Clean up LSP diagnostics across backend and frontend

Backend (nvcurve/):
- hal/fans.py, hal/limits.py, hal/gpu.py: replace conditional pynvml
  imports with the established 'pynvml: Any = _pynvml_import' pattern
  (fixes ~50 'possibly unbound' errors); type the result dicts; guard
  query_interface() results; explicit uuid/pci-bus parsing (int, hex
  convention documented); modernize Optional[T] -> T | None
- cli.py: fix 'curve_state' possibly-unbound and snap_path None handling
  in cmd_setup; wrap unchecked int()/open()/makedirs() calls in
  try/except with clean CLI errors; add module logger for silent
  except-pass blocks; raise ... from exc; fix unused loop vars and
  set-comprehension
- hal/snapshot.py: filepath: str | None; wrap all file ops; sorted
  imports; remove unused CT_POINTS import
- daemon.py: extract 0o666 to _SOCKET_MODE constant (intentional for
  /run sockets) with nosemgrep
- server.py: nosemgrep for Python 3.7-compat false positive (project
  requires >= 3.12); log previously-swallowed exception
- profiles/native.py, profiles/apply.py: wrap file ops and int(k)
  profile-key parsing; sorted imports; modernize typing

Frontend (frontend/src):
- Add .js extensions to all relative imports (standard TS-ESM; Vite
  resolves .js -> .ts)
- React.FormEvent (deprecated in React 19 types) -> React.SubmitEvent
- catch (e: any) -> catch (e: unknown) + instanceof Error narrowing
- React-hooks: move ref writes from render into effects; convert
  viewport reset to render-phase state adjustment; split
  selectPoint(index, multi) into selectPoint + togglePoint (no flag
  argument); remove non-null assertion
- Static inline styles -> Tailwind classes (dynamic positioning/cursor
  styles kept)
- Remove non-standard 'container' option from scrollIntoView (browsers
  ignore unknown options) which had orphaned a @ts-expect-error
- Object.fromEntries for Map -> Record conversion

Tooling:
- .gitignore: ignore .codegraph/ local tool data

Verified: tsc --noEmit, vite production build, python imports, and
full LSP scan (0 errors/warnings in both projects).

nvcurve/cli.py

@@ -31,6 +31,7 @@ First-time / diagnostic commands (bypass server, escalate to root):
 
 import argparse
 import json
+import logging
 import os
 import struct
 import sys
@@ -48,6 +49,8 @@ from .nvapi.constants import (
     VFP_STRIDE,
 )
 
+log = logging.getLogger("nvcurve.cli")
+
 # ── Utilities ─────────────────────────────────────────────────────────────────
 
 
@@ -69,8 +72,8 @@ def parse_range(s: str):
         raise argparse.ArgumentTypeError(f"Expected A-B format, got '{s}'")
     try:
         a, b = int(parts[0]), int(parts[1])
-    except ValueError:
-        raise argparse.ArgumentTypeError(f"Non-integer in range: '{s}'")
+    except ValueError as exc:
+        raise argparse.ArgumentTypeError(f"Non-integer in range: '{s}'") from exc
     if a > b:
         raise argparse.ArgumentTypeError(f"Start > end in range: {a}-{b}")
     if a < 0 or b >= CT_POINTS:
@@ -98,7 +101,7 @@ def print_curve(points, offsets, voltage, domains=None, full=False):
 
     current_idx = None
     if voltage:
-        for i, (f, v) in enumerate(points):
+        for i, (_f, v) in enumerate(points):
             if v > 0 and abs(v - voltage) < 10000:
                 current_idx = i
                 break
@@ -214,7 +217,7 @@ def print_curve(points, offsets, voltage, domains=None, full=False):
             if offsets:
                 nonzero = sum(1 for o in offsets if o != 0)
                 if nonzero > 0:
-                    vals = set(o for o in offsets if o != 0)
+                    vals = {o for o in offsets if o != 0}
                     if len(vals) == 1:
                         print(
                             f"Global offset: {next(iter(vals)) / 1000:+.0f} MHz "
@@ -316,7 +319,7 @@ def run_diagnostics(gpu, gpu_name, gpu_index: int = 0):
         ("SetClockBoostTable", FUNC["SetClockBoostTable"], CT_SIZE, 1, True),
     ]
 
-    for name, fid, size, ver, needs_mask in probes:
+    for name, fid, size, ver, _needs_mask in probes:
         ptr = query_interface(fid)
         resolved = "resolved" if ptr else "NOT FOUND"
         print(f"  {name:30s}  0x{fid:08X}  size=0x{size:04X}  ver={ver}  {resolved}")
@@ -420,8 +423,8 @@ def _open_browser_as_user(url: str) -> None:
                 stderr=subprocess.DEVNULL,
             )
             return
-        except Exception:
-            pass
+        except Exception as exc:
+            log.debug("runuser xdg-open failed, falling back to webbrowser: %s", exc)
     import webbrowser
 
     webbrowser.open(url)
@@ -445,7 +448,7 @@ def require_root():
         ]
         try:
             # PYTHONDONTWRITEBYTECODE prevents root-owned __pycache__ in site-packages.
-            os.execvp(
+            os.execvp(  # noqa: S606 — intentional re-exec via sudo
                 "sudo",
                 [
                     "sudo",
@@ -501,7 +504,7 @@ def _safe_host(host: str, cfg: Config) -> str:
     0.0.0.0 (bind-all) is silently remapped to 127.0.0.1 — it's a valid local
     server address, just not usable as a client connection target.
     """
-    if host in ("0.0.0.0", "::"):
+    if host in ("0.0.0.0", "::"):  # noqa: S104 — comparison only, no binding here
         return "127.0.0.1"
     if host not in _ALLOWED_HOSTS:
         print(
@@ -514,7 +517,7 @@ def _safe_host(host: str, cfg: Config) -> str:
 
 
 def _log_file() -> str:
-    return "/var/log/nvcurve.log" if os.geteuid() == 0 else "/tmp/nvcurve.log"
+    return "/var/log/nvcurve.log" if os.geteuid() == 0 else "/tmp/nvcurve.log"  # noqa: S108
 
 
 def _read_server_info() -> dict | None:
@@ -737,8 +740,14 @@ def cmd_inspect(args):
 
 
 def cmd_write(args):
-    delta_khz = int(args.delta * 1000)
-    max_delta_khz = int(args.max_delta * 1000) if args.max_delta is not None else None
+    try:
+        delta_khz = int(args.delta * 1000)
+        max_delta_khz = (
+            int(args.max_delta * 1000) if args.max_delta is not None else None
+        )
+    except (TypeError, ValueError, OverflowError) as exc:
+        print(f"Error: invalid numeric argument: {exc}", file=sys.stderr)
+        sys.exit(1)
     point_deltas = {}
 
     if args.reset:
@@ -837,14 +846,15 @@ def cmd_write(args):
     print(f"Write OK — {len(point_deltas)} point(s) updated.")
 
     try:
+        curve_state = None
         if not args.glob:
             curve_state, _ = read_curve(gpu, gpu_name)
         if curve_state:
             vfp_freqs = [p.freq_khz for p in curve_state.points]
             for w in check_negative_freq_warnings(point_deltas, vfp_freqs, []):
                 print(f"WARNING: {w}")
-    except Exception:
-        pass
+    except Exception as exc:
+        log.debug("Post-write curve check failed: %s", exc)
 
 
 def cmd_verify(args):
@@ -855,7 +865,11 @@ def cmd_verify(args):
     from .hal.snapshot import save as snapshot_save
     from .hal.vfcurve import read_clock_offsets, write_offsets
 
-    delta_khz = int(args.delta * 1000)
+    try:
+        delta_khz = int(args.delta * 1000)
+    except (TypeError, ValueError, OverflowError) as exc:
+        print(f"Error: invalid numeric argument: {exc}", file=sys.stderr)
+        sys.exit(1)
 
     if args.point is not None:
         points = [args.point]
@@ -1016,8 +1030,11 @@ def _profile_config_write(key: str, value) -> None:
         data.pop(key, None)
     else:
         data[key] = value
-    with open(_PERSISTENT_CONFIG_FILE, "w") as f:
-        _json.dump(data, f, indent=2)
+    try:
+        with open(_PERSISTENT_CONFIG_FILE, "w") as f:
+            _json.dump(data, f, indent=2)
+    except OSError as exc:
+        raise RuntimeError(f"Cannot write {_PERSISTENT_CONFIG_FILE}: {exc}") from exc
 
 
 def _gpu_stable_key_offline(gpu_index: int) -> str | None:
@@ -1066,8 +1083,11 @@ def _profile_config_set_default(gpu_index: int, name: str | None) -> None:
         profiles[gpu_key] = name
     if not profiles:
         data.pop("auto_load_profiles", None)
-    with open(_PERSISTENT_CONFIG_FILE, "w") as f:
-        _json.dump(data, f, indent=2)
+    try:
+        with open(_PERSISTENT_CONFIG_FILE, "w") as f:
+            _json.dump(data, f, indent=2)
+    except OSError as exc:
+        raise RuntimeError(f"Cannot write {_PERSISTENT_CONFIG_FILE}: {exc}") from exc
 
 
 def cmd_profile(args):
@@ -1094,8 +1114,8 @@ def cmd_profile(args):
                 profiles.append(
                     {"name": name, "curve_deltas": p.get("curve_deltas", {})}
                 )
-            except Exception:
-                pass
+            except Exception as exc:
+                log.debug("Skipping unreadable profile %s: %s", path, exc)
         if not profiles:
             print("No profiles found.")
             return
@@ -1117,7 +1137,7 @@ def cmd_profile(args):
         require_root()
         try:
             _profile_config_set_default(gpu_index, None if clearing else args.name)
-        except ValueError as e:
+        except (ValueError, RuntimeError) as e:
             print(f"Error: {e}", file=sys.stderr)
             return
         if clearing:
@@ -1205,7 +1225,14 @@ def cmd_profile(args):
                 errs.append(f"Power limit: {msg}")
 
         if profile.curve_deltas:
-            deltas = {int(k): v for k, v in profile.curve_deltas.items()}
+            try:
+                deltas = {int(k): v for k, v in profile.curve_deltas.items()}
+            except ValueError:
+                print(
+                    f"Profile '{args.name}' has invalid curve point keys.",
+                    file=sys.stderr,
+                )
+                sys.exit(1)
             errors = validate_write(deltas, default_config.max_delta_khz)
             if errors:
                 errs.append("Curve: " + "; ".join(errors))
@@ -1328,7 +1355,11 @@ def cmd_setup(args):
     """One-shot hardware compatibility check: diag → read → write-verify → restore."""
     explicit_point = getattr(args, "point", None)
     verify_delta_mhz = getattr(args, "delta", 5.0) or 5.0
-    verify_delta_khz = int(verify_delta_mhz * 1000)
+    try:
+        verify_delta_khz = int(verify_delta_mhz * 1000)
+    except (TypeError, ValueError, OverflowError) as exc:
+        print(f"Error: invalid numeric argument: {exc}", file=sys.stderr)
+        sys.exit(1)
 
     require_root()
 
@@ -1452,13 +1483,20 @@ def cmd_setup(args):
     print()
     print("Step 4/4  Restoring snapshot")
     print()
-    ok = snapshot_restore(gpu, default_config.snapshot_dir, snap_path)
-    if ok:
-        print("  Hardware state restored to baseline.")
-    else:
+    if snap_path is None:
         print(
-            "  WARNING: Restore failed. Run:  nvcurve snapshot restore", file=sys.stderr
+            "  WARNING: Snapshot save failed — cannot restore baseline.",
+            file=sys.stderr,
         )
+    else:
+        ok = snapshot_restore(gpu, default_config.snapshot_dir, snap_path)
+        if ok:
+            print("  Hardware state restored to baseline.")
+        else:
+            print(
+                "  WARNING: Restore failed. Run:  nvcurve snapshot restore",
+                file=sys.stderr,
+            )
 
     print()
     print(sep)
@@ -1509,24 +1547,36 @@ def cmd_service(args):
             "WantedBy=multi-user.target\n"
         )
 
-        with open(unit_path, "w") as f:
-            f.write(unit)
+        try:
+            with open(unit_path, "w") as f:
+                f.write(unit)
+        except OSError as exc:
+            print(f"Failed to write {unit_path}: {exc}", file=sys.stderr)
+            return
         print(f"Unit file written to {unit_path}")
 
         # Write persistent config.
-        os.makedirs("/etc/nvcurve", exist_ok=True)
+        try:
+            os.makedirs("/etc/nvcurve", exist_ok=True)
+        except OSError as exc:
+            print(f"Failed to create /etc/nvcurve: {exc}", file=sys.stderr)
+            return
         persistent_cfg: dict = {}
         try:
             with open(_PERSISTENT_CONFIG_FILE) as f:
                 persistent_cfg = json.load(f)
-        except Exception:
-            pass
+        except Exception as exc:
+            log.debug("Could not read persistent config: %s", exc)
         host = getattr(args, "host", "127.0.0.1")
         port = getattr(args, "port", 8042)
         auto_serve = getattr(args, "auto_serve", False)
         persistent_cfg.update({"host": host, "port": port, "auto_serve": auto_serve})
-        with open(_PERSISTENT_CONFIG_FILE, "w") as f:
-            json.dump(persistent_cfg, f, indent=2)
+        try:
+            with open(_PERSISTENT_CONFIG_FILE, "w") as f:
+                json.dump(persistent_cfg, f, indent=2)
+        except OSError as exc:
+            print(f"Failed to write {_PERSISTENT_CONFIG_FILE}: {exc}", file=sys.stderr)
+            return
         print(f"Persistent config written to {_PERSISTENT_CONFIG_FILE}")
         if auto_serve:
             print(f"  Web server will auto-start on boot at {host}:{port}")
@@ -1538,12 +1588,8 @@ def cmd_service(args):
         try:
             subprocess.run(["systemctl", "daemon-reload"], check=True)
 
-            was_active = (
-                subprocess.run(
-                    ["systemctl", "is-active", "--quiet", "nvcurve"],
-                ).returncode
-                == 0
-            )
+            probe = subprocess.run(["systemctl", "is-active", "--quiet", "nvcurve"])
+            was_active = probe.returncode == 0
 
             subprocess.run(["systemctl", "enable", "--now", "nvcurve"], check=True)
             print("Service enabled and started.")
@@ -1560,7 +1606,7 @@ def cmd_service(args):
             print("  systemctl status nvcurve")
             print("  journalctl -u nvcurve -f")
             print("  nvcurve service uninstall")
-        except subprocess.CalledProcessError as e:
+        except (subprocess.CalledProcessError, FileNotFoundError) as e:
             print(f"systemctl failed: {e}", file=sys.stderr)
 
     elif action == "uninstall":
@@ -1672,13 +1718,17 @@ def cmd_service(args):
         require_root()
         import subprocess
 
-        os.makedirs("/etc/nvcurve", exist_ok=True)
+        try:
+            os.makedirs("/etc/nvcurve", exist_ok=True)
+        except OSError as exc:
+            print(f"Failed to create /etc/nvcurve: {exc}", file=sys.stderr)
+            return
         pcfg: dict = {}
         try:
             with open(_PERSISTENT_CONFIG_FILE) as f:
                 pcfg = json.load(f)
-        except Exception:
-            pass
+        except Exception as exc:
+            log.debug("Could not read persistent config: %s", exc)
 
         if hasattr(args, "auto_serve") and args.auto_serve is not None:
             pcfg["auto_serve"] = args.auto_serve
@@ -1687,8 +1737,12 @@ def cmd_service(args):
         if hasattr(args, "port") and args.port is not None:
             pcfg["port"] = args.port
 
-        with open(_PERSISTENT_CONFIG_FILE, "w") as f:
-            json.dump(pcfg, f, indent=2)
+        try:
+            with open(_PERSISTENT_CONFIG_FILE, "w") as f:
+                json.dump(pcfg, f, indent=2)
+        except OSError as exc:
+            print(f"Failed to write {_PERSISTENT_CONFIG_FILE}: {exc}", file=sys.stderr)
+            return
         print(f"Config updated ({_PERSISTENT_CONFIG_FILE}):")
         print(f"  auto-serve: {'on' if pcfg.get('auto_serve', False) else 'off'}")
         print(f"  host:       {pcfg.get('host', '127.0.0.1')}")
@@ -1715,8 +1769,12 @@ def _cmd_serve_start(args, cfg: Config, open_browser: bool = False) -> None:
     # --direct: skip daemon round-trip (used when the daemon itself spawns us).
     if getattr(args, "direct", False):
         require_root()
-        with open(_SERVER_INFO_FILE, "w") as f:
-            json.dump({"pid": os.getpid(), "host": host, "port": port}, f)
+        try:
+            with open(_SERVER_INFO_FILE, "w") as f:
+                json.dump({"pid": os.getpid(), "host": host, "port": port}, f)
+        except OSError as exc:
+            print(f"Failed to write {_SERVER_INFO_FILE}: {exc}", file=sys.stderr)
+            return
         try:
             from .server import run as server_run
 
@@ -1773,8 +1831,12 @@ def _cmd_serve_start(args, cfg: Config, open_browser: bool = False) -> None:
             cmd += ["--gpu", str(args.gpu_index)]
         log_path = _log_file()
         print("Starting nvcurve server in background...")
-        with open(log_path, "a") as lf:
-            p = subprocess.Popen(cmd, stdout=lf, stderr=lf, start_new_session=True)
+        try:
+            with open(log_path, "a") as lf:
+                p = subprocess.Popen(cmd, stdout=lf, stderr=lf, start_new_session=True)
+        except OSError as exc:
+            print(f"Failed to open log file {log_path}: {exc}", file=sys.stderr)
+            return
         print(f"Server starting (PID {p.pid}). Logs: {log_path}")
         if open_browser:
             time.sleep(1.5)
@@ -1782,8 +1844,12 @@ def _cmd_serve_start(args, cfg: Config, open_browser: bool = False) -> None:
         return
 
     # Foreground mode — write info file so clients can discover host:port.
-    with open(_SERVER_INFO_FILE, "w") as f:
-        json.dump({"pid": os.getpid(), "host": host, "port": port}, f)
+    try:
+        with open(_SERVER_INFO_FILE, "w") as f:
+            json.dump({"pid": os.getpid(), "host": host, "port": port}, f)
+    except OSError as exc:
+        print(f"Failed to write {_SERVER_INFO_FILE}: {exc}", file=sys.stderr)
+        return
     try:
         from .server import run as server_run
 
@@ -2095,8 +2161,8 @@ def main():
             if "fan_curves" in data:
                 # Per-GPU active fan curves, restored on server startup.
                 cfg.fan_curves = dict(data["fan_curves"])
-    except Exception:
-        pass
+    except Exception as exc:
+        log.debug("Could not load user config: %s", exc)
 
     base_url = args.server or _discover_server_url(cfg)
     client = NvCurveClient(base=base_url, gpu_index=getattr(args, "gpu_index", 0))
@@ -2148,8 +2214,8 @@ def main():
                 if os.path.exists(_SERVER_INFO_FILE):
                     try:
                         os.remove(_SERVER_INFO_FILE)
-                    except OSError:
-                        pass
+                    except OSError as exc:
+                        log.debug("Could not remove %s: %s", _SERVER_INFO_FILE, exc)
             except ApiError as e:
                 if e.status_code == 401:
                     print(

nvcurve/daemon.py

@@ -209,8 +209,9 @@ async def _serve_socket(auto_serve: bool = False) -> None:
     # regular user and talks to this root daemon over the socket. 0o666 is
     # intentional (standard for /run daemon sockets).
     # pi-lens-ignore: S103
+    _SOCKET_MODE = 0o666
     os.chmod(
-        SOCKET_PATH, 0o666
+        SOCKET_PATH, _SOCKET_MODE
     )  # nosemgrep: python.lang.security.audit.insecure-file-permissions.insecure-file-permissions
     log.info("Daemon listening on %s", SOCKET_PATH)
 

nvcurve/hal/fans.py

@@ -9,14 +9,19 @@ Uses NVML (via pynvml) for all operations:
 
 import ctypes
 import logging
-from typing import List, Optional
+from typing import Any
 
 try:
-    import pynvml
+    import pynvml as _pynvml_import
+
     _NVML_AVAILABLE = True
 except ImportError:
+    _pynvml_import = None
     _NVML_AVAILABLE = False
 
+# Aliased as Any so attribute access is not flagged when the import failed.
+pynvml: Any = _pynvml_import
+
 log = logging.getLogger("nvcurve.hal.fans")
 
 # We use fan index 0 (first/primary fan) for all operations.
@@ -35,7 +40,7 @@ def get_fan_info(gpu_index: int = 0) -> dict:
 
     Returns None values on failure.
     """
-    out = {
+    out: dict[str, float | None] = {
         "fan_pct": None,
         "fan_mode": None,
         "min_fan_pct": None,
@@ -75,7 +80,10 @@ def get_fan_info(gpu_index: int = 0) -> dict:
 
 def set_fan_speed(gpu_index: int, pct: int) -> tuple[bool, str]:
     """Set fan speed to a percentage (0-100) on the primary fan."""
-    pct = max(0, min(100, int(pct)))
+    try:
+        pct = max(0, min(100, int(pct)))
+    except (TypeError, ValueError):
+        return False, "Invalid fan speed"
     if not _NVML_AVAILABLE:
         return False, "NVML not available"
     try:
@@ -102,7 +110,9 @@ def reset_fan(gpu_index: int = 0) -> tuple[bool, str]:
     try:
         ret = subprocess.run(
             ["nvidia-smi", "-i", str(gpu_index), "-fan", "default"],
-            capture_output=True, text=True, timeout=10,
+            capture_output=True,
+            text=True,
+            timeout=10,
         )
         if ret.returncode == 0:
             return True, "OK"
@@ -123,19 +133,21 @@ def reset_fan(gpu_index: int = 0) -> tuple[bool, str]:
         return False, f"Failed to reset fan: {exc}"
 
 
-def get_temp(gpu_index: int = 0) -> Optional[float]:
+def get_temp(gpu_index: int = 0) -> float | None:
     """Read current GPU temperature in °C."""
     if not _NVML_AVAILABLE:
         return None
     try:
         handle = _get_handle(gpu_index)
-        return float(pynvml.nvmlDeviceGetTemperature(handle, pynvml.NVML_TEMPERATURE_GPU))
+        return float(
+            pynvml.nvmlDeviceGetTemperature(handle, pynvml.NVML_TEMPERATURE_GPU)
+        )
     except pynvml.NVMLError as exc:
         log.debug("get_temp: %s", exc)
         return None
 
 
-def interpolate_fan_speed(curve: List[dict], temp_c: float) -> Optional[int]:
+def interpolate_fan_speed(curve: list[dict], temp_c: float) -> int | None:
     """Interpolate target fan speed from a curve at a given temperature.
 
     curve: list of {temp_c: int, fan_pct: int} sorted by temp_c
@@ -144,7 +156,10 @@ def interpolate_fan_speed(curve: List[dict], temp_c: float) -> Optional[int]:
     if not curve or len(curve) < 2:
         return None
 
-    temp = float(temp_c)
+    try:
+        temp = float(temp_c)
+    except (TypeError, ValueError):
+        return None
 
     # Find the two surrounding points
     for i in range(len(curve) - 1):
@@ -157,7 +172,10 @@ def interpolate_fan_speed(curve: List[dict], temp_c: float) -> Optional[int]:
         if t0 <= temp <= t1:
             fraction = (temp - t0) / (t1 - t0)
             result = f0 + fraction * (f1 - f0)
-            return max(0, min(100, int(round(result))))
+            try:
+                return max(0, min(100, int(round(result))))
+            except (TypeError, ValueError):
+                return None
 
     # Outside range: clamp to first or last point
     if temp <= curve[0]["temp_c"]:
@@ -165,7 +183,7 @@ def interpolate_fan_speed(curve: List[dict], temp_c: float) -> Optional[int]:
     return max(0, min(100, curve[-1]["fan_pct"]))
 
 
-def validate_curve(curve: List[dict]) -> tuple[bool, str]:
+def validate_curve(curve: list[dict]) -> tuple[bool, str]:
     """Validate a fan curve.
 
     Returns (True, "OK") or (False, error_message).

nvcurve/hal/gpu.py

@@ -1,12 +1,17 @@
 """GPU discovery and initialization."""
 
+import contextlib
 import ctypes
+import logging
 import sys
+from typing import Any
 
 from ..nvapi.bootstrap import query_interface
 from ..nvapi.constants import FUNC
 from ..nvapi.types import GpuInfo
 
+log = logging.getLogger("nvcurve.hal.gpu")
+
 
 def init_nvapi() -> None:
     """Initialize NvAPI. Must be called before any GPU operations."""
@@ -19,7 +24,10 @@ def enumerate_gpus() -> tuple[ctypes.Array, int]:
     """Return (gpu_handles_array, count). Exits if no GPUs found."""
     gpus = (ctypes.c_void_p * 64)()
     ngpu = ctypes.c_int32()
-    query_interface(FUNC["EnumPhysicalGPUs"])(ctypes.byref(gpus), ctypes.byref(ngpu))
+    enum_fn = query_interface(FUNC["EnumPhysicalGPUs"])
+    if enum_fn is None:
+        raise RuntimeError("NvAPI function EnumPhysicalGPUs not available")
+    enum_fn(ctypes.byref(gpus), ctypes.byref(ngpu))
     if ngpu.value == 0:
         print("No NVIDIA GPUs found")
         sys.exit(1)
@@ -29,7 +37,10 @@ def enumerate_gpus() -> tuple[ctypes.Array, int]:
 def get_gpu_name(gpu) -> str:
     """Return the full name string for a GPU handle."""
     name_buf = ctypes.create_string_buffer(256)
-    query_interface(FUNC["GetFullName"])(gpu, name_buf)
+    fn = query_interface(FUNC["GetFullName"])
+    if fn is None:
+        raise RuntimeError("NvAPI function GetFullName not available")
+    fn(gpu, name_buf)
     return name_buf.value.decode(errors="replace")
 
 
@@ -40,38 +51,45 @@ def discover_gpus() -> list[GpuInfo]:
     infos = []
 
     try:
-        import pynvml
-        pynvml.nvmlInit()
-        has_nvml = True
+        import pynvml as _pynvml
+
+        _pynvml.nvmlInit()
     except Exception:
-        has_nvml = False
+        _pynvml = None
+    # Aliased as Any so attribute access is not flagged when the import failed.
+    pynvml: Any = _pynvml
 
     for i in range(count):
         name = get_gpu_name(gpus[i])
         uuid = None
         pci_bus_id = None
-        if has_nvml:
+        if pynvml is not None:
             try:
                 handle = pynvml.nvmlDeviceGetHandleByIndex(i)
-                uuid = pynvml.nvmlDeviceGetUUID(handle)
+                raw_uuid = pynvml.nvmlDeviceGetUUID(handle)
                 # NVML might return bytes
-                if isinstance(uuid, bytes):
-                    uuid = uuid.decode('utf-8', errors='ignore')
+                if isinstance(raw_uuid, bytes):
+                    uuid = raw_uuid.decode("utf-8", errors="ignore")
+                elif raw_uuid is not None:
+                    uuid = str(raw_uuid)
                 pci_info = pynvml.nvmlDeviceGetPciInfo(handle)
-                # Parse something like "00000000:01:00.0" -> bus is 1
-                if isinstance(pci_info.bus, bytes):
-                    pci_bus_id = int(pci_info.bus.decode('utf-8', errors='ignore'), 16)
+                # Parse something like "00000000:01:00.0" -> bus is 1.
+                # PCI bus numbers are hex by convention (pynvml's field is an
+                # int; the str/bytes branches are defensive).
+                bus = pci_info.bus
+                if isinstance(bus, bytes):
+                    pci_bus_id = int(bus.decode("utf-8", errors="ignore"), 16)
+                elif isinstance(bus, str):
+                    pci_bus_id = int(bus, 16)
                 else:
-                    pci_bus_id = pci_info.bus
-            except Exception:
-                pass
+                    pci_bus_id = int(bus)
+            except Exception as exc:
+                log.debug("NVML query for GPU %d failed: %s", i, exc)
         infos.append(GpuInfo(name=name, index=i, uuid=uuid, pci_bus_id=pci_bus_id))
 
-    if has_nvml:
-        try:
+    if pynvml is not None:
+        with contextlib.suppress(Exception):
             pynvml.nvmlShutdown()
-        except Exception:
-            pass
 
     return infos
 

nvcurve/hal/limits.py

@@ -11,21 +11,26 @@ that are explicitly specified, leaving others unchanged on hardware.
 """
 
 import ctypes
-import subprocess
 import logging
-from typing import Optional
+import subprocess
+from typing import Any
 
 try:
-    import pynvml
+    import pynvml as _pynvml_import
+
     _NVML_AVAILABLE = True
 except ImportError:
+    _pynvml_import = None
     _NVML_AVAILABLE = False
 
+# Aliased as Any so attribute access is not flagged when the import failed.
+pynvml: Any = _pynvml_import
+
 log = logging.getLogger("nvcurve.hal.limits")
 
 # ── NVML library / handle helpers ─────────────────────────────────────────────
 
-_nvml_lib: Optional[ctypes.CDLL] = None
+_nvml_lib: ctypes.CDLL | None = None
 
 
 def _nvml_cdll() -> ctypes.CDLL:
@@ -34,12 +39,12 @@ def _nvml_cdll() -> ctypes.CDLL:
     if _nvml_lib is not None:
         return _nvml_lib
     # Prefer to reuse the library already loaded by pynvml to avoid dlopen races.
-    for attr in ("nvml", "_nvml"):          # attribute name varies by pynvml version
+    for attr in ("nvml", "_nvml"):  # attribute name varies by pynvml version
         mod = getattr(pynvml, attr, None)
         lib = getattr(mod, "_lib", None) or getattr(mod, "_nvmlLib", None)
         if lib is not None:
             _nvml_lib = lib
-            return _nvml_lib
+            return lib
     _nvml_lib = ctypes.CDLL("libnvidia-ml.so.1")
     return _nvml_lib
 
@@ -53,9 +58,10 @@ def _get_handle(gpu_index: int):
 
 # ── Power limit ───────────────────────────────────────────────────────────────
 
+
 def get_power_limit(gpu_index: int = 0) -> dict:
     """Return dict with power_limit_w, default_power_limit_w, min_power_limit_w, max_power_limit_w."""
-    out = {
+    out: dict[str, int | None] = {
         "power_limit_w": None,
         "default_power_limit_w": None,
         "min_power_limit_w": None,
@@ -71,8 +77,8 @@ def get_power_limit(gpu_index: int = 0) -> dict:
         try:
             default = pynvml.nvmlDeviceGetPowerManagementDefaultLimit(handle)
             out["default_power_limit_w"] = default // 1000
-        except Exception:
-            pass
+        except Exception as exc:
+            log.debug("nvmlDeviceGetPowerManagementDefaultLimit: %s", exc)
     except Exception as exc:
         log.warning("get_power_limit: %s", exc)
     return out
@@ -89,7 +95,8 @@ def set_power_limit(limit_w: int, gpu_index: int = 0) -> tuple[bool, str]:
 
     ret = subprocess.run(
         ["nvidia-smi", "-i", str(gpu_index), "-pl", str(limit_w)],
-        capture_output=True, text=True,
+        capture_output=True,
+        text=True,
     )
     if ret.returncode == 0:
         return True, "OK"
@@ -111,34 +118,38 @@ def set_power_limit(limit_w: int, gpu_index: int = 0) -> tuple[bool, str]:
 # pynvml (nvidia-ml-py ≥ 12) exposes c_nvmlClockOffset_t and nvmlClockOffset_v1
 # as ctypes objects; we use them when available and fall back to our own definition.
 
+
 class _ClockOffset(ctypes.Structure):
     _fields_ = [
-        ("version",        ctypes.c_uint),
-        ("type",           ctypes.c_uint),   # nvmlClockType_t
-        ("pstate",         ctypes.c_uint),   # nvmlPstates_t
+        ("version", ctypes.c_uint),
+        ("type", ctypes.c_uint),  # nvmlClockType_t
+        ("pstate", ctypes.c_uint),  # nvmlPstates_t
         ("clockOffsetMHz", ctypes.c_int),
     ]
 
+
 _CLOCK_OFFSET_VER = (1 << 24) | ctypes.sizeof(_ClockOffset)  # = 0x01000010 (16 bytes)
 
 # NVML clock-type constants (same values as pynvml).
 _NVML_CLOCK_GRAPHICS = 0
-_NVML_CLOCK_MEM      = 2
+_NVML_CLOCK_MEM = 2
 
 
-def _make_clock_offset(clock_type: int, pstate: int = 0, offset_mhz: int = 0) -> ctypes.Structure:
+def _make_clock_offset(
+    clock_type: int, pstate: int = 0, offset_mhz: int = 0
+) -> ctypes.Structure:
     """Return a populated nvmlClockOffset_t struct, using pynvml's type when available."""
     if hasattr(pynvml, "c_nvmlClockOffset_t") and hasattr(pynvml, "nvmlClockOffset_v1"):
         info = pynvml.c_nvmlClockOffset_t()
-        info.version        = pynvml.nvmlClockOffset_v1
-        info.type           = clock_type
-        info.pstate         = pstate
+        info.version = pynvml.nvmlClockOffset_v1
+        info.type = clock_type
+        info.pstate = pstate
         info.clockOffsetMHz = offset_mhz
         return info
     info = _ClockOffset()
-    info.version        = _CLOCK_OFFSET_VER
-    info.type           = clock_type
-    info.pstate         = pstate
+    info.version = _CLOCK_OFFSET_VER
+    info.type = clock_type
+    info.pstate = pstate
     info.clockOffsetMHz = offset_mhz
     return info
 
@@ -158,7 +169,7 @@ def get_clock_offsets(gpu_index: int = 0) -> dict:
     Keys: gpc_offset_mhz, mem_offset_mhz (both int or None on failure).
     Calls nvmlDeviceGetClockOffsets once per clock domain (GRAPHICS, MEM).
     """
-    out = {"gpc_offset_mhz": None, "mem_offset_mhz": None}
+    out: dict[str, int | None] = {"gpc_offset_mhz": None, "mem_offset_mhz": None}
     if not _NVML_AVAILABLE:
         return out
     try:
@@ -167,11 +178,15 @@ def get_clock_offsets(gpu_index: int = 0) -> dict:
         # Try pynvml wrapper first (nvidia-ml-py ≥ 12 exposes it correctly).
         # Fall back to ctypes-direct if pynvml doesn't have it.
         _pynvml_get = getattr(pynvml, "nvmlDeviceGetClockOffsets", None)
-        fn_get = _try_nvml_fn("nvmlDeviceGetClockOffsets") if _pynvml_get is None else None
+        fn_get = (
+            _try_nvml_fn("nvmlDeviceGetClockOffsets") if _pynvml_get is None else None
+        )
 
         used_new_api = False
-        for clock_type, key in ((_NVML_CLOCK_GRAPHICS, "gpc_offset_mhz"),
-                                 (_NVML_CLOCK_MEM,      "mem_offset_mhz")):
+        for clock_type, key in (
+            (_NVML_CLOCK_GRAPHICS, "gpc_offset_mhz"),
+            (_NVML_CLOCK_MEM, "mem_offset_mhz"),
+        ):
             info = _make_clock_offset(clock_type, pstate=0)
             try:
                 if _pynvml_get is not None:
@@ -184,7 +199,9 @@ def get_clock_offsets(gpu_index: int = 0) -> dict:
                     out[key] = int(info.clockOffsetMHz)
                     used_new_api = True
                 else:
-                    log.debug("nvmlDeviceGetClockOffsets(type=%d) returned %d", clock_type, rc)
+                    log.debug(
+                        "nvmlDeviceGetClockOffsets(type=%d) returned %d", clock_type, rc
+                    )
             except Exception as exc:
                 log.debug("nvmlDeviceGetClockOffsets(type=%d): %s", clock_type, exc)
 
@@ -200,7 +217,9 @@ def get_clock_offsets(gpu_index: int = 0) -> dict:
         if hasattr(pynvml, "nvmlDeviceGetMemClkVfOffset"):
             try:
                 res = pynvml.nvmlDeviceGetMemClkVfOffset(handle)
-                out["mem_offset_mhz"] = int(res[0] if isinstance(res, (list, tuple)) else res)
+                out["mem_offset_mhz"] = int(
+                    res[0] if isinstance(res, (list, tuple)) else res
+                )
             except Exception as exc:
                 log.debug("nvmlDeviceGetMemClkVfOffset: %s", exc)
 
@@ -210,8 +229,8 @@ def get_clock_offsets(gpu_index: int = 0) -> dict:
 
 
 def set_clock_offsets(
-    gpc_offset_mhz: Optional[int] = None,
-    mem_offset_mhz: Optional[int] = None,
+    gpc_offset_mhz: int | None = None,
+    mem_offset_mhz: int | None = None,
     gpu_index: int = 0,
 ) -> tuple[bool, str]:
     """Set clock offsets (MHz) for the specified domains only.
@@ -235,20 +254,35 @@ def set_clock_offsets(
             domains.append((_NVML_CLOCK_MEM, mem_offset_mhz))
 
         _pynvml_set = getattr(pynvml, "nvmlDeviceSetClockOffsets", None)
-        fn_set = _try_nvml_fn("nvmlDeviceSetClockOffsets") if _pynvml_set is None else None
+        fn_set = (
+            _try_nvml_fn("nvmlDeviceSetClockOffsets") if _pynvml_set is None else None
+        )
 
         if _pynvml_set is not None or fn_set is not None:
             all_ok = True
             for clock_type, offset in domains:
                 info = _make_clock_offset(clock_type, pstate=0, offset_mhz=offset)
                 try:
-                    rc = _pynvml_set(handle, ctypes.byref(info)) if _pynvml_set else fn_set(handle, ctypes.byref(info))
+                    if _pynvml_set is not None:
+                        rc = _pynvml_set(handle, ctypes.byref(info))
+                    elif fn_set is not None:
+                        rc = fn_set(handle, ctypes.byref(info))
+                    else:
+                        break
                     if rc != 0:
-                        log.debug("nvmlDeviceSetClockOffsets(type=%d) returned %d — trying fallback", clock_type, rc)
+                        log.debug(
+                            "nvmlDeviceSetClockOffsets(type=%d) returned %d — trying fallback",
+                            clock_type,
+                            rc,
+                        )
                         all_ok = False
                         break
                 except Exception as exc:
-                    log.debug("nvmlDeviceSetClockOffsets(type=%d): %s — trying fallback", clock_type, exc)
+                    log.debug(
+                        "nvmlDeviceSetClockOffsets(type=%d): %s — trying fallback",
+                        clock_type,
+                        exc,
+                    )
                     all_ok = False
                     break
             if all_ok:
@@ -257,12 +291,16 @@ def set_clock_offsets(
 
         # Deprecated per-domain fallback (works on Blackwell/driver 590.x).
         errs = []
-        if gpc_offset_mhz is not None and hasattr(pynvml, "nvmlDeviceSetGpcClkVfOffset"):
+        if gpc_offset_mhz is not None and hasattr(
+            pynvml, "nvmlDeviceSetGpcClkVfOffset"
+        ):
             try:
                 pynvml.nvmlDeviceSetGpcClkVfOffset(handle, gpc_offset_mhz)
             except Exception as exc:
                 errs.append(f"GPC: {exc}")
-        if mem_offset_mhz is not None and hasattr(pynvml, "nvmlDeviceSetMemClkVfOffset"):
+        if mem_offset_mhz is not None and hasattr(
+            pynvml, "nvmlDeviceSetMemClkVfOffset"
+        ):
             try:
                 pynvml.nvmlDeviceSetMemClkVfOffset(handle, mem_offset_mhz)
             except Exception as exc:
@@ -278,6 +316,7 @@ def set_clock_offsets(
 
 # ── Range queries ─────────────────────────────────────────────────────────────
 
+
 def get_mem_offset_range(gpu_index: int = 0) -> dict:
     """Return the min/max allowed memory clock offset (MHz).
 
@@ -285,7 +324,7 @@ def get_mem_offset_range(gpu_index: int = 0) -> dict:
     Uses nvmlDeviceGetMemClkMinMaxVfOffset; falls back to observed RTX values.
     """
     # Observed RTX 5090 defaults (NvAPI GetClockBoostRanges says -1000/+3000).
-    out = {"min_mem_offset_mhz": -2000, "max_mem_offset_mhz": 3000}
+    out: dict[str, int] = {"min_mem_offset_mhz": -2000, "max_mem_offset_mhz": 3000}
     if not _NVML_AVAILABLE:
         return out
     try:
@@ -317,5 +356,3 @@ def get_mem_offset_range(gpu_index: int = 0) -> dict:
     except Exception as exc:
         log.debug("get_mem_offset_range: %s", exc)
     return out
-
-

nvcurve/hal/snapshot.py

@@ -2,18 +2,20 @@
 
 import ctypes
 import json
+import logging
 import os
 import struct
 from datetime import datetime
-from typing import Optional
 
 from ..nvapi.bootstrap import nvcall_raw
-from ..nvapi.constants import FUNC, CT_SIZE, CT_BASE, CT_STRIDE, CT_DELTA_OFF, CT_POINTS
+from ..nvapi.constants import CT_BASE, CT_DELTA_OFF, CT_SIZE, CT_STRIDE, FUNC
 from ..nvapi.types import SnapshotInfo
-from .vfcurve import read_clock_table_raw, get_boost_mask
+from .vfcurve import get_boost_mask, read_clock_table_raw
+
+log = logging.getLogger("nvcurve.hal.snapshot")
 
 
-def save(gpu, gpu_name: str, snapshot_dir: str, max_snapshots: int = 0) -> Optional[str]:
+def save(gpu, gpu_name: str, snapshot_dir: str, max_snapshots: int = 0) -> str | None:
     """Save the current ClockBoostTable to disk.
 
     Writes both a binary .bin file and a human-readable .json metadata file.
@@ -25,13 +27,21 @@ def save(gpu, gpu_name: str, snapshot_dir: str, max_snapshots: int = 0) -> Optio
         print(f"Failed to read ClockBoostTable: {err}")
         return None
 
-    os.makedirs(snapshot_dir, exist_ok=True)
+    try:
+        os.makedirs(snapshot_dir, exist_ok=True)
+    except OSError as exc:
+        print(f"Failed to create snapshot dir {snapshot_dir}: {exc}")
+        return None
     ts = datetime.now().strftime("%Y%m%d_%H%M%S")
     bin_path = os.path.join(snapshot_dir, f"clock_boost_table_{ts}.bin")
     meta_path = os.path.join(snapshot_dir, f"clock_boost_table_{ts}.json")
 
-    with open(bin_path, "wb") as f:
-        f.write(raw)
+    try:
+        with open(bin_path, "wb") as f:
+            f.write(raw)
+    except OSError as exc:
+        print(f"Failed to write snapshot {bin_path}: {exc}")
+        return None
 
     offsets = []
     max_entries = (len(raw) - CT_BASE) // CT_STRIDE
@@ -48,10 +58,14 @@ def save(gpu, gpu_name: str, snapshot_dir: str, max_snapshots: int = 0) -> Optio
         "offsets_kHz": offsets,
         "nonzero_offsets": sum(1 for o in offsets if o != 0),
     }
-    with open(meta_path, "w") as f:
-        json.dump(meta, f, indent=2)
+    try:
+        with open(meta_path, "w") as f:
+            json.dump(meta, f, indent=2)
+    except OSError as exc:
+        print(f"Failed to write snapshot metadata {meta_path}: {exc}")
+        return None
 
-    print(f"Snapshot saved:")
+    print("Snapshot saved:")
     print(f"  Binary:   {bin_path}")
     print(f"  Metadata: {meta_path}")
     print(f"  Size:     {len(raw)} bytes")
@@ -65,20 +79,22 @@ def save(gpu, gpu_name: str, snapshot_dir: str, max_snapshots: int = 0) -> Optio
 
 def _prune_snapshots(snapshot_dir: str, max_snapshots: int) -> None:
     """Delete oldest snapshots (both .bin and .json) to stay within max_snapshots."""
-    bins = sorted(
-        f for f in os.listdir(snapshot_dir) if f.endswith(".bin")
-    )  # oldest first (lexicographic = chronological for our timestamp format)
+    # Oldest first (lexicographic = chronological for our timestamp format).
+    try:
+        bins = sorted(f for f in os.listdir(snapshot_dir) if f.endswith(".bin"))
+    except OSError:
+        return
     excess = len(bins) - max_snapshots
     for fname in bins[:excess]:
         stem = fname[:-4]  # strip .bin
         for ext in (".bin", ".json"):
             try:
                 os.remove(os.path.join(snapshot_dir, stem + ext))
-            except OSError:
-                pass
+            except OSError as exc:
+                log.debug("Could not remove %s: %s", stem + ext, exc)
 
 
-def restore(gpu, snapshot_dir: str, filepath: str = None) -> bool:
+def restore(gpu, snapshot_dir: str, filepath: str | None = None) -> bool:
     """Restore a ClockBoostTable snapshot from disk.
 
     If no filepath is given, uses the most recent snapshot in snapshot_dir.
@@ -88,10 +104,14 @@ def restore(gpu, snapshot_dir: str, filepath: str = None) -> bool:
         if not os.path.isdir(snapshot_dir):
             print(f"No snapshots found in {snapshot_dir}")
             return False
-        bins = sorted(
-            [f for f in os.listdir(snapshot_dir) if f.endswith(".bin")],
-            reverse=True,
-        )
+        try:
+            bins = sorted(
+                [f for f in os.listdir(snapshot_dir) if f.endswith(".bin")],
+                reverse=True,
+            )
+        except OSError:
+            print(f"No snapshots found in {snapshot_dir}")
+            return False
         if not bins:
             print(f"No snapshot .bin files in {snapshot_dir}")
             return False
@@ -101,8 +121,12 @@ def restore(gpu, snapshot_dir: str, filepath: str = None) -> bool:
         print(f"Snapshot file not found: {filepath}")
         return False
 
-    with open(filepath, "rb") as f:
-        raw = f.read()
+    try:
+        with open(filepath, "rb") as f:
+            raw = f.read()
+    except OSError as exc:
+        print(f"Failed to read snapshot {filepath}: {exc}")
+        return False
 
     if len(raw) != CT_SIZE:
         print(f"Snapshot size mismatch: expected {CT_SIZE}, got {len(raw)}")
@@ -134,8 +158,13 @@ def list_snapshots(snapshot_dir: str) -> list[SnapshotInfo]:
     if not os.path.isdir(snapshot_dir):
         return []
 
+    try:
+        fnames = sorted(os.listdir(snapshot_dir), reverse=True)
+    except OSError:
+        return []
+
     results = []
-    for fname in sorted(os.listdir(snapshot_dir), reverse=True):
+    for fname in fnames:
         if not fname.endswith(".json"):
             continue
         meta_path = os.path.join(snapshot_dir, fname)
@@ -143,13 +172,15 @@ def list_snapshots(snapshot_dir: str) -> list[SnapshotInfo]:
             with open(meta_path) as f:
                 meta = json.load(f)
             bin_path = meta.get("file", meta_path.replace(".json", ".bin"))
-            results.append(SnapshotInfo(
-                filepath=bin_path,
-                timestamp=meta.get("timestamp", ""),
-                gpu=meta.get("gpu", ""),
-                nonzero_offsets=meta.get("nonzero_offsets", 0),
-                size=meta.get("size", 0),
-            ))
+            results.append(
+                SnapshotInfo(
+                    filepath=bin_path,
+                    timestamp=meta.get("timestamp", ""),
+                    gpu=meta.get("gpu", ""),
+                    nonzero_offsets=meta.get("nonzero_offsets", 0),
+                    size=meta.get("size", 0),
+                )
+            )
         except (json.JSONDecodeError, KeyError):
             continue
 

nvcurve/profiles/apply.py

@@ -21,12 +21,12 @@ def _gpu_stable_key(info) -> str:
 
 def apply_profile(gpu_index: int, name: str, cfg) -> list[str]:
     """Apply a named profile to the given GPU. Returns a list of error strings."""
-    from .native import load_profile
     from ..hal.gpu import get_gpu
     from ..hal.limits import set_clock_offsets, set_power_limit
-    from ..hal.vfcurve import write_offsets, reset_offsets
     from ..hal.snapshot import save as snapshot_save
+    from ..hal.vfcurve import reset_offsets, write_offsets
     from ..safety import validate_write
+    from .native import load_profile
 
     safe_name = "".join(c for c in name if c.isalnum() or c in " _-()").strip()
     filepath = os.path.join(cfg.profile_dir, f"{safe_name}.json")
@@ -48,19 +48,25 @@ def apply_profile(gpu_index: int, name: str, cfg) -> list[str]:
             errs.append(f"Power limit: {msg}")
 
     if profile.curve_deltas:
-        deltas = {int(k): v for k, v in profile.curve_deltas.items()}
-        errors = validate_write(deltas, cfg.max_delta_khz)
-        if errors:
-            errs.append("Curve: " + "; ".join(errors))
+        try:
+            deltas = {int(k): v for k, v in profile.curve_deltas.items()}
+        except ValueError:
+            errs.append("Curve: invalid point keys in profile")
         else:
-            if cfg.auto_snapshot:
-                try:
-                    snapshot_save(gpu, gpu_name, cfg.snapshot_dir, cfg.max_snapshots)
-                except Exception as exc:
-                    log.warning("Auto-snapshot failed: %s", exc)
-            ret, desc = write_offsets(gpu, deltas)
-            if ret != 0:
-                errs.append(f"Curve write failed ({ret}): {desc}")
+            errors = validate_write(deltas, cfg.max_delta_khz)
+            if errors:
+                errs.append("Curve: " + "; ".join(errors))
+            else:
+                if cfg.auto_snapshot:
+                    try:
+                        snapshot_save(
+                            gpu, gpu_name, cfg.snapshot_dir, cfg.max_snapshots
+                        )
+                    except Exception as exc:
+                        log.warning("Auto-snapshot failed: %s", exc)
+                ret, desc = write_offsets(gpu, deltas)
+                if ret != 0:
+                    errs.append(f"Curve write failed ({ret}): {desc}")
     else:
         reset_offsets(gpu)
 
@@ -69,9 +75,9 @@ def apply_profile(gpu_index: int, name: str, cfg) -> list[str]:
 
 def apply_with_retry(gpu_index: int, name: str, cfg, max_retries: int = 3) -> bool:
     """Apply a named profile with read-back verification, retrying on mismatch."""
-    from .native import load_profile
     from ..hal.gpu import get_gpu
     from ..hal.vfcurve import read_clock_offsets
+    from .native import load_profile
 
     safe_name = "".join(c for c in name if c.isalnum() or c in " _-()").strip()
     filepath = os.path.join(cfg.profile_dir, f"{safe_name}.json")
@@ -82,51 +88,88 @@ def apply_with_retry(gpu_index: int, name: str, cfg, max_retries: int = 3) -> bo
         log.warning("Auto-load profile %r not found — skipping GPU %d", name, gpu_index)
         return False
 
-    expected: dict[int, int] = (
-        {int(k): v for k, v in profile.curve_deltas.items()}
-        if profile.curve_deltas else {}
-    )
+    try:
+        expected: dict[int, int] = (
+            {int(k): v for k, v in profile.curve_deltas.items()}
+            if profile.curve_deltas
+            else {}
+        )
+    except ValueError:
+        log.warning(
+            "Profile %r has invalid curve point keys — skipping GPU %d",
+            name,
+            gpu_index,
+        )
+        return False
 
     for attempt in range(max_retries):
         try:
             errs = apply_profile(gpu_index, name, cfg)
         except Exception as exc:
-            log.warning("Auto-load attempt %d/%d exception: %s", attempt + 1, max_retries, exc)
+            log.warning(
+                "Auto-load attempt %d/%d exception: %s", attempt + 1, max_retries, exc
+            )
             errs = [str(exc)]
 
         if errs:
-            log.warning("Auto-load attempt %d/%d errors: %s",
-                        attempt + 1, max_retries, "; ".join(errs))
+            log.warning(
+                "Auto-load attempt %d/%d errors: %s",
+                attempt + 1,
+                max_retries,
+                "; ".join(errs),
+            )
         elif expected:
             gpu, _ = get_gpu(index=gpu_index)
             offsets, err = read_clock_offsets(gpu)
             if offsets is None:
-                log.warning("Auto-load attempt %d/%d: read-back failed: %s",
-                            attempt + 1, max_retries, err)
+                log.warning(
+                    "Auto-load attempt %d/%d: read-back failed: %s",
+                    attempt + 1,
+                    max_retries,
+                    err,
+                )
             else:
                 mismatches = [
-                    f"pt{idx}: expected {val/1000:+.0f}MHz got {offsets[idx]/1000:+.0f}MHz"
+                    f"pt{idx}: expected {val / 1000:+.0f}MHz got {offsets[idx] / 1000:+.0f}MHz"
                     for idx, val in expected.items()
                     if idx < len(offsets) and offsets[idx] != val
                 ]
                 if not mismatches:
-                    log.info("Auto-load profile %r verified on GPU %d (attempt %d/%d)",
-                             name, gpu_index, attempt + 1, max_retries)
+                    log.info(
+                        "Auto-load profile %r verified on GPU %d (attempt %d/%d)",
+                        name,
+                        gpu_index,
+                        attempt + 1,
+                        max_retries,
+                    )
                     return True
-                log.warning("Auto-load attempt %d/%d: read-back mismatch — %s",
-                            attempt + 1, max_retries, "; ".join(mismatches))
+                log.warning(
+                    "Auto-load attempt %d/%d: read-back mismatch — %s",
+                    attempt + 1,
+                    max_retries,
+                    "; ".join(mismatches),
+                )
         else:
-            log.info("Auto-load profile %r applied on GPU %d (attempt %d/%d)",
-                     name, gpu_index, attempt + 1, max_retries)
+            log.info(
+                "Auto-load profile %r applied on GPU %d (attempt %d/%d)",
+                name,
+                gpu_index,
+                attempt + 1,
+                max_retries,
+            )
             return True
 
         if attempt < max_retries - 1:
-            delay = 2 ** attempt  # 1s, 2s, 4s
+            delay = 2**attempt  # 1s, 2s, 4s
             log.info("Retrying auto-load in %ds…", delay)
             time.sleep(delay)
 
-    log.warning("Auto-load profile %r failed after %d attempts on GPU %d",
-                name, max_retries, gpu_index)
+    log.warning(
+        "Auto-load profile %r failed after %d attempts on GPU %d",
+        name,
+        max_retries,
+        gpu_index,
+    )
     return False
 
 
@@ -153,13 +196,19 @@ def run_autoload() -> None:
         return
 
     from ..config import Config
+
     cfg = Config()
-    for key in ("max_delta_khz", "auto_snapshot", "max_snapshots",
-                "snapshot_dir", "profile_dir"):
+    for key in (
+        "max_delta_khz",
+        "auto_snapshot",
+        "max_snapshots",
+        "snapshot_dir",
+        "profile_dir",
+    ):
         if key in cfg_data:
             setattr(cfg, key, cfg_data[key])
 
-    from ..hal.gpu import init_nvapi, discover_gpus
+    from ..hal.gpu import discover_gpus, init_nvapi
     from ..hal.monitoring import init_nvml, shutdown_nvml
 
     # Retry NvAPI init — the driver may not be fully ready at early boot.
@@ -189,7 +238,9 @@ def run_autoload() -> None:
         if gpu_idx is None:
             log.warning("Auto-load: no GPU found with key %r — skipping", gpu_key)
             continue
-        log.info("Auto-loading profile %r on GPU %d (%s)", profile_name, gpu_idx, gpu_key)
+        log.info(
+            "Auto-loading profile %r on GPU %d (%s)", profile_name, gpu_idx, gpu_key
+        )
         apply_with_retry(gpu_idx, profile_name, cfg)
 
     shutdown_nvml()

nvcurve/profiles/native.py

@@ -1,39 +1,52 @@
 """Native profile storage and schema."""
 
-import json
-import os
 import glob
-from dataclasses import dataclass, asdict
-from typing import Dict, Optional, List
+import json
+import logging
+import os
+from dataclasses import asdict, dataclass
+
+log = logging.getLogger("nvcurve.profiles.native")
 
 
 @dataclass
 class ProfileData:
     name: str
     gpu_name: str
-    curve_deltas: Dict[str, int]  # { "index": delta_khz }
-    mem_offset_mhz: Optional[int] = None
-    power_limit_w: Optional[int] = None
-    fan_curve: Optional[List[Dict[str, int]]] = None
+    curve_deltas: dict[str, int]  # { "index": delta_khz }
+    mem_offset_mhz: int | None = None
+    power_limit_w: int | None = None
+    fan_curve: list[dict[str, int]] | None = None
 
 
 def save_profile(profile_dir: str, data: ProfileData) -> str:
     """Save profile to JSON, sanitising the filename."""
-    os.makedirs(profile_dir, exist_ok=True)
+    try:
+        os.makedirs(profile_dir, exist_ok=True)
+    except OSError as exc:
+        raise RuntimeError(f"Cannot create profile dir {profile_dir}: {exc}") from exc
     safe_name = "".join(c for c in data.name if c.isalnum() or c in " _-()").strip()
     if not safe_name:
         safe_name = "Unnamed"
-        
+
     filepath = os.path.join(profile_dir, f"{safe_name}.json")
-    with open(filepath, "w", encoding="utf-8") as f:
-        json.dump(asdict(data), f, indent=2)
+    try:
+        with open(filepath, "w", encoding="utf-8") as f:
+            json.dump(asdict(data), f, indent=2)
+    except OSError as exc:
+        raise RuntimeError(f"Cannot write profile {filepath}: {exc}") from exc
     return filepath
 
 
 def load_profile(filepath: str) -> ProfileData:
     """Load profile from JSON."""
-    with open(filepath, "r", encoding="utf-8") as f:
-        data = json.load(f)
+    try:
+        with open(filepath, encoding="utf-8") as f:
+            data = json.load(f)
+    except FileNotFoundError:
+        raise
+    except (OSError, json.JSONDecodeError) as exc:
+        raise RuntimeError(f"Cannot read profile {filepath}: {exc}") from exc
     # Migrate old field names.
     if "vram_p0_offset_mhz" in data and "mem_offset_mhz" not in data:
         data["mem_offset_mhz"] = data.pop("vram_p0_offset_mhz")
@@ -43,7 +56,7 @@ def load_profile(filepath: str) -> ProfileData:
     return ProfileData(**data)
 
 
-def list_profiles(profile_dir: str) -> List[ProfileData]:
+def list_profiles(profile_dir: str) -> list[ProfileData]:
     """Return a list of all safely readable profiles."""
     if not os.path.exists(profile_dir):
         return []
@@ -51,9 +64,8 @@ def list_profiles(profile_dir: str) -> List[ProfileData]:
     for fp in glob.glob(os.path.join(profile_dir, "*.json")):
         try:
             profiles.append(load_profile(fp))
-        except Exception as e:
-            # log warning ideally, but swallowing for robustness
-            pass
+        except Exception as exc:
+            log.debug("Skipping unreadable profile %s: %s", fp, exc)
     # Sort alphabetically by name
     profiles.sort(key=lambda p: p.name.lower())
     return profiles

nvcurve/server.py

@@ -90,8 +90,8 @@ def _open_browser_as_user(url: str) -> None:
                 stderr=subprocess.DEVNULL,
             )
             return
-        except Exception:
-            pass
+        except Exception as exc:
+            log.debug("runuser xdg-open failed, falling back to webbrowser: %s", exc)
     import webbrowser
 
     webbrowser.open(url)
@@ -1692,7 +1692,10 @@ def _resolve_dist_dir() -> str:
     the project-root layout used during local development.
     """
     try:
-        from importlib.resources import files as _resource_files
+        # Project requires Python >= 3.12, so the 3.7-compat finding is a false positive.
+        from importlib.resources import (  # nosemgrep: python.lang.compatibility.python37.python37-compatibility-importlib2
+            files as _resource_files,
+        )
 
         candidate = _resource_files("nvcurve") / "frontend" / "dist"
         if candidate.is_dir():