fix: 第10轮搜索审计修复 — 缓存键PubMed标志、NOT OR语义、affiliation ILIKE、前端布尔OR切换和精确短语

- 后端: _search_cache_key和_facet_cache_key加入"pm": _is_pm(query)防止PubMed/纯文本路径缓存碰撞
- 后端: OR模式下neg_conds不再AND到所有条件，改用or_(*term_conditions)直接组合
- 后端: _field_condition("all")所有5条分支添加affiliation ILIKE通过jsonb_array_elements(authors)子查询
- 后端: AdvancedSearchRequest.query添加max_length=2000
- 前端: SearchView搜索栏添加AND/OR切换器和精确短语复选框
- 前端: exact_phrase参数从SearchView传递到API请求
- 前端: booleanOp和exactPhrase同步到URL并可从URL恢复

backend/app/api/v1/features.py

@@ -49,7 +49,7 @@ async def test_rule(req: TestRuleRequest, pmid: int = Query(...), db: AsyncSessi
 # ─── 高级搜索 ───
 
 class AdvancedSearchRequest(BaseModel):
-    query: str = ""
+    query: str = Field("", max_length=2000)
     field: str = "all"
     boolean: str = "and"
     exact_phrase: bool = False

backend/app/services/search_engine.py

@@ -55,8 +55,10 @@ class AdvancedSearchEngine:
     ) -> str:
         """归一化查询参数 → 确定性缓存 key（所有 list 排序后参与哈希）"""
         import hashlib, json
+        from app.services.pubmed_query_parser import is_pubmed_syntax as _is_pm
         norm = {
             "q": query.strip().lower(),
+            "pm": _is_pm(query),
             "f": field, "b": boolean, "ep": exact_phrase,
             "yf": year_from, "yt": year_to,
             "df": date_from, "dt": date_to,
@@ -109,8 +111,10 @@ class AdvancedSearchEngine:
         与 _search_cache_key 的区别：不含 p/ps/s/cd/ci。
         """
         import hashlib, json
+        from app.services.pubmed_query_parser import is_pubmed_syntax as _is_pm
         norm = {
             "q": query.strip().lower(),
+            "pm": _is_pm(query),
             "f": field, "b": boolean, "ep": exact_phrase,
             "yf": year_from, "yt": year_to,
             "df": date_from, "dt": date_to,
@@ -1144,10 +1148,13 @@ class AdvancedSearchEngine:
 
         # 将 term_conditions 加入 conditions
         if term_conditions:
-            if pp.boolean_operator in ("or", "mixed"):
+            if pp.boolean_operator == "or":
+                # OR 模式：所有条件（含 NOT）OR 在一起
+                conditions.append(or_(*term_conditions))
+            elif pp.boolean_operator == "mixed":
+                # mixed 模式下 NOT 项应独立 AND（PubMed: A OR B NOT C = (A OR B) AND NOT C）
                 from sqlalchemy.sql.elements import UnaryExpression
                 from sqlalchemy.sql import operators as _sa_ops
-                # OR/mixed 模式下 NOT 项应独立 AND（PubMed: A OR B NOT C = (A OR B) AND NOT C）
                 pos_conds = [c for c in term_conditions
                              if not (isinstance(c, UnaryExpression) and c.modifier == _sa_ops.inv)]
                 neg_conds = [c for c in term_conditions
@@ -1382,7 +1389,11 @@ class AdvancedSearchEngine:
             pat = _pt()
             if exact and not _wildcard:
                 # P4: 精确短语 → phraseto_tsquery（利用 GIN 索引，保留词序）
-                return GlobalLiterature.search_tsv.op("@@")(func.phraseto_tsquery("english", term))
+                return or_(
+                    GlobalLiterature.search_tsv.op("@@")(func.phraseto_tsquery("english", term)),
+                    text("EXISTS (SELECT 1 FROM jsonb_array_elements(global_literature.authors) AS _e "
+                         "WHERE _e->>'affiliation' ILIKE :aff_pat)").bindparams(aff_pat=pat),
+                )
             if _wildcard:
                 # wildcard → ILIKE 右截断（tsvector 不支持 *），多字段覆盖
                 return or_(
@@ -1393,6 +1404,8 @@ class AdvancedSearchEngine:
                     GlobalLiterature.journal_iso.ilike(pat),
                     cast(GlobalLiterature.pmid, String).ilike(pat),
                     GlobalLiterature.doi.ilike(pat),
+                    text("EXISTS (SELECT 1 FROM jsonb_array_elements(global_literature.authors) AS _e "
+                         "WHERE _e->>'affiliation' ILIKE :aff_pat)").bindparams(aff_pat=pat),
                 )
             like_val = f"%{_escaped}%"
             if "/" in term:
@@ -1400,6 +1413,8 @@ class AdvancedSearchEngine:
                     return or_(
                         GlobalLiterature.doi.ilike(_escape_ilike(term)),
                         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),
                     )
                 return or_(
                     GlobalLiterature.title.ilike(like_val),
@@ -1408,6 +1423,8 @@ class AdvancedSearchEngine:
                     GlobalLiterature.abstract.ilike(like_val),
                     GlobalLiterature.author_names_text.ilike(like_val),
                     GlobalLiterature.journal.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),
                 )
             # P7-D2: Chinese → ILIKE fallback (tsvector is English-only)
             if re.search(r'[一-鿿㐀-䶿豈-﫿]', term):
@@ -1416,6 +1433,8 @@ class AdvancedSearchEngine:
                     GlobalLiterature.abstract.ilike(like_val),
                     GlobalLiterature.author_names_text.ilike(like_val),
                     GlobalLiterature.journal.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),
                 )
             # tsvector 索引主覆盖 title/abstract/author_names/chemicals/genes/mesh/keywords
             # journal/journal_iso/affiliation 不在 tsvector 中，以 ILIKE 兜底
@@ -1423,6 +1442,8 @@ class AdvancedSearchEngine:
                 GlobalLiterature.search_tsv.op("@@")(func.plainto_tsquery("english", term)),
                 GlobalLiterature.journal.ilike(like_val),
                 GlobalLiterature.journal_iso.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),
             )
 
     @staticmethod

frontend/src/views/app/SearchView.vue

@@ -44,6 +44,8 @@ const negativeResult = ref('')  // '' = all, 'yes', 'no', 'only'
 const selectedSpecies = ref<string[]>([])
 const selectedSex = ref<string[]>([])
 const selectedAge = ref<string[]>([])
+const booleanOp = ref('and')   // "and" | "or"
+const exactPhrase = ref(false)
 
 // ── PubMed 筛选器状态 ──
 const hasAbstract = ref(false)
@@ -277,8 +279,9 @@ const { page, total, goToPage } = usePagination({
       }
       const body: SearchRequestBody = {
         query: query.value, page: p, page_size: pageSize.value, sort: sort.value,
-        boolean: 'and',
+        boolean: booleanOp.value,
       }
+      if (exactPhrase.value) body.exact_phrase = true
       if (field.value !== 'all') body.field = field.value
       // P3-6: precision_mode 不再发送（后端已忽略）
       // ── Keyset 游标分页（所有排序模式通用，跳过 COUNT + OFFSET） ──
@@ -401,6 +404,8 @@ function restoreFromQuery() {
   if (route.query.tier) selectedTiers.value = String(route.query.tier).split(',')
   if (route.query.pub_type) pubTypes.value = String(route.query.pub_type).split(',')
   if (route.query.lang) languages.value = String(route.query.lang).split(',')
+  if (route.query.boolean === 'or') booleanOp.value = 'or'
+  if (route.query.exact_phrase === 'true') exactPhrase.value = true
   if (route.query.subset) nlmSubsets.value = String(route.query.subset).split(',')
   if (route.query.retracted) retracted.value = String(route.query.retracted)
   if (route.query.negative) negativeResult.value = String(route.query.negative)
@@ -514,6 +519,8 @@ function syncSearchToUrl() {
   if (hasAssociatedData.value) q.associated_data = 'true'
   if (medlineOnly.value) q.medline_only = 'true'
   if (excludePreprints.value) q.exclude_preprints = 'true'
+  if (booleanOp.value !== 'and') q.boolean = booleanOp.value
+  if (exactPhrase.value) q.exact_phrase = 'true'
   if (pageSize.value !== 20) q.page_size = String(pageSize.value)
   // 页码持久化到 URL（Keyset 排序使用 keyset 页码，offset 排序使用 offset 页码）
   const currentPage = KEYSET_SORTS.has(sort.value) ? keysetPage.value : page.value
@@ -544,6 +551,8 @@ function resetAllFilters() {
   query.value = ''
   field.value = 'all'
   sort.value = 'date'
+  booleanOp.value = 'and'
+  exactPhrase.value = false
   goToPage(1)
 }
 
@@ -829,7 +838,9 @@ const specialTags = computed(() => filterOptions.value?.special_tags || {})
     <div style="flex:1;min-width:0">
       <div class="search-bar">
         <div class="search-input-wrap"><NInput v-model:value="query" placeholder='搜索标题、摘要、作者、MeSH词... 支持PubMed语法如 "lung cancer"[TI]' size="small" clearable aria-label="搜索文献" @keyup.enter="goToPage(1)" /></div>
-        <NSelect v-model:value="field" :options="[{label:'全部',value:'all'},{label:'标题',value:'title'},{label:'摘要',value:'abstract'},{label:'作者',value:'author'},{label:'机构',value:'affiliation'},{label:'期刊',value:'journal'}]" size="small" />
+        <NSelect v-model:value="booleanOp" :options="[{label:'AND',value:'and'},{label:'OR',value:'or'}]" size="small" style="width:75px;flex-shrink:0" />
+        <NCheckbox v-model:checked="exactPhrase" size="small" style="flex-shrink:0;font-size:13px;white-space:nowrap">精确短语</NCheckbox>
+        <NSelect v-model:value="field" :options="[{label:'全部',value:'all'},{label:'标题',value:'title'},{label:'摘要',value:'abstract'},{label:'作者',value:'author'},{label:'机构',value:'affiliation'},{label:'期刊',value:'journal'}]" size="small" style="width:90px;flex-shrink:0" />
         <NButton class="sky-btn" size="small" :loading="loading" :disabled="loading" @click="goToPage(1)"><template #icon><NIcon size="14"><SearchOutline /></NIcon></template>搜索</NButton>
         <NButton class="sky-btn" :ghost="showFilters" size="small" @click="showFilters=!showFilters"><template #icon><NIcon size="14"><component :is="showFilters ? EyeOffOutline : FilterOutline" /></NIcon></template>{{ showFilters?'隐藏筛选':'筛选' }}</NButton>
       </div>