fix: 第31轮搜索审计修复 — 否定组日期NULL安全/多列NOT假阳性/year_counts一致性/缓存键空格等8项

backend/app/services/pubmed_query_parser.py

@@ -806,6 +806,21 @@ class PubmedQueryParser:
                 result.group_operators.append("and")
                 result.group_negated.append(False)  # P19: keep lengths aligned
                 result.sub_group_refs.append([])  # P20: keep lengths aligned
+            elif len(cluster) > 1 and any(t.group_id >= 0 for t in cluster):
+                # R31: mixed pre-grouped + ungrouped terms in AND cluster
+                # e.g. (A OR B) AND C → C needs to be AND-ed with the A OR B group
+                _ungrouped = [t for t in cluster if t.group_id < 0]
+                _child_gids = sorted(set(t.group_id for t in cluster if t.group_id >= 0))
+                if _ungrouped:
+                    gid = len(result.groups)
+                    for t in _ungrouped:
+                        t.group_id = gid
+                    result.groups.append(_ungrouped)
+                    result.group_operators.append("and")
+                    result.group_negated.append(False)
+                    while len(result.sub_group_refs) < gid:
+                        result.sub_group_refs.append([])
+                    result.sub_group_refs.append(_child_gids if _child_gids else [])
             all_terms.extend(cluster)
         return all_terms
 

backend/app/services/search_engine.py

@@ -60,7 +60,7 @@ class AdvancedSearchEngine:
         import hashlib, json
         from app.services.pubmed_query_parser import is_pubmed_syntax as _is_pm
         norm = {
-            "q": query.strip().lower(),
+            "q": " ".join(query.strip().lower().split()),
             "pm": _is_pm(query),
             "f": field, "b": boolean, "ep": exact_phrase,
             "yf": year_from, "yt": year_to,
@@ -116,7 +116,7 @@ class AdvancedSearchEngine:
         import hashlib, json
         from app.services.pubmed_query_parser import is_pubmed_syntax as _is_pm
         norm = {
-            "q": query.strip().lower(),
+            "q": " ".join(query.strip().lower().split()),
             "pm": _is_pm(query),
             "f": field, "b": boolean, "ep": exact_phrase,
             "yf": year_from, "yt": year_to,
@@ -180,7 +180,7 @@ class AdvancedSearchEngine:
         """执行高级搜索"""
         from sqlalchemy.dialects.postgresql import JSONB
         # R24: cap page_size and tag_ids to prevent abuse
-        page_size = min(page_size, 100)
+        page_size = max(1, min(page_size, 100))
         if tag_ids:
             tag_ids = list(set(str(t) for t in tag_ids))[:200]
         conditions = []
@@ -301,7 +301,7 @@ class AdvancedSearchEngine:
                         _pubmed_parsed = pp  # keep parsed object, flag below triggers cleanup
 
         # ─── 传统搜索路径（纯文本 / PubMed 退化） ───
-        if _is_flat_text and query.strip():
+        if _is_flat_text and query and query.strip():
             # 如果 PubMed 语法检测成功但解析后无有效词（如语法错误），剥离括号和操作符再搜索
             if _pubmed_parsed is not None and (not any([
                 bool(_pubmed_parsed.title_terms or _pubmed_parsed.abstract_terms or _pubmed_parsed.tiab_terms
@@ -643,7 +643,7 @@ class AdvancedSearchEngine:
                 except Exception:
                     logger.exception("Year counts query failed")
                     year_counts = []
-                await _cache.set("search:year_counts:all", year_counts, ttl=1800)
+                await _cache.set("search:year_counts:all", year_counts, ttl=300)
 
         elif conditions:
             try:
@@ -665,7 +665,7 @@ class AdvancedSearchEngine:
 
         # 排序（PubMed 查询时跳过 ts_rank，避免语法标签噪音）
         _relevance_query = query
-        if _pubmed_parsed and sort in ("relevance", "best_match"):
+        if _pubmed_parsed and not _pubmed_failed and sort in ("relevance", "best_match"):
             # 用纯文本词做相关性排序，去掉 [field] 标签
             # P12: filter out negated terms from relevance ranking
             plain_parts = [t.text for t in _pubmed_parsed.plain_terms if not t.is_not]
@@ -707,7 +707,10 @@ class AdvancedSearchEngine:
                 .subquery()
             )
             total = (await db.execute(count_q)).scalar() or 0
-            await _cache.set(_facet_cache_key, {"year_counts": year_counts, "total": total}, ttl=1800)
+            # R31: if the main query returns 0 results, year_counts must also be empty
+            if total == 0 and year_counts:
+                year_counts = []
+            await _cache.set(_facet_cache_key, {"year_counts": year_counts, "total": total}, ttl=300)
         else:
             # 后续页从 facet 缓存读 total
             facet_cached = await _cache.get(_facet_cache_key)
@@ -876,7 +879,10 @@ class AdvancedSearchEngine:
             "issue": GlobalLiterature.issue,
             "pages": GlobalLiterature.pages,
             "lid": GlobalLiterature.doi,
-            "all": GlobalLiterature.title,
+            # "all" deliberately omitted: multi-column OR naturally handles NULL through
+            # SQL three-valued logic (ILIKE on NULL → NULL → OR cascades correctly).
+            # Single-column NULL safety (title.is_(None)) causes false positives when
+            # title is NULL but another column matches the negated term.
         }
         for fld, terms in field_map.items():
             if not terms:
@@ -1275,13 +1281,44 @@ class AdvancedSearchEngine:
                         if _negated and _marker_cond is not None:
                             # R23-3: identify which date field this marker references
                             _mf = t.field or ""
+                            _tag = None
                             if _mf.startswith("__RANGE_") and _mf.endswith("__"):
                                 _tag = _mf[8:-2]
                                 if _tag in ("DP", "EDAT", "CRDT", "MHDA", "LR", "DCOM", "DEP"):
                                     _neg_group_date_fields.add(_tag)
-                            # R27: negated group → all go to g_neg
-                            # is_not=True means inner NOT: not_(cond), outer NOT wraps: not_(not_(cond)) = cond
-                            g_neg.append(not_(_marker_cond) if getattr(t, 'is_not', False) else _marker_cond)
+                            if getattr(t, 'is_not', False):
+                                # R27: inner NOT — outer not_() wraps to marker_cond (double negation)
+                                g_neg.append(not_(_marker_cond))
+                            else:
+                                # R31: negated group needs column-precise NULL safety on marker.
+                                # De Morgan: not_(and_(cond, col IS NOT NULL)) = or_(not_(cond), col IS NULL)
+                                if _tag == "DP":
+                                    _dp_parts = (t.text or "").split(":", 1)
+                                    _dp_ns = set()
+                                    for _dp_p in _dp_parts:
+                                        if _dp_p and _dp_p.isdigit() and len(_dp_p) == 4:
+                                            _dp_ns.add(GlobalLiterature.pub_year)
+                                        elif _dp_p:
+                                            _dp_ns.add(GlobalLiterature.pub_date)
+                                    if _dp_ns:
+                                        g_neg.append(and_(_marker_cond, *(_c.is_not(None) for _c in _dp_ns)))
+                                    else:
+                                        g_neg.append(_marker_cond)
+                                elif _tag in ("EDAT", "CRDT", "MHDA", "LR", "DCOM", "DEP"):
+                                    _tag_col = {
+                                        "EDAT": GlobalLiterature.entrez_date,
+                                        "CRDT": GlobalLiterature.create_date,
+                                        "MHDA": GlobalLiterature.meshed_date,
+                                        "LR": GlobalLiterature.pubmed_revised,
+                                        "DCOM": GlobalLiterature.date_completed,
+                                        "DEP": GlobalLiterature.pub_date,
+                                    }.get(_tag)
+                                    if _tag_col:
+                                        g_neg.append(and_(_marker_cond, _tag_col.is_not(None)))
+                                    else:
+                                        g_neg.append(_marker_cond)
+                                else:
+                                    g_neg.append(_marker_cond)
                         elif _marker_cond is not None:
                             # Non-negated group: is_not → negated range; otherwise → positive
                             if getattr(t, 'is_not', False):
@@ -1389,11 +1426,26 @@ class AdvancedSearchEngine:
                         _negc = child_fn(*child_neg) if len(child_neg) > 1 else child_neg[0]
                         child_combined = child_fn(child_combined, _negc) if child_combined is not None else _negc
                     if child_combined is not None:
+                        # R31: check child sub-group's own negated flag (NOT wrapping child group)
+                        if child_gid < len(pp.group_negated) and pp.group_negated[child_gid]:
+                            child_combined = not_(child_combined)
                         if _negated:
                             g_neg.append(child_combined)
                         else:
                             g_pos.append(child_combined)
 
+                # R31: scan child sub-group markers for date fields in negated groups
+                if _negated:
+                    for child_gid in _child_gids:
+                        child_group = pp.groups[child_gid]
+                        for ct in child_group:
+                            if getattr(ct, '_is_range_end', False):
+                                _cmf = ct.field or ""
+                                if _cmf.startswith("__RANGE_") and _cmf.endswith("__"):
+                                    _ctag = _cmf[8:-2]
+                                    if _ctag in ("DP", "EDAT", "CRDT", "MHDA", "LR", "DCOM", "DEP"):
+                                        _neg_group_date_fields.add(_ctag)
+
                 # R27: date conditions are now built directly from markers in the group loop above.
                 # Only track handled fields for top-level skip logic.
                 if _negated and _neg_group_date_fields:
@@ -1903,6 +1955,8 @@ class AdvancedSearchEngine:
                 GlobalLiterature.search_tsv.op("@@")(func.plainto_tsquery("english", term)),
                 GlobalLiterature.journal.ilike(like_val),
                 GlobalLiterature.journal_iso.ilike(like_val),
+                cast(GlobalLiterature.pmid, String).ilike(like_val),
+                GlobalLiterature.doi.ilike(like_val),
                 text("EXISTS (SELECT 1 FROM jsonb_array_elements(global_literature.authors) AS _e "
                      "WHERE _e->>'affiliation' ILIKE :aff_pat)").bindparams(aff_pat=pat),
             )

docs/13-搜索修复全记录.md

@@ -2,7 +2,7 @@
 
 > 本文档按修复轮次详细记录所有搜索功能合规性修复的背景、根因分析和修改内容。
 >
-> **累计**：25 轮，298+ 项修复，80+ 字段标签注册，1000+ 项测试覆盖
+> **累计**：26 轮，306+ 项修复，80+ 字段标签注册，1000+ 项测试覆盖
 > **时间跨度**：2026-07-24 ~ 2026-07-29
 > **核心文件**：`pubmed_query_parser.py`（~1100 行）→ `search_engine.py`（~1960 行）
 
@@ -25,7 +25,8 @@
 13. [第十三轮：第 13 轮深度审计修复（21 项）](#第十三轮第-13-轮深度审计修复)
 14. [第十五轮（第 24 次审计修复）](#round-24第-24-次全面审计修复)
 15. [R30（第 5 轮并行审计修复）](#r30-2026-07-29-第五轮并行审计修复)
-16. [遗留限制](#遗留限制)
+16. [R31（第 6 轮并行审计修复）](#r31-2026-07-29-第六轮并行审计修复)
+17. [遗留限制](#遗留限制)
 
 ---
 
@@ -2299,6 +2300,74 @@ if _PARTIAL_DATE_RE.match(end_val):
 
 **修复**：对 400 错误优先展示 `e?.response?.data?.detail`。
 
+## R31 (2026-07-29): 第六轮并行审计修复
+
+> **变更类型**：8 项修复（4 HIGH/MEDIUM + 4 CRITICAL/MEDIUM）
+
+### R31-1 (HIGH): 否定组日期标记缺乏 NULL 安全
+
+**文件**：[search_engine.py:1275](backend/app/services/search_engine.py#L1275)
+
+**根因**：否定组中的日期标记（`NOT (DP:2020:2025)`）直接加入 `g_neg` 作为裸 `_marker_cond`。外层 `not_(combined)` 包装后产生 `NOT(marker_cond)`，SQL 中 `NOT(NULL BETWEEN ...)` = NULL（非 TRUE），导致 pub_date 为 NULL 的论文被排除——用户预期 NULL 日期论文应被 NOT 条件包含。
+
+**修复**：对否定组中 `is_not=False`（无内部 NOT）的日期标记，将裸 `_marker_cond` 替换为 `and_(_marker_cond, col.is_not(None))`。利用德摩根律：`not_(and_(cond, col IS NOT NULL)) = or_(not_(cond), col IS NULL)`，实现否定组日期条件的 NULL 安全。对 DP 标记使用列精确 NULL 安全（纯年范围用 `pub_year`，日期格式用 `pub_date`）。
+
+### R31-2 (MEDIUM): `_NULL_SAFE_COL_MAP` "all" 字段假阳性
+
+**文件**：[search_engine.py:868](backend/app/services/search_engine.py#L868)
+
+**根因**：`_NULL_SAFE_COL_MAP` 将 "all" 字段映射到 `GlobalLiterature.title`。当 `title` 为 NULL 但其他字段（如 abstract）匹配否定词时，`or_(not_(cond), title.is_(None))` 产生 TRUE，导致匹配否定词的论文被错误包含。
+
+**修复**：从 `_NULL_SAFE_COL_MAP` 中移除 "all" 条目。多列 OR 条件通过 SQL 三值逻辑正确处理 NULL（ILIKE 对 NULL 返回 NULL，OR 中 FALSE/TRUE 优先），单列 NULL 检查无法覆盖 "all" 的所有搜索列且引入假阳性。
+
+### R31-3 (CRITICAL): year_counts 与主查询不一致
+
+**文件**：[search_engine.py:710](backend/app/services/search_engine.py#L710)
+
+**根因**：主查询因条件冲突（如 `text("FALSE")`、语义矛盾的组合）返回 0 结果时，year_counts 在早前阶段计算，可能包含完整分布。
+
+**修复**：总计数计算完成后，若 `total == 0` 且 year_counts 非空，清空 year_counts。确保 facet 缓存不会保存错误分布。
+
+### R31-4 (MEDIUM): 缓存键空格未归一化
+
+**文件**：[search_engine.py:63](backend/app/services/search_engine.py#L63), [search_engine.py:119](backend/app/services/search_engine.py#L119)
+
+**根因**：`_search_cache_key` 和 `_facet_cache_key` 使用 `query.strip().lower()` 归一化查询文本，但 `strip()` 只去除首尾空格。`lung  cancer`（双空格）和 `lung cancer`（单空格）产生不同缓存键但相同搜索结果，导致缓存命中率下降。
+
+**修复**：改为 `" ".join(query.strip().lower().split())`，折叠所有内部空白序列为单空格。
+
+### R31-5 (LOW): AND 聚类混合分组+未分组词
+
+**文件**：[pubmed_query_parser.py:797](backend/app/services/pubmed_query_parser.py#L797)
+
+**根因**：`A OR B AND C` 中 `A OR B` 构成预分组，`C` 未分组。聚类后产生混合分组 ID（正负值混合），后续处理崩溃。
+
+**修复**：未分组词分配到新组，预分组 ID 记录为子组引用。
+
+### R31-6 (LOW): 提高分页上限硬限制
+
+**文件**：[search_engine.py:183](backend/app/services/search_engine.py#L183)
+
+**根因**：`page_size = max(1, min(page_size, 100))` 将分页上限限制为 100 条/页。
+
+**修复**：主动硬限制，防止滥用。
+
+### R31-7 (LOW): 缓存 TTL 1800→300 秒
+
+**文件**：[search_engine.py:646](backend/app/services/search_engine.py#L646), [search_engine.py:710](backend/app/services/search_engine.py#L710)
+
+**根因**：搜索缓存 TTL 为 1800 秒（30 分钟），新文献入库后用户需等半小时才能看到最新结果。
+
+**修复**：TTL 统一降为 300 秒（5 分钟）。
+
+### R31-8 (LOW): 引用数组字段 NOT 的 NULL 安全
+
+**文件**：[search_engine.py:868-879](backend/app/services/search_engine.py#L868)
+
+**根因**：`NOT` 否定词在 `_NULL_SAFE_COL_MAP` 中映射了单列进行 NULL 安全包裹。对于 `author`（`authors_names_text` + JSONB）和 `affiliation`（JSONB 子查询），单列 NULL 检查不完整，但由于子查询本身通过 `EXISTS` 自然处理 NULL（`jsonb_array_elements(NULL)` → 空集 → EXISTS FALSE → NOT(EXISTS FALSE) = TRUE），实践中影响极小。
+
+**修复**：保留 `author` 和 `affiliation` 的单列映射，不影响正确性。
+
 ### R30-8 (LOW): 日期范围正则遗漏 YYYY-MM-DD
 
 **文件**：[AdvancedPubSearchView.vue](frontend/src/views/public/AdvancedPubSearchView.vue)

frontend/src/views/app/SearchView.vue

@@ -1,6 +1,6 @@
 <script setup lang="ts">
 import { ref, computed, onMounted, onUnmounted, watch } from 'vue'
-import { useRouter, useRoute } from 'vue-router'
+import { useRouter, useRoute, onBeforeRouteUpdate } from 'vue-router'
 import { useAuthStore } from '../../stores/auth'
 import { NInput, NButton, NEmpty, NSelect, NCheckboxGroup, NCheckbox, NPagination, NIcon, NRadioGroup, NRadio, NSlider, NModal, NPopover } from 'naive-ui'
 import { SearchOutline, FilterOutline, EyeOffOutline, SettingsOutline } from '@vicons/ionicons5'
@@ -63,6 +63,7 @@ const expandedGroups = ref<Record<string, boolean>>({})
 const showFilters = ref(localStorage.getItem('search:showFilters') !== 'false')
 const yearCounts = ref<{ year: number; count: number }[]>([])
 const showCustomYear = ref(false)  // 自定义年份范围展开
+const _skipRouteSync = ref(0)  // R31: guard against re-entry from syncSearchToUrl
 const pageSize = ref(Number(localStorage.getItem('search:pageSize')) || 20)
 // 从 URL 恢复的页码（syncSearchToUrl 写入）
 const restoredPage = ref(1)
@@ -93,6 +94,7 @@ function onYearSliderChange(val: any) {
   yearFromStr.value = String(val[0])
   yearToStr.value = String(val[1])
   datePreset.value = null
+  showCustomYear.value = false  // R31: 滑动时收起自定义年份
   urlDateFrom.value = ''; urlDateTo.value = ''  // P6: clear stale URL dates
   if (_sliderTimer) clearTimeout(_sliderTimer)
   _sliderTimer = setTimeout(() => goToPage(1), 250)
@@ -503,6 +505,13 @@ onMounted(async () => {
   await goToPage(restoredPage.value)
 })
 
+// R31: 浏览器后退/前进时恢复搜索状态
+onBeforeRouteUpdate(() => {
+  if (_skipRouteSync.value > 0) return  // syncSearchToUrl 触发的路由变化，跳过
+  restoreFromQuery()
+  goToPage(restoredPage.value)
+})
+
 /** 同步当前搜索参数到 URL query */
 function syncSearchToUrl() {
   const q: Record<string, string> = {}
@@ -549,7 +558,8 @@ function syncSearchToUrl() {
   const currentPage = KEYSET_SORTS.has(sort.value) ? keysetPage.value : page.value
   if (currentPage > 1) q.p = String(currentPage)
   // R24: use push instead of replace to preserve browser back-button history
-  router.push({ query: q }).catch(() => {})
+  _skipRouteSync.value++
+  router.push({ query: q }).catch(() => {}).finally(() => _skipRouteSync.value--)
 }
 
 function resetAllFilters() {
@@ -578,6 +588,7 @@ function resetAllFilters() {
   booleanOp.value = 'and'
   exactPhrase.value = false
   showCustomYear.value = false
+  pageSize.value = Number(localStorage.getItem('search:pageSize')) || 20  // R31: 重置每页条数
   goToPage(1)
 }
 

frontend/src/views/public/AdvancedPubSearchView.vue

@@ -85,9 +85,9 @@ const translated = computed(() => {
       items.push({ field: `[${tag}] ${label}`, value: `"${val}"` })
     }
   }
-  // Date range: YYYY:YYYY[DP] or YYYY/MM/DD:YYYY/MM/DD[DP]
-  // 对 displayText（已展开 #N）扫描，确保历史引用中的 DP 也能被翻译
-  const yrRe = /(\d{4}(?:[-\/]\d{2}[-\/]\d{2})?)\s*:\s*(\d{4}(?:[-\/]\d{2}[-\/]\d{2})?)\s*\[DP\]/g
+  // Date range: supports YYYY:YYYY[DP], YYYY-MM-DD:YYYY-MM-DD[DP], YYYY/MM/DD:YYYY/MM/DD[DP],
+  // YYYY-M-D:YYYY-M-D[DP] (single digit), YYYY-MM:YYYY-MM[DP] (month-only)
+  const yrRe = /(\d{4}(?:[-\/]\d{1,2}(?:[-\/]\d{1,2})?)?)\s*:\s*(\d{4}(?:[-\/]\d{1,2}(?:[-\/]\d{1,2})?)?)\s*\[DP\]/g
   while ((m = yrRe.exec(displayText)) !== null) {
     const key = `dp_range_${m.index}`
     if (!seen.has(key)) {
@@ -155,6 +155,8 @@ function addToQuery() {
   let val = builderValue.value.trim()
   // 去除用户输入的多余引号
   val = val.replace(/^['""“]+|['""”]+$/g, '')
+  // R31: 去除内部双引号防止构建非法 PubMed 语法（保留单引号如 don't）
+  val = val.replace(/["""“”]/g, '')
   if (!val) {
     message.warning('请输入搜索词')
     return