stapp: stabilize long-session streaming and add runtime stats

frontends/stapp.py

@@ -13,8 +13,9 @@ sys.path.append(os.path.abspath(script_dir))
 
 import streamlit as st
 import time, json, re, threading, queue
+from functools import lru_cache
 from datetime import timedelta
-import agentmain
+import agentmain, llmcore
 from agentmain import GenericAgent
 import chatapp_common  # activate /continue command (monkey patches GeneraticAgent)
 from continue_cmd import handle_frontend_command, reset_conversation, list_sessions, extract_ui_messages
@@ -178,7 +179,7 @@ def render_sidebar():
         field-sizing: content; min-height: 1.6em !important; height: auto !important;
     }
     </style>""", unsafe_allow_html=True)
-    st.text_area("Loop prompt", value=st.session_state.get('loop_prompt_input', "继续" if LANG=='zh' else 'next'), key="loop_prompt_input", height=1)
+    st.text_area("Loop prompt", value=st.session_state.get('loop_prompt_input', "继续" if LANG=='zh' else 'next'), key="loop_prompt_input", height=68)
     if st.session_state.get('loop_enabled'):
         if st.button("⏹️ Stop Loop"):
             st.session_state.loop_enabled = False
@@ -215,7 +216,7 @@ def render_sidebar():
             _sp(); st.rerun(scope="app")
 with st.sidebar: render_sidebar()
 
-def fold_turns(text):
+def _fold_turns_impl(text):
     """Return list of segments: [{'type':'text','content':...}, {'type':'fold','title':...,'content':...}]"""
     # 先把4+反引号块替换为占位符，避免误切子agent嵌套的 LLM Running
     _ph = []
@@ -246,6 +247,13 @@ def fold_turns(text):
             segments.append({'type': 'fold', 'title': title, 'content': content})
         else: segments.append({'type': 'text', 'content': marker + content})
     return segments
+
+@st.cache_resource
+def _get_fold_turns():
+    """Keep parsed history across Streamlit reruns."""
+    return lru_cache(maxsize=128)(_fold_turns_impl)
+
+fold_turns = _get_fold_turns()
 _SUMMARY_TAG_RE = re.compile(r'<summary>.*?</summary>\s*', re.DOTALL)
 
 def render_segments(segments, suffix=''):
@@ -262,6 +270,8 @@ def _start_main_task(prompt):
     """Start a task whose queue can be drained across Streamlit reruns."""
     st.session_state.display_queue = agent.put_task(prompt, source="user")
     st.session_state.partial_response = ""
+    st.session_state.task_start_ts = time.time()
+    st.session_state.pop('task_end_ts', None)
 
 
 def _cancel_main_task():
@@ -283,32 +293,48 @@ def _poll_main_task(max_items=256):
             st.session_state.partial_response = item["next"]
         if "done" in item:
             done = item["done"]
+            st.session_state.task_end_ts = time.time()
             st.session_state.display_queue = None
             st.session_state.partial_response = ""
             break
     return done
 
 
-@st.fragment(run_every=timedelta(milliseconds=200))
-def render_main_stream():
-    """Poll and redraw briefly, leaving the script thread free for sidebar events."""
+@st.fragment(run_every=timedelta(seconds=1))
+def render_main_stream(frozen_host, live_slot, render_state):
+    """Append completed turns outside the fragment; redraw only the active turn."""
     done = _poll_main_task()
-    response = done if done is not None else st.session_state.get("partial_response", "")
-    # Avoid a marker-only gray line at a turn boundary. It reappears with content.
-    if done is None:
-        response = re.sub(r'\**LLM Running \(Turn \d+\) \.\.\.\**\s*$', '', response).rstrip()
-    with st.chat_message("assistant"):
-        render_segments(fold_turns(response), suffix="" if done is not None else " ▌")
-    if done is None: return
+    if done is not None:
+        if done:
+            st.session_state.messages.append({"role": "assistant", "content": done})
+            st.session_state.last_reply_time = int(time.time())
+            # ── 循环回调：回答完成戳醒 controller 决策(去程,现取最新objective) ──
+            if st.session_state.get('loop_enabled'):
+                b = get_controller()
+                b['obj'] = st.session_state.get('loop_prompt_input', ''); b['ready'] = False; b['job'] = b['epoch']; b['ev'].set()
+        # The final response is rendered once from history on the full-app rerun.
+        st.rerun(scope="app")
 
-    if response:
-        st.session_state.messages.append({"role": "assistant", "content": response})
-        st.session_state.last_reply_time = int(time.time())
-        # ── 循环回调：回答完成戳醒 controller 决策(去程,现取最新objective) ──
-        if st.session_state.get('loop_enabled'):
-            b = get_controller()
-            b['obj'] = st.session_state.get('loop_prompt_input', ''); b['ready'] = False; b['job'] = b['epoch']; b['ev'].set()
-    st.rerun(scope="app")
+    response = st.session_state.get("partial_response", "")
+    # Avoid a marker-only gray line at a turn boundary. It reappears with content.
+    response = re.sub(r'\**LLM Running \(Turn \d+\) \.\.\.\**\s*$', '', response).rstrip()
+    segments = fold_turns(response)
+    n_done = max(0, len(segments) - 1)
+    while render_state['frozen'] < n_done:
+        with frozen_host:
+            render_segments([segments[render_state['frozen']]])
+        render_state['frozen'] += 1
+    # This external slot keeps a stable delta path; only the growing final segment changes.
+    with live_slot.container():
+        render_segments([segments[-1]], suffix=" ▌")
+
+
+def mount_main_stream():
+    """Create persistent targets captured by the fragment until the next full rerun."""
+    with st.chat_message("assistant"):
+        frozen_host = st.container()
+        live_slot = st.empty()
+        render_main_stream(frozen_host, live_slot, {'frozen': 0})
 
 if not hasattr(agent, "_ui_messages"): agent._ui_messages = st.session_state.get("messages", [])
 if "messages" not in st.session_state: st.session_state.messages = agent._ui_messages
@@ -428,22 +454,41 @@ if prompt:
     if hasattr(agent, '_pet_req') and not prompt.startswith('/'): agent._pet_req('state=walk')
     with st.chat_message("user"): st.markdown(prompt)
     _start_main_task(prompt)
-    render_main_stream()
+    mount_main_stream()
 elif st.session_state.get('display_queue') is not None:
     # No new prompt but a task is mid-flight (typically a /btw rerun) — resume drain.
-    render_main_stream()
+    mount_main_stream()
 
-# Faux RUNNING badge: rendered in the main body (NOT the polling fragment) so its DOM node
-# survives fragment ticks — no animation restart/flicker. The app rerun on done wipes it.
-if st.session_state.get('display_queue') is not None:
+# Usage/time label remains after completion; while running only its fragment updates once/sec.
+_has_task_stats = 'task_start_ts' in st.session_state
+_is_running = st.session_state.get('display_queue') is not None
+if _has_task_stats:
+    def _render_stat_badge():
+        end_ts = time.time() if _is_running else st.session_state.get('task_end_ts', time.time())
+        secs = max(0, int(end_ts - st.session_state.task_start_ts))
+        stats = dict(llmcore.STATS)
+        short = lambda n: f'{n / 1000:.0f}k' if n >= 1000 else str(n)
+        usage = (f"{short(stats['ctx'])} chars·{stats['msgs']}msgs │ "
+                 f"in {short(stats.get('inp', 0))} toks·cached{short(stats.get('cached', 0))}·out{short(stats.get('out', 0))} │ "
+                 if 'ctx' in stats else '')
+        st.markdown(f'<div class="ga-stat-badge">{usage}{secs // 60}:{secs % 60:02d}</div>',
+                    unsafe_allow_html=True)
+    if _is_running:
+        @st.fragment(run_every=timedelta(seconds=1))
+        def _live_stat_badge(): _render_stat_badge()
+        _live_stat_badge()
+    else:
+        _render_stat_badge()
+
+if _has_task_stats or _is_running:
     st.markdown(
-        '<div class="ga-run-badge">RUNNING</div>'
-        '<style>.ga-run-badge{position:fixed;top:1.25rem;right:2.8rem;z-index:1000001;'
-        'padding:1px 10px;border-radius:12px;background:rgba(255,75,75,.10);'
-        'color:#ff4b4b;font-size:.72rem;font-weight:600;letter-spacing:.05em;'
-        'animation:gaPulse 1.2s ease-in-out infinite}'
-        '@keyframes gaPulse{50%{opacity:.35}}</style>',
-        unsafe_allow_html=True)
+        ('<div class="ga-run-badge">RUNNING</div>' if _is_running else '') +
+        '<style>.ga-run-badge,.ga-stat-badge{position:fixed;top:1.25rem;z-index:1000001;'
+        'padding:1px 10px;border-radius:12px;font-size:.72rem;font-weight:600;letter-spacing:.05em}'
+        '.ga-run-badge{right:2.8rem;background:rgba(255,75,75,.10);color:#ff4b4b;animation:gaPulse 1.2s ease-in-out infinite}'
+        '.ga-stat-badge{right:8.4rem;background:rgba(128,128,128,.1);color:#8a8a8a;'
+        'font-variant-numeric:tabular-nums;letter-spacing:0}'
+        '@keyframes gaPulse{50%{opacity:.35}}</style>', unsafe_allow_html=True)
 
 # ── 空闲自主行动：fragment 定时检测，替代 launch.pyw 的 idle_monitor ──
 @st.fragment(run_every=timedelta(minutes=1))

ga.py

@@ -208,15 +208,15 @@ def file_patch(path: str, old_content: str, new_content: str):
     """在文件中寻找唯一的 old_content 块并替换为 new_content"""
     path = str(Path(path).resolve())
     try:
-        if not os.path.exists(path): return {"status": "error", "msg": "文件不存在"}
+        if not os.path.exists(path): return {"status": "error", "msg": "file not found"}
         with open(path, 'r', encoding='utf-8') as f: full_text = f.read()
-        if not old_content: return {"status": "error", "msg": "old_content 为空，请确认 arguments"}
+        if not old_content: return {"status": "error", "msg": "old_content is blank"}
         count = full_text.count(old_content)
-        if count == 0: return {"status": "error", "msg": "未找到匹配的旧文本块，建议：先用 file_read 确认当前内容，再分小段进行 patch。若多次失败则询问用户，严禁自行使用 overwrite 或代码替换。"}
-        if count > 1: return {"status": "error", "msg": f"找到 {count} 处匹配，无法确定唯一位置。请提供更长、更具体的旧文本块以确保唯一性。建议：包含上下文行来增强特征，或分小段逐个修改。"}
+        if count == 0: return {"status": "error", "msg": "old_content is not found. Suggestion: use file_read to check current file content, make more small patches. Don't huge overwrite (even with code)"}
+        if count > 1: return {"status": "error", "msg": f"find {count} matches, unable to determine unique position. Provide a longer, more specific old_content to ensure uniqueness. Suggestion: include context lines to enhance features, or modify in smaller segments."}
         updated_text = full_text.replace(old_content, new_content)
         with open(path, 'w', encoding='utf-8', newline=_file_newline(path)) as f: f.write(updated_text)
-        return {"status": "success", "msg": "文件局部修改成功"}
+        return {"status": "success", "msg": "file patched successfully"}
     except Exception as e: return {"status": "error", "msg": str(e)}
 
 _read_dirs = set()
@@ -312,7 +312,8 @@ class GenericAgentHandler(BaseHandler):
         cwd = os.path.normpath(os.path.abspath(raw_path))
         code_cwd = os.path.normpath(self.cwd)
         maxlen = self._get_tool_maxlen(10000, args)
-        if code_type == 'python' and _arg(args, "inline_eval", False, bool):
+        if timeout > 600: result = '[ERROR] Timeout must be <= 600 seconds, run time-consuming code in background subprocess and monitor it!'
+        elif code_type == 'python' and _arg(args, "inline_eval", False, bool):
             ns = {'handler':self, 'parent':self.parent, 'history':json.dumps(self.parent.llmclient.backend.history)}
             old_cwd = os.getcwd()
             try:

llmcore.py

@@ -97,14 +97,16 @@ def safeprint(*argv):
     except OSError: pass
 print = safeprint
 
+STATS = {}
+
 def trim_messages_history(history, sess):
     cap = sess.context_win * 3
     target = int(cap * getattr(sess, 'trim_keep_rate', 0.6))
     kp = sess.trim_keep_prefix
     def cost(ms): return sum(len(json.dumps(m, ensure_ascii=False)) for m in ms)
     compress_history_tags(history, interval=getattr(sess, 'cut_msg_interval', 5))
-    print(f'[Debug] Current context: {cost(history)} chars, {len(history)} messages.')
-    if cost(history) <= cap: return
+    STATS.update(ctx=(c := cost(history)), msgs=len(history)); print(f'[Debug] Current context: {c} chars, {len(history)} messages.')
+    if c <= cap: return
     compress_history_tags(history, keep_recent=4, force=True)
     if cost(history) <= target: return
     pre, post = history[:kp], history[kp:]
@@ -120,7 +122,7 @@ def trim_messages_history(history, sess):
         gap = [{"role": "assistant", "content": _d()}] if m.get('role') == 'user' else [{"role": "user", "content": _d()}, {"role": "assistant", "content": _d()}]
         history[:] = pre + gap + post
     else: history[:] = pre + post
-    print(f'[Debug] Trimmed context, current: {cost(history)} chars, {len(history)} messages.')
+    STATS.update(ctx=(c := cost(history)), msgs=len(history)); print(f'[Debug] Trimmed context, current: {c} chars, {len(history)} messages.')
 
 def auto_make_url(base, path):
     b, p = base.rstrip('/'), path.strip('/')
@@ -189,7 +191,7 @@ def _parse_claude_sse(resp_lines):
             stop_reason = delta.get("stop_reason", stop_reason)
             out_usage = evt.get("usage", {})
             out_tokens = out_usage.get("output_tokens", 0)
-            if out_tokens: print(f"[Output] tokens={out_tokens} stop_reason={stop_reason}")
+            if out_tokens: STATS['out'] = out_tokens; print(f"[Output] tokens={out_tokens} stop_reason={stop_reason}")
         elif evt_type == "message_stop": got_message_stop = True
         elif evt_type == "error":
             err = evt.get("error", {})
@@ -333,7 +335,10 @@ def _record_usage(usage, api_mode):
         if out: print(f"[Output] tokens={out}")
     elif api_mode == 'messages':
         ci, cr, inp = usage.get("cache_creation_input_tokens", 0), usage.get("cache_read_input_tokens", 0), usage.get("input_tokens", 0)
+        cached, out = cr, 0
         print(f"[Cache] input={inp} creation={ci} read={cr}")
+    else: return
+    STATS.update(inp=inp, cached=cached, out=out)
     
 def _parse_openai_json(data, api_mode="chat_completions"):
     blocks = []