feat(mcp): 申购单/出库单生成与抽取结果局部修正(2a)

- src/infra/com_thread.py:Office COM 调用的单线程串行 actor。Word/Excel 并发驱动会互相抢
  ActiveDocument 与模态框,因此所有 COM 操作排队在同一线程执行(与 P3 的 Playwright actor 分开,
  COM 套间与 greenlet 约束不同)
- mcp__arb__doc_requisition:Excel 申购单,产物写 cases/<id>/artifacts/,支持 exclude 关键词与
  project 覆盖。实机验证:2 张普通发票 + 1 笔支付 → 排除「万用表」后生成 19456 字节 .xls,5.0s
- mcp__arb__doc_consumable:Word 出库单;纯差旅返回 skipped 而非报错。当前被缺模板挡住(见下)
- mcp__arb__case_patch_extraction:只改传入的键,回显每个字段的 from/to,未知键告警但仍写入,
  与现值相同时不写(避免无意义重抽),金额/日期仍走 normalize
- server.py:工具数 15 → 18,playbook 增加第 6 步(文档产物)与核对修正说明

发现(未处理,需产品决定):Word 模板「易耗品、出库单.doc」在 4fbb4c3(2026-08-31,同一提交新增
requisition.py)被删除,至今不在仓库里。因此 consumable.py 的出库单路径自那以后一直不可用;
Web 端 _try_fill_consumable_doc 捕获异常后只记日志,所以普通发票提交一直在静默跳过出库单生成。
已确认历史版本是空白校用表单(无姓名、无卡号、无长数字串),是否恢复或改为配置模板路径待定。

门禁:ruff / format --check / mypy / deptry 全绿,pytest 132 passed。
This commit is contained in:
wandering
2026-10-19 20:07:38 +08:00
parent 2f08bbd6e8
commit b2d271c5ed
4 changed files with 219 additions and 2 deletions
+42
View File
@@ -90,6 +90,48 @@ def case_save_extraction(case_id: str, source_file: str, data: dict[str, Any]) -
return ok(cached=str(path), invoice_type=normalized.get("invoice_type"), warnings=warnings)
def case_patch_extraction(case_id: str, source_file: str, patch: dict[str, Any]) -> dict[str, Any]:
"""局部修正已落盘的抽取结果(人工核对后改字段用)。
只合并传入的键,返回每个被改字段的 from/to,避免静默覆盖;
未在 schema 中出现的键会告警但仍写入(有些字段是给下游自定义用的)。
"""
d = _store.resolve(case_id)
if d is None:
return fail(f"会话不存在: {case_id}")
if not isinstance(patch, dict) or not patch:
return fail("patch 必须是非空对象")
from ...core.extraction.cache import load_cache as _load
from ...core.extraction.cache import save_extraction
current = _load(d).get(source_file)
if current is None:
return fail(f"该材料还没有抽取结果: {source_file}", next_action="先 case_save_extraction 落一份完整结果")
from ..schemas import DOCUMENT_SCHEMAS, INVOICE_TYPES
doc_type = str(current.get("invoice_type", ""))
schema = DOCUMENT_SCHEMAS.get(doc_type, {})
known = set(schema.get("required", [])) | set(schema.get("optional", [])) | {"invoice_type", "_source_file"}
warnings: list[str] = []
changed: dict[str, dict[str, Any]] = {}
merged = dict(current)
for key, value in patch.items():
if key.startswith("_"):
warnings.append(f"忽略内部字段 {key} 的修改")
continue
if key not in changed and merged.get(key) != value:
changed[key] = {"from": merged.get(key), "to": value}
if doc_type in INVOICE_TYPES and known and key not in known:
warnings.append(f"{key} 不在 {doc_type} 的字段定义里,确认拼写无误")
merged[key] = value
if not changed:
return ok(source_file=source_file, changed={}, warnings=warnings, note="patch 与现值相同,未做写入")
normalized, normalize_warnings = normalize_extraction(merged)
normalized["_source_file"] = source_file
path = save_extraction(d, source_file, normalized)
return ok(source_file=source_file, changed=changed, warnings=warnings + normalize_warnings, cached=str(path))
def extraction_spec(kind: str | None = None) -> dict[str, Any]:
"""返回抽取字段定义。kind 省略给全部文档类型;'info' 给 travel_info/normal_info 骨架。"""
return ok(**schema_doc(kind))