echotools 2.3.97__tar.gz → 2.3.99__tar.gz

{echotools-2.3.97 → echotools-2.3.99}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: echotools
-Version: 2.3.97
+Version: 2.3.99
 Summary: 通用基础设施 SDK：配置、日志、事件、调度、插件、协议、调用链
 Home-page: https://github.com/nichengfuben/echotools
 Author: nichengfuben

{echotools-2.3.97 → echotools-2.3.99}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "echotools"
-version = "2.3.97"
+version = "2.3.99"
 description = "通用基础设施 SDK：配置、日志、事件、调度、插件、协议、调用链"
 readme = "README.md"
 requires-python = ">=3.8"

{echotools-2.3.97 → echotools-2.3.99}/src/echotools/exec/fncall/protocols/entml.py

@@ -349,6 +349,11 @@ class EntmlProtocol(ToolProtocol):
         if tool_calls:
             clean = strip_actionable_entml_invoke_blocks(clean, known_names=known)
         clean = strip_tool_entml_residue(clean, known_names=known)
+        from echotools.exec.fncall.protocols.entml_fake_structure_markup import (
+            strip_orphan_entml_close_tags,
+        )
+
+        clean, _ = strip_orphan_entml_close_tags(clean)
         return (clean, normalize_tool_calls(tool_calls, tools))
 
     def parse_fragment(

{echotools-2.3.97 → echotools-2.3.99}/src/echotools/exec/fncall/protocols/entml_fake_structure_markup.py

@@ -26,17 +26,21 @@ _TAG_ONLY_OPEN_RES = (
     re.compile(r"<entml:funtions_results\b[^>]*>", re.IGNORECASE),
     re.compile(r"<entml:conversation_history\b[^>]*>", re.IGNORECASE),
     re.compile(r"<entml:calls\b[^>]*>", re.IGNORECASE),
+    re.compile(r"<entml:call\b[^>]*>", re.IGNORECASE),
     re.compile(r"<function_calling_behavior\b[^>]*>", re.IGNORECASE),
     re.compile(r"<thinking_behavior\b[^>]*>", re.IGNORECASE),
     re.compile(r"<entml:result\b[^>]*>", re.IGNORECASE),
+    re.compile(r"<entml:todo\b[^>]*>", re.IGNORECASE),
 )
 _TAG_ONLY_CLOSE_RES = (
     re.compile(r"</entml:funtions_results\s*>", re.IGNORECASE),
     re.compile(r"</entml:conversation_history\s*>", re.IGNORECASE),
     re.compile(r"</entml:calls\s*>", re.IGNORECASE),
+    re.compile(r"</entml:call\s*>", re.IGNORECASE),
     re.compile(r"</function_calling_behavior\s*>", re.IGNORECASE),
     re.compile(r"</thinking_behavior\s*>", re.IGNORECASE),
     re.compile(r"</entml:result\s*>", re.IGNORECASE),
+    re.compile(r"</entml:todo\s*>", re.IGNORECASE),
 )
 
 _FAKE_ENTML_TAG_PREFIXES: Tuple[str, ...] = (
@@ -48,10 +52,14 @@ _FAKE_ENTML_TAG_PREFIXES: Tuple[str, ...] = (
     "</entml:conversation_history",
     "<entml:calls",
     "</entml:calls",
+    "<entml:call",
+    "</entml:call",
     "<function_calling_behavior",
     "</function_calling_behavior",
     "<thinking_behavior",
     "</thinking_behavior",
+    "<entml:todo",
+    "</entml:todo",
     "</entml:invoke",
     "</entml:parameter",
     "</entml:thinking",
@@ -66,6 +74,15 @@ _TRAILING_INCOMPLETE_ENTML_CLOSE_RE = re.compile(
     r"</entml:?[a-z0-9_-]*\s*$",
     re.IGNORECASE,
 )
+# 完整闭标签泄漏：须在 invoke 块移除之后剥离（独立行或文末）。
+_ORPHAN_ENTML_CLOSE_LINE_COMPLETE_RE = re.compile(
+    r"(?m)^\s*(?:●\s*)?</entml:?[a-z0-9_-]*>\s*(?:\n|$)",
+    re.IGNORECASE,
+)
+_TRAILING_ORPHAN_ENTML_CLOSE_RE = re.compile(
+    r"(?:\n|\A)\s*</entml:?[a-z0-9_-]*>\s*$",
+    re.IGNORECASE,
+)
 
 
 def _strip_fake_entml_result_id_blocks(text: str) -> Tuple[str, bool]:
@@ -88,6 +105,15 @@ def _strip_orphan_entml_close_leaks(text: str) -> Tuple[str, bool]:
     return text, (n1 + n2) > 0
 
 
+def strip_orphan_entml_close_tags(content: str) -> Tuple[str, bool]:
+    """剥离独立行或文末的完整 ``</entml:…>`` 泄漏（invoke 块已移除后调用）。"""
+    if not content:
+        return content, False
+    text, n1 = _ORPHAN_ENTML_CLOSE_LINE_COMPLETE_RE.subn("", content)
+    text, n2 = _TRAILING_ORPHAN_ENTML_CLOSE_RE.subn("", text)
+    return text, (n1 + n2) > 0
+
+
 def _could_be_fake_entml_tag_prefix(fragment: str) -> bool:
     frag = fragment.lower()
     if not frag.startswith("<"):

{echotools-2.3.97 → echotools-2.3.99}/src/echotools.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: echotools
-Version: 2.3.97
+Version: 2.3.99
 Summary: 通用基础设施 SDK：配置、日志、事件、调度、插件、协议、调用链
 Home-page: https://github.com/nichengfuben/echotools
 Author: nichengfuben

{echotools-2.3.97 → echotools-2.3.99}/src/tests/test_entml_fake_structure_markup.py

@@ -7,6 +7,7 @@ from echotools.exec.fncall.parsers.stream import FncallStreamParser
 from echotools.exec.fncall.protocols.entml_fake_structure_markup import (
     strip_fake_entml_structure_markup,
     strip_fake_entml_structure_markup_for_display,
+    strip_orphan_entml_close_tags,
 )
 from echotools.exec.fncall.protocols.entml_tool_result_comment import (
     trailing_partial_tool_result_id_comment_len,
@@ -52,6 +53,12 @@ READ_TOOL = {
             "conversation_history",
             "完",
         ),
+        (
+            "前\n<entml:call>\n调用块正文\n</entml:call>\n后",
+            "调用块正文",
+            "entml:call",
+            "后",
+        ),
         (
             "前\n<entml:calls>\n调用块正文\n</entml:calls>\n后",
             "调用块正文",
@@ -76,6 +83,18 @@ READ_TOOL = {
             "entml:result",
             "ok",
         ),
+        (
+            "除此之外</entml:todo>，<entml:todo>只过滤标签",
+            "只过滤标签",
+            "entml:todo",
+            "除此之外",
+        ),
+        (
+            "前\n<entml:todo>\n正文保留\n</entml:todo>\n后",
+            "正文保留",
+            "entml:todo",
+            "后",
+        ),
     ],
 )
 def test_strip_fake_entml_structure_batch(
@@ -135,6 +154,16 @@ def test_bare_result_tags_only_not_body() -> None:
     assert "尾" in cleaned
 
 
+def test_orphan_complete_entml_close_after_invoke_strip() -> None:
+    cleaned, found = strip_orphan_entml_close_tags("完成编辑\n</entml:invoke>")
+    assert found
+    assert cleaned.strip() == "完成编辑"
+    rest, found2 = strip_orphan_entml_close_tags("前\n后\n</entml:invoke>")
+    assert found2
+    assert "前" in rest and "后" in rest
+    assert "entml" not in rest.lower()
+
+
 def test_orphan_incomplete_entml_close_leak() -> None:
     raw = (
         "说明正文\n"
@@ -188,6 +217,81 @@ def test_batch_parse_strips_fake_structure() -> None:
     assert "funtions_results" not in clean
 
 
+EDIT_TOOL = {
+    "type": "function",
+    "function": {
+        "name": "Edit",
+        "parameters": {
+            "type": "object",
+            "properties": {
+                "path": {"type": "string"},
+                "old_string": {"type": "string"},
+                "new_string": {"type": "string"},
+                "replace_all": {"type": "boolean"},
+            },
+            "required": ["path", "old_string", "new_string"],
+        },
+    },
+}
+
+
+def test_rogator_edit_triple_invoke_corpus_batch() -> None:
+    from pathlib import Path
+
+    log = (
+        Path(__file__).resolve().parent
+        / "fixtures"
+        / "rogator_edit_triple_orphan_close.txt"
+    )
+    raw = log.read_text(encoding="utf-8")
+    proto = get_protocol("entml")
+    clean, calls = proto.parse(raw, [EDIT_TOOL])
+    assert len(calls) == 3
+    assert all(c["function"]["name"] == "Edit" for c in calls)
+    assert "entml" not in clean.lower()
+    assert "main.py" in clean
+
+
+@pytest.mark.parametrize("chunk", [1, 4, 8, 17, 32])
+def test_rogator_edit_triple_invoke_corpus_stream(chunk: int) -> None:
+    from pathlib import Path
+
+    log = (
+        Path(__file__).resolve().parent
+        / "fixtures"
+        / "rogator_edit_triple_orphan_close.txt"
+    )
+    raw = log.read_text(encoding="utf-8")
+    parser = FncallStreamParser(protocol=get_protocol("entml"), tools=[EDIT_TOOL])
+    for i in range(0, len(raw), chunk):
+        parser.feed(raw[i : i + chunk])
+        pt = parser.partial_text.lower()
+        assert "entml:invoke" not in pt
+        assert "entml:parameter" not in pt
+        assert "</entml:invoke" not in pt
+        assert "entml:todo" not in pt
+    clean, calls = parser.finalize()
+    assert len(calls) == 3
+    assert "entml" not in clean.lower()
+
+
+def test_corpus_with_inline_todo_tags_strips_tags_only() -> None:
+    from pathlib import Path
+
+    log = (
+        Path(__file__).resolve().parent
+        / "fixtures"
+        / "rogator_edit_triple_orphan_close.txt"
+    )
+    raw = log.read_text(encoding="utf-8") + "\n除此之外</entml:todo>，<entml:todo>只过滤标签"
+    proto = get_protocol("entml")
+    clean, calls = proto.parse(raw, [EDIT_TOOL])
+    assert len(calls) == 3
+    assert "entml" not in clean.lower()
+    assert "除此之外" in clean
+    assert "只过滤标签" in clean
+
+
 @pytest.mark.parametrize("chunk", [1, 4, 17, 32])
 def test_stream_partial_no_fake_structure_leak(chunk: int) -> None:
     text = (