_vals: sizet* = None
 _g_vpi_wf_count: int = 0
 # 跨模块常量特化（P3 收尾）：同一函数可有多条 `escg=`（打包 p|k|m，见 StructGenMixin._PackEscapeClear），
-# 所以是"一键多行"表：键列与 wf 表同构，值列是打包整数。
 define VPI_ESG_MAX = 8192
 define VPI_ESG_KEY_LEN = 224
 # 常量环境的形参位上限：必须与 StructGenMixin.EC_ENV_MAX 一致（越界位一律不信）
@@ -1124,7 +1122,6 @@ def _ensure_vpi_ret_annot_loaded(src_sha1: str, temp_dir: str) -> int:


 # _cache_vpi_esc_clears_text - 抽取行尾**所有** `escg=<packed>` 标记（跨模块常量特化）。
-# 与 wf 表同构，但同一行可有多条（一条一个打包值），故是"一键多行"。值 = p|k|m 打包整数
 # （见 StructGenMixin._PackEscapeClear：p=守卫形参位、k=常量、m=可清掉的位掩码）。
 def _cache_vpi_esc_clears_text(vpi_buf: bytes, src_sha1: str) -> int:
    """抽取 vpi 行尾的 escg= 标记进缓存；1=处理完成 / 0=缓冲区或分配失败"""
@@ -1395,18 +1392,27 @@ def _cache_vpi_annots_text(vpi_buf: bytes, src_sha1: str) -> int:
    return 1


-# _cache_vpi_escape_annots_text - extract the trailing `# esc=<mask> ret=<ret>` markers.
-# v3 跨模块逃逸摘要：不依赖 AST，按行扫描（与 _cache_vpi_return_annots_text 同构）。
-def _cache_vpi_escape_annots_text(vpi_buf: bytes, src_sha1: str) -> int:
-    """抽取 vpi 行尾逃逸摘要标记进缓存；1=处理完成 / 0=缓冲区或分配失败"""
+# _cache_vpi_marker_annots_text - scan .vpi def lines and cache per-method markers.
+# kind 0 = escape summary (`# esc=<mask> ret=<ret>`), kind 1 = willFree (`# ... wf=<mask>`).
+def _cache_vpi_marker_annots_text(vpi_buf: bytes, src_sha1: str, kind: int) -> int:
    global _g_vpi_esc_keys, _g_vpi_esc_vals, _g_vpi_esc_count
+    global _g_vpi_wf_keys, _g_vpi_wf_vals, _g_vpi_wf_count
    if vpi_buf is None or src_sha1 is None: return 0
-    if _g_vpi_esc_keys is None:
-        _g_vpi_esc_keys = stdlib.malloc(VPI_ESC_MAX * VPI_ESC_KEY_LEN)
-        _g_vpi_esc_vals = sizet*(stdlib.malloc(VPI_ESC_MAX * 8))
-        if _g_vpi_esc_keys is None or _g_vpi_esc_vals is None: return 0
-        string.memset(_g_vpi_esc_keys, 0, VPI_ESC_MAX * VPI_ESC_KEY_LEN)
-        string.memset(_g_vpi_esc_vals, 0, VPI_ESC_MAX * 8)
+    if kind == 0:
+        if _g_vpi_esc_keys is None:
+            _g_vpi_esc_keys = stdlib.malloc(VPI_ESC_MAX * VPI_ESC_KEY_LEN)
+            _g_vpi_esc_vals = sizet*(stdlib.malloc(VPI_ESC_MAX * 8))
+            if _g_vpi_esc_keys is None or _g_vpi_esc_vals is None: return 0
+            string.memset(_g_vpi_esc_keys, 0, VPI_ESC_MAX * VPI_ESC_KEY_LEN)
+            string.memset(_g_vpi_esc_vals, 0, VPI_ESC_MAX * 8)
+    else:
+        if _g_vpi_wf_keys is None:
+            _g_vpi_wf_keys = stdlib.malloc(VPI_WF_MAX * VPI_WF_KEY_LEN)
+            _g_vpi_wf_vals = sizet*(stdlib.malloc(VPI_WF_MAX * 8))
+            if _g_vpi_wf_keys is None or _g_vpi_wf_vals is None: return 0
+            string.memset(_g_vpi_wf_keys, 0, VPI_WF_MAX * VPI_WF_KEY_LEN)
+            string.memset(_g_vpi_wf_vals, 0, VPI_WF_MAX * 8)
+
    sha1_len: sizet = string.strlen(src_sha1)
    buf_len: sizet = string.strlen(vpi_buf)
    cur_cls: bytes = None
@@ -1418,7 +1424,6 @@ def _cache_vpi_escape_annots_text(vpi_buf: bytes, src_sha1: str) -> int:
        i: sizet = ls
        while i < le and vpi_buf[i] == ' ': i += 1
        if i < le:
-            # class X: -> 记录当前类上下文（与方法键一致；兼容 virtual/novirtual 前缀）
            cls_off_e: int = _vpi_class_name_offset(vpi_buf, i, le)
            if cls_off_e >= 0:
                cstart: sizet = sizet(cls_off_e)
@@ -1430,34 +1435,65 @@ def _cache_vpi_escape_annots_text(vpi_buf: bytes, src_sha1: str) -> int:
                    if cur_cls is not None:
                        string.strncpy(cur_cls, vpi_buf + cstart, cend - cstart)
                        cur_cls[cend - cstart] = '\0'
-            # def name(...) ... # esc=<mask> ret=<ret>
            elif le - i >= 4 and _vpi_func_name_offset(vpi_buf, i, le) >= 0:
-                # 模块级函数（无缩进）清掉类上下文，避免上一类泄漏
                if i == ls and cur_cls is not None:
                    stdlib.free(cur_cls)
                    cur_cls = None
-                # 找同行 "# esc=" 标记
                mark: sizet = le
+                hm: sizet = le
                j: sizet = i
-                while j + 6 <= le:
-                    if vpi_buf[j] == '#' and vpi_buf[j + 1] == ' ' and string.strncmp(vpi_buf + j + 2, "esc=", 4) == 0:
-                        mark = j
+                while j < le:
+                    if vpi_buf[j] == '#':
+                        hm = j
                        break
                    j += 1
-                if mark < le:
-                    kp: sizet = mark + 6
-                    esc_v: int = 0
-                    while kp < le and vpi_buf[kp] >= '0' and vpi_buf[kp] <= '9':
-                        esc_v = esc_v * 10 + (int(vpi_buf[kp]) - 48)
-                        kp += 1
-                    ret_v: int = 0
-                    while kp < le and vpi_buf[kp] != 'e': kp += 1
-                    if kp + 4 <= le:
-                        kp += 4
+                esc_v: int = 0
+                ret_v: int = 0
+                wf_v: int = 0
+                if kind == 0:
+                    j2: sizet = i
+                    while j2 + 6 <= le:
+                        if vpi_buf[j2] == '#' and vpi_buf[j2 + 1] == ' ' and string.strncmp(vpi_buf + j2 + 2, "esc=", 4) == 0:
+                            mark = j2
+                            break
+                        j2 += 1
+                    if mark < le:
+                        kp: sizet = mark + 6
                        while kp < le and vpi_buf[kp] >= '0' and vpi_buf[kp] <= '9':
-                            ret_v = ret_v * 10 + (int(vpi_buf[kp]) - 48)
+                            esc_v = esc_v * 10 + (int(vpi_buf[kp]) - 48)
                            kp += 1
-                    if (esc_v != 0 or ret_v != 0) and _g_vpi_esc_count < VPI_ESC_MAX:
+                        while kp < le and vpi_buf[kp] != 'e': kp += 1
+                        if kp + 4 <= le:
+                            kp += 4
+                            while kp < le and vpi_buf[kp] >= '0' and vpi_buf[kp] <= '9':
+                                ret_v = ret_v * 10 + (int(vpi_buf[kp]) - 48)
+                                kp += 1
+                else:
+                    if hm < le:
+                        j3: sizet = hm + 1
+                        while j3 + 3 <= le:
+                            if vpi_buf[j3] == 'w' and vpi_buf[j3 + 1] == 'f' and vpi_buf[j3 + 2] == '=':
+                                mark = j3
+                                break
+                            j3 += 1
+                    if mark < le:
+                        kp2: sizet = mark + 3
+                        while kp2 < le and vpi_buf[kp2] >= '0' and vpi_buf[kp2] <= '9':
+                            wf_v = wf_v * 10 + (int(vpi_buf[kp2]) - 48)
+                            kp2 += 1
+
+                has_marker: int = 0
+                if kind == 0:
+                    if esc_v != 0 or ret_v != 0: has_marker = 1
+                else:
+                    if wf_v != 0: has_marker = 1
+                if has_marker != 0:
+                    can_store: int = 0
+                    if kind == 0:
+                        if _g_vpi_esc_count < VPI_ESC_MAX: can_store = 1
+                    else:
+                        if _g_vpi_wf_count < VPI_WF_MAX: can_store = 1
+                    if can_store != 0:
                        dstart: sizet = sizet(_vpi_func_name_offset(vpi_buf, i, le))
                        dend: sizet = dstart
                        while dend < le and vpi_buf[dend] != '(': dend += 1
@@ -1466,7 +1502,12 @@ def _cache_vpi_escape_annots_text(vpi_buf: bytes, src_sha1: str) -> int:
                            cls_len_e: sizet = 0
                            if cur_cls is not None: cls_len_e = string.strlen(cur_cls)
                            key_len_e: sizet = sha1_len + 1 + cls_len_e + 1 + mname_len
-                            if key_len_e < VPI_ESC_KEY_LEN:
+                            key_ok: int = 0
+                            if kind == 0:
+                                if key_len_e < VPI_ESC_KEY_LEN: key_ok = 1
+                            else:
+                                if key_len_e < VPI_WF_KEY_LEN: key_ok = 1
+                            if key_ok != 0:
                                mname_buf_e: bytes = stdlib.malloc(mname_len + 1)
                                key_buf_e: bytes = stdlib.malloc(key_len_e + 1)
                                if mname_buf_e is not None and key_buf_e is not None:
@@ -1474,9 +1515,14 @@ def _cache_vpi_escape_annots_text(vpi_buf: bytes, src_sha1: str) -> int:
                                    mname_buf_e[mname_len] = '\0'
                                    if cur_cls is not None: viperlib.snprintf(key_buf_e, key_len_e + 1, "%s.%s.%s", src_sha1, cur_cls, mname_buf_e)
                                    else: viperlib.snprintf(key_buf_e, key_len_e + 1, "%s.%s", src_sha1, mname_buf_e)
-                                    string.strcpy(_g_vpi_esc_keys + sizet(_g_vpi_esc_count) * VPI_ESC_KEY_LEN, key_buf_e)
-                                    _g_vpi_esc_vals[_g_vpi_esc_count] = sizet(esc_v) | (sizet(ret_v) << 32)
-                                    _g_vpi_esc_count += 1
+                                    if kind == 0:
+                                        string.strcpy(_g_vpi_esc_keys + sizet(_g_vpi_esc_count) * VPI_ESC_KEY_LEN, key_buf_e)
+                                        _g_vpi_esc_vals[_g_vpi_esc_count] = sizet(esc_v) | (sizet(ret_v) << 32)
+                                        _g_vpi_esc_count += 1
+                                    else:
+                                        string.strcpy(_g_vpi_wf_keys + sizet(_g_vpi_wf_count) * VPI_WF_KEY_LEN, key_buf_e)
+                                        _g_vpi_wf_vals[_g_vpi_wf_count] = sizet(wf_v)
+                                        _g_vpi_wf_count += 1
                                if mname_buf_e is not None: stdlib.free(mname_buf_e)
                                if key_buf_e is not None: stdlib.free(key_buf_e)
        pos = le + 1
@@ -1484,6 +1530,12 @@ def _cache_vpi_escape_annots_text(vpi_buf: bytes, src_sha1: str) -> int:
    return 1


+# _cache_vpi_escape_annots_text - cache the escape summary markers.
+def _cache_vpi_escape_annots_text(vpi_buf: bytes, src_sha1: str) -> int:
+    """Cache escape-summary markers from one .vpi text."""
+    return _cache_vpi_marker_annots_text(vpi_buf, src_sha1, 0)
+
+
 # _lookup_vpi_escape_packed - query the cross-module escape summary cache (v3).
 def _lookup_vpi_escape_packed(src_sha1: str, cls_name: str, method_name: str) -> int:
    """查跨模块逃逸摘要；返回打包值（低32位 mask | ret<<32），0 = 无摘要"""
@@ -1504,94 +1556,10 @@ def _lookup_vpi_escape_packed(src_sha1: str, cls_name: str, method_name: str) ->
    return res


-# _cache_vpi_willfree_annots_text - extract the trailing `wf=<mask>` marker (P2 willFree).
-# 与逃逸摘要同构：不依赖 AST，按行扫描；只在有 wf 标记的 def 上建条目。
+# _cache_vpi_willfree_annots_text - cache the P2 willFree markers.
 def _cache_vpi_willfree_annots_text(vpi_buf: bytes, src_sha1: str) -> int:
-    """抽取 vpi 行尾 P2 willFree 标记（`# ... wf=<mask>`）进缓存；1=处理完成 / 0=缓冲区或分配失败"""
-    global _g_vpi_wf_keys, _g_vpi_wf_vals, _g_vpi_wf_count
-    if vpi_buf is None or src_sha1 is None: return 0
-    if _g_vpi_wf_keys is None:
-        _g_vpi_wf_keys = stdlib.malloc(VPI_WF_MAX * VPI_WF_KEY_LEN)
-        _g_vpi_wf_vals = sizet*(stdlib.malloc(VPI_WF_MAX * 8))
-        if _g_vpi_wf_keys is None or _g_vpi_wf_vals is None: return 0
-        string.memset(_g_vpi_wf_keys, 0, VPI_WF_MAX * VPI_WF_KEY_LEN)
-        string.memset(_g_vpi_wf_vals, 0, VPI_WF_MAX * 8)
-    sha1_len: sizet = string.strlen(src_sha1)
-    buf_len: sizet = string.strlen(vpi_buf)
-    cur_cls: bytes = None
-    pos: sizet = 0
-    while pos < buf_len:
-        ls: sizet = pos
-        le: sizet = pos
-        while le < buf_len and vpi_buf[le] != '\n': le += 1
-        i: sizet = ls
-        while i < le and vpi_buf[i] == ' ': i += 1
-        if i < le:
-            # class X: -> 记录当前类上下文（与方法键一致；兼容 virtual/novirtual 前缀）
-            cls_off_e: int = _vpi_class_name_offset(vpi_buf, i, le)
-            if cls_off_e >= 0:
-                cstart: sizet = sizet(cls_off_e)
-                cend: sizet = cstart
-                while cend < le and vpi_buf[cend] != ':' and vpi_buf[cend] != ' ' and vpi_buf[cend] != '(': cend += 1
-                if cend > cstart:
-                    if cur_cls is not None: stdlib.free(cur_cls)
-                    cur_cls = stdlib.malloc(cend - cstart + 1)
-                    if cur_cls is not None:
-                        string.strncpy(cur_cls, vpi_buf + cstart, cend - cstart)
-                        cur_cls[cend - cstart] = '\0'
-            # def name(...) ... # [esc=<mask> ret=<ret>] wf=<mask>
-            elif le - i >= 4 and _vpi_func_name_offset(vpi_buf, i, le) >= 0:
-                # 模块级函数（无缩进）清掉类上下文，避免上一类泄漏
-                if i == ls and cur_cls is not None:
-                    stdlib.free(cur_cls)
-                    cur_cls = None
-                # 只在注释段里找 "wf="：避开参数名（如 def f(wf: int)）
-                mark: sizet = le
-                hm: sizet = le
-                j: sizet = i
-                while j < le:
-                    if vpi_buf[j] == '#':
-                        hm = j
-                        break
-                    j += 1
-                if hm < le:
-                    j2: sizet = hm + 1
-                    while j2 + 3 <= le:
-                        if vpi_buf[j2] == 'w' and vpi_buf[j2 + 1] == 'f' and vpi_buf[j2 + 2] == '=':
-                            mark = j2
-                            break
-                        j2 += 1
-                if mark < le:
-                    kp: sizet = mark + 3
-                    wf_v: int = 0
-                    while kp < le and vpi_buf[kp] >= '0' and vpi_buf[kp] <= '9':
-                        wf_v = wf_v * 10 + (int(vpi_buf[kp]) - 48)
-                        kp += 1
-                    if wf_v != 0 and _g_vpi_wf_count < VPI_WF_MAX:
-                        dstart: sizet = sizet(_vpi_func_name_offset(vpi_buf, i, le))
-                        dend: sizet = dstart
-                        while dend < le and vpi_buf[dend] != '(': dend += 1
-                        if dend > dstart:
-                            mname_len: sizet = dend - dstart
-                            cls_len_w: sizet = 0
-                            if cur_cls is not None: cls_len_w = string.strlen(cur_cls)
-                            key_len_w: sizet = sha1_len + 1 + cls_len_w + 1 + mname_len
-                            if key_len_w < VPI_WF_KEY_LEN:
-                                mname_buf_w: bytes = stdlib.malloc(mname_len + 1)
-                                key_buf_w: bytes = stdlib.malloc(key_len_w + 1)
-                                if mname_buf_w is not None and key_buf_w is not None:
-                                    string.strncpy(mname_buf_w, vpi_buf + dstart, mname_len)
-                                    mname_buf_w[mname_len] = '\0'
-                                    if cur_cls is not None: viperlib.snprintf(key_buf_w, key_len_w + 1, "%s.%s.%s", src_sha1, cur_cls, mname_buf_w)
-                                    else: viperlib.snprintf(key_buf_w, key_len_w + 1, "%s.%s", src_sha1, mname_buf_w)
-                                    string.strcpy(_g_vpi_wf_keys + sizet(_g_vpi_wf_count) * VPI_WF_KEY_LEN, key_buf_w)
-                                    _g_vpi_wf_vals[_g_vpi_wf_count] = sizet(wf_v)
-                                    _g_vpi_wf_count += 1
-                                if mname_buf_w is not None: stdlib.free(mname_buf_w)
-                                if key_buf_w is not None: stdlib.free(key_buf_w)
-        pos = le + 1
-    if cur_cls is not None: stdlib.free(cur_cls)
-    return 1
+    """Cache willFree markers from one .vpi text."""
+    return _cache_vpi_marker_annots_text(vpi_buf, src_sha1, 1)


 # _lookup_vpi_willfree - query the cross-module P2 willFree cache (key shape = escape summary).
@@ -114,7 +114,7 @@ Retained: `ClassGenEnum.RenumTableDump` is referenced from a diagnostic f-string
 The following duplicate implementations were merged or redirected:

 - `Phase2`: one directory walker for `.vp` and `.s` sources.
- `VpiReader`: shared `_vpi_method_key` for all cached-annotation lookups, a shared pre-registration walker for defaults/enums, and a shared `.vpi` read helper for the two annotation cache loaders.
+- `VpiReader`: shared `_vpi_method_key` for all cached-annotation lookups, a shared pre-registration walker for defaults/enums, a shared `.vpi` read helper for the two annotation cache loaders, and one marker scanner for escape/willFree annotations.
 - `StubMergerStore`: shared store initialization/append and alias parsing.
 - `AssignGenMixin` / `ExprGenCdef`: one `TypeGenCDef.NormalizeDefineBits`.
 - `AssignGenMixin` / `ExprGenCls`: removed the weaker duplicate class-annotation helpers; calls now use `ExprGenCls`.
@@ -128,7 +128,9 @@ Measured source-only reduction for the main consolidation batch: **235 added / 7

 Follow-up batch: VLogger level filtering and the VpiReader pre-registration/read helpers, **53 added / 81 deleted = net -28 lines**.

-Remaining candidates: `VpiReader` escape/willfree annotation scanners, `BuildPipeline` MLIR path setup, `LSP _SliceStr/_ByteAt`, and the list/dict iteration helpers.
+Scanner batch: one VpiReader marker scanner for escape/willFree annotations, **91 added / 123 deleted = net -32 lines**.
+
+Remaining candidates: `BuildPipeline` MLIR path setup, `LSP _SliceStr/_ByteAt`, the list/dict iteration helpers, and the nested-function rescan walkers.

 ---