FIX SOME BUGS

server/server.py

@@ -24,6 +24,7 @@ import cmd
 import hashlib
 import hmac
 import json
+import os
 import sqlite3
 import sys
 import threading
@@ -35,7 +36,9 @@ from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
 from pathlib import Path
 
 ROOT = Path(__file__).resolve().parent
-CONFIG_FILE = ROOT / "config.toml"
+# SJCODE_CONFIG lets you run a second daemon (tests, staging) without
+# touching the regular config; the database is created next to that file.
+CONFIG_FILE = Path(os.environ.get("SJCODE_CONFIG") or (ROOT / "config.toml"))
 DB_FILE = ROOT / "sjcode.db"  # overwritten by config value once loaded
 
 PEPPER = b"sjcode-v1"
@@ -43,7 +46,8 @@ SESSION_TTL = 24 * 3600  # login token valid for one day
 
 CONF: dict = {
     "server": {"host": "127.0.0.1", "port": 8420},
-    "ollama": {"base_url": "http://127.0.0.1:11434", "model": "llama3.2"},
+    "ollama": {"base_url": "http://127.0.0.1:11434", "model": "llama3.2",
+               "num_ctx": 16384},
     "database": {"path": "sjcode.db"},
     "limits": {"user": 120, "admin": -1},
 }
@@ -66,7 +70,7 @@ def load_config() -> None:
     for section, values in loaded.items():
         if isinstance(values, dict) and isinstance(CONF.get(section), dict):
             CONF[section].update(values)
-    DB_FILE = ROOT / CONF["database"]["path"]
+    DB_FILE = CONFIG_FILE.parent / CONF["database"]["path"]
 
 
 # --------------------------------------------------------------------------
@@ -232,15 +236,62 @@ def ollama_json(path: str, payload: dict | None = None, timeout: int = 300) -> t
         return 502, {"error": f"ollama not reachable: {exc}"}
 
 
-def ollama_generate(prompt: str, model: str | None) -> tuple[int, str]:
-    status, body = ollama_json(
-        "/api/generate",
-        {
-            "model": model or CONF["ollama"]["model"],
-            "prompt": prompt,
-            "stream": False,
-        },
+def ollama_stream(path: str, payload: dict) -> tuple[int, object]:
+    """POST to ollama and return the raw response to iterate line by line."""
+    url = CONF["ollama"]["base_url"].rstrip("/") + path
+    data = json.dumps(payload).encode()
+    req = urllib.request.Request(
+        url, data=data, headers={"Content-Type": "application/json"}
     )
+    try:
+        resp = urllib.request.urlopen(req, timeout=600)
+    except urllib.error.HTTPError as exc:
+        try:
+            body = json.loads(exc.read().decode())
+        except Exception:
+            body = {"error": str(exc)}
+        return exc.code, str(body.get("error", "ollama error"))
+    except (urllib.error.URLError, TimeoutError, ConnectionError) as exc:
+        return 502, {"error": f"ollama not reachable: {exc}"}
+    return resp.status, resp  # caller iterates the lines and closes it
+
+
+def ollama_generate(prompt: str, model: str | None, stream: bool = False):
+    """Generate a reply. Returns (status, text) or (200, token-iterator)."""
+    options = {"num_ctx": int(CONF["ollama"].get("num_ctx", 16384))}
+    payload = {
+        "model": model or CONF["ollama"]["model"],
+        "prompt": prompt,
+        "stream": stream,
+        "options": options,
+    }
+    if stream:
+        status, body = ollama_stream("/api/generate", payload)
+        if status != 200:
+            return status, str(body)
+
+        def tokens():
+            try:
+                for raw in body:
+                    line = raw.decode() if isinstance(raw, bytes) else raw
+                    if not line.strip():
+                        continue
+                    try:
+                        obj = json.loads(line)
+                    except json.JSONDecodeError:
+                        continue
+                    if obj.get("error"):
+                        yield f"\n[ollama error: {obj['error']}]"
+                        return
+                    piece = obj.get("response", "")
+                    if piece:
+                        yield piece
+            finally:
+                body.close()
+
+        return 200, tokens()
+
+    status, body = ollama_json("/api/generate", payload)
     if status != 200:
         return status, str(body.get("error", "ollama error"))
     return 200, str(body.get("response", ""))
@@ -259,7 +310,8 @@ def list_models() -> tuple[int, list[str]]:
 # --------------------------------------------------------------------------
 
 class ApiHandler(BaseHTTPRequestHandler):
-    server_version = "sjcode-daemon/0.1"
+    server_version = "sjcode-daemon/0.2"
+    protocol_version = "HTTP/1.1"  # required for chunked streamed replies
 
     # -- helpers ----------------------------------------------------------
     def _json(self, status: int, payload: dict) -> None:
@@ -277,8 +329,48 @@ class ApiHandler(BaseHTTPRequestHandler):
         except (ValueError, json.JSONDecodeError):
             return {}
 
-    def log_message(self, fmt: str, *args) -> None:  # quiet by default
-        return
+    def _respond_stream(self, tokens, model: str, started: float,
+                        username: str) -> None:
+        """Stream a reply as JSON lines: {"chunk": ...} ... {"response": ...}."""
+        self.send_response(200)
+        self.send_header("Content-Type", "application/x-ndjson")
+        self.send_header("Transfer-Encoding", "chunked")
+        self.end_headers()
+
+        def send_line(obj: dict) -> None:
+            data = (json.dumps(obj) + "\n").encode()
+            self.wfile.write(f"{len(data):x}\r\n".encode() + data + b"\r\n")
+            self.wfile.flush()
+
+        parts: list[str] = []
+        abandoned = False
+        try:
+            for piece in tokens:
+                parts.append(piece)
+                send_line({"chunk": piece})
+        except (BrokenPipeError, ConnectionResetError):
+            abandoned = True  # client went away; still bill the usage below
+        except OSError as exc:
+            parts.append(f"\n[stream error: {exc}]")
+            send_line({"chunk": f"\n[stream error: {exc}]"})
+        finally:
+            text = "".join(parts)
+            elapsed = time.monotonic() - started
+            used_min = elapsed / 60.0
+            add_used_minutes(username, used_min)
+            if not abandoned:
+                left_now = remaining_minutes(username)
+                send_line({
+                    "response": text,
+                    "model": model,
+                    "used_minutes": round(used_min, 4),
+                    "remaining_minutes": None if left_now == float("inf")
+                    else round(left_now, 2),
+                })
+                self.wfile.write(b"0\r\n\r\n")
+                self.wfile.flush()
+                print(f"[chat] {username}: {len(text)} chars in "
+                      f"{elapsed:.1f}s (streamed)")
 
     # -- routes -----------------------------------------------------------
     def do_GET(self) -> None:  # noqa: N802 (http.server API)
@@ -353,9 +445,15 @@ class ApiHandler(BaseHTTPRequestHandler):
             return
 
         model = str(data.get("model") or "") or CONF["ollama"]["model"]
+        want_stream = bool(data.get("stream"))
 
         started = time.monotonic()
-        status, answer = ollama_generate(message, model)
+        status, answer = ollama_generate(message, model, stream=want_stream)
+
+        if want_stream and status == 200:
+            self._respond_stream(answer, model, started, username)
+            return
+
         elapsed = time.monotonic() - started
 
         if status != 200:
@@ -375,6 +473,9 @@ class ApiHandler(BaseHTTPRequestHandler):
             },
         )
 
+    def log_message(self, fmt: str, *args) -> None:  # quiet by default
+        return
+
 
 # --------------------------------------------------------------------------
 # admin CLI