WIP: Regression hell

2026-03-06 21:22:21 -05:00
parent 528f0a04c3
commit f65e9b40b2
14 changed files with 359 additions and 97 deletions
--- a/src/ai_client.py
+++ b/src/ai_client.py
@@ -252,7 +252,12 @@ def _classify_gemini_error(exc: Exception) -> ProviderError:
 return ProviderError("unknown", "gemini", exc)

 def _classify_deepseek_error(exc: Exception) -> ProviderError:
- body = str(exc).lower()
+ body = ""
+ if isinstance(exc, requests.exceptions.HTTPError) and exc.response is not None:
+  body = exc.response.text.lower()
+ else:
+  body = str(exc).lower()
+ 
 if "429" in body or "rate" in body:
  return ProviderError("rate_limit", "deepseek", exc)
 if "401" in body or "403" in body or "auth" in body or "api key" in body:
@@ -263,6 +268,13 @@ def _classify_deepseek_error(exc: Exception) -> ProviderError:
  return ProviderError("quota", "deepseek", exc)
 if "connection" in body or "timeout" in body or "network" in body:
  return ProviderError("network", "deepseek", exc)
+ 
+ if "400" in body or "bad request" in body:
+  # Try to wrap the original error with the response body for better debugging
+  if body:
+   new_exc = Exception(f"Bad Request (400): {body}")
+   return ProviderError("unknown", "deepseek", new_exc)
+
 return ProviderError("unknown", "deepseek", exc)

 def set_provider(provider: str, model: str) -> None:
@@ -637,6 +649,54 @@ def _build_file_diff_text(changed_items: list[dict[str, Any]]) -> str:
    parts.append(f"### `{path}` (no changes detected)")
 return "\n\n---\n\n".join(parts)

+def _build_deepseek_tools() -> list[dict[str, Any]]:
+ mcp_tools: list[dict[str, Any]] = []
+ for spec in mcp_client.MCP_TOOL_SPECS:
+  if _agent_tools.get(spec["name"], True):
+   mcp_tools.append({
+     "type": "function",
+     "function": {
+      "name":        spec["name"],
+      "description": spec["description"],
+      "parameters":  spec["parameters"],
+     }
+    })
+ tools_list = mcp_tools
+ if _agent_tools.get(TOOL_NAME, True):
+  powershell_tool: dict[str, Any] = {
+   "type": "function",
+   "function": {
+    "name": TOOL_NAME,
+    "description": (
+     "Run a PowerShell script within the project base_dir. "
+     "Use this to create, edit, rename, or delete files and directories. "
+     "The working directory is set to base_dir automatically. "
+     "Always prefer targeted edits over full rewrites where possible. "
+     "stdout and stderr are returned to you as the result."
+    ),
+    "parameters": {
+     "type": "object",
+     "properties": {
+      "script": {
+       "type": "string",
+       "description": "The PowerShell script to execute."
+      }
+     },
+     "required": ["script"]
+    }
+   }
+  }
+  tools_list.append(powershell_tool)
+ return tools_list
+
+_CACHED_DEEPSEEK_TOOLS: Optional[list[dict[str, Any]]] = None
+
+def _get_deepseek_tools() -> list[dict[str, Any]]:
+ global _CACHED_DEEPSEEK_TOOLS
+ if _CACHED_DEEPSEEK_TOOLS is None:
+  _CACHED_DEEPSEEK_TOOLS = _build_deepseek_tools()
+ return _CACHED_DEEPSEEK_TOOLS
+
 def _content_block_to_dict(block: Any) -> dict[str, Any]:
 if isinstance(block, dict):
  return block
@@ -1356,33 +1416,56 @@ def _send_deepseek(md_content: str, user_message: str, base_dir: str,
   "Authorization": f"Bearer {api_key}",
   "Content-Type": "application/json",
  }
-  current_api_messages: list[dict[str, Any]] = []
+  
+  is_reasoner = _model == "deepseek-reasoner"
+  
+  # Update history following Anthropic pattern
  with _deepseek_history_lock:
-   for msg in _deepseek_history:
-    current_api_messages.append(msg)
-  initial_user_message_content = user_message
-  if discussion_history:
-   initial_user_message_content = f"[DISCUSSION HISTORY]\n\n{discussion_history}\n\n---\n\n{user_message}"
-  current_api_messages.append({"role": "user", "content": initial_user_message_content})
-  request_payload: dict[str, Any] = {
-   "model": _model,
-   "messages": current_api_messages,
-   "temperature": _temperature,
-   "max_tokens": _max_tokens,
-   "stream": stream,
-  }
-  sys_msg = {"role": "system", "content": f"{_get_combined_system_prompt()}\n\n<context>\n{md_content}\n</context>"}
-  request_payload["messages"].insert(0, sys_msg)
+   if discussion_history and not _deepseek_history:
+    user_content = f"[DISCUSSION HISTORY]\n\n{discussion_history}\n\n---\n\n{user_message}"
+   else:
+    user_content = user_message
+   _deepseek_history.append({"role": "user", "content": user_content})
+  
  all_text_parts: list[str] = []
  _cumulative_tool_bytes = 0
-  round_idx = 0
-  while round_idx <= MAX_TOOL_ROUNDS + 1:
+  
+  for round_idx in range(MAX_TOOL_ROUNDS + 2):
+   current_api_messages: list[dict[str, Any]] = []
+   with _deepseek_history_lock:
+    for msg in _deepseek_history:
+     current_api_messages.append(msg)
+   
+   sys_msg = {"role": "system", "content": f"{_get_combined_system_prompt()}\n\n<context>\n{md_content}\n</context>"}
+   current_api_messages.insert(0, sys_msg)
+   
+   request_payload: dict[str, Any] = {
+    "model": _model,
+    "messages": current_api_messages,
+    "stream": stream,
+   }
+   
+   if not is_reasoner:
+    request_payload["temperature"] = _temperature
+    request_payload["max_tokens"] = _max_tokens
+    tools = _get_deepseek_tools()
+    if tools:
+     request_payload["tools"] = tools
+   
   events.emit("request_start", payload={"provider": "deepseek", "model": _model, "round": round_idx, "streaming": stream})
+   
   try:
-    response = requests.post(api_url, headers=headers, json=request_payload, timeout=60, stream=stream)
+    response = requests.post(api_url, headers=headers, json=request_payload, timeout=120, stream=stream)
    response.raise_for_status()
   except requests.exceptions.RequestException as e:
    raise _classify_deepseek_error(e) from e
+
+   assistant_text = ""
+   tool_calls_raw = []
+   reasoning_content = ""
+   finish_reason = "stop"
+   usage = {}
+
   if stream:
    aggregated_content = ""
    aggregated_tool_calls: list[dict[str, Any]] = []
@@ -1443,10 +1526,12 @@ def _send_deepseek(md_content: str, user_message: str, base_dir: str,
    reasoning_content = message.get("reasoning_content", "")
    finish_reason = choice.get("finish_reason", "stop")
    usage = response_data.get("usage", {})
+
   thinking_tags = ""
   if reasoning_content:
    thinking_tags = f"<thinking>\n{reasoning_content}\n</thinking>\n"
   full_assistant_text = thinking_tags + assistant_text
+   
   with _deepseek_history_lock:
    msg_to_store: dict[str, Any] = {"role": "assistant", "content": assistant_text}
    if reasoning_content:
@@ -1454,8 +1539,10 @@ def _send_deepseek(md_content: str, user_message: str, base_dir: str,
    if tool_calls_raw:
     msg_to_store["tool_calls"] = tool_calls_raw
    _deepseek_history.append(msg_to_store)
+   
   if full_assistant_text:
    all_text_parts.append(full_assistant_text)
+   
   _append_comms("IN", "response", {
     "round": round_idx,
     "stop_reason": finish_reason,
@@ -1464,6 +1551,7 @@ def _send_deepseek(md_content: str, user_message: str, base_dir: str,
     "usage": usage,
     "streaming": stream
    })
+   
   if finish_reason != "tool_calls" and not tool_calls_raw:
    break
   if round_idx > MAX_TOOL_ROUNDS:
@@ -1507,16 +1595,11 @@ def _send_deepseek(md_content: str, user_message: str, base_dir: str,
      "content": f"SYSTEM WARNING: Cumulative tool output exceeded {_MAX_TOOL_OUTPUT_BYTES // 1000}KB budget. Provide your final answer now."
     })
    _append_comms("OUT", "request", {"message": f"[TOOL OUTPUT BUDGET EXCEEDED: {_cumulative_tool_bytes} bytes]"})
+   
   with _deepseek_history_lock:
    for tr in tool_results_for_history:
     _deepseek_history.append(tr)
-   next_messages: list[dict[str, Any]] = []
-   with _deepseek_history_lock:
-    for msg in _deepseek_history:
-     next_messages.append(msg)
-   next_messages.insert(0, sys_msg)
-   request_payload["messages"] = next_messages
-   round_idx += 1
+   
  return "\n\n".join(all_text_parts) if all_text_parts else "(No text returned)"
 except Exception as e:
  raise _classify_deepseek_error(e) from e