Let a decode failure degrade instead of escaping a fail-closed helper (#7487)

* Let a decode failure degrade instead of escaping a fail-closed helper

Pinning utf-8 makes a read that used to return mojibake on Windows raise
instead. 33 of those reads sit under a handler catching OSError or
json.JSONDecodeError but not UnicodeDecodeError, which subclasses
ValueError, so a corrupt file would now escape a helper written to return
a default. Adds UnicodeDecodeError to those tuples only.

* Treat an undecodable install lock as stale instead of retrying forever

studio/backend/colab.py

@@ -109,7 +109,7 @@ def _load_colab_login_credentials() -> "tuple[str, str] | None":
         lines = path.read_text(encoding = "utf-8").splitlines()
         if len(lines) >= 2 and lines[0] and lines[1]:
             return lines[0], lines[1]
-    except OSError as e:
+    except (OSError, UnicodeDecodeError) as e:
         logger.info(f"Could not load Colab login credentials ({e}).")
     return None
 

studio/backend/core/inference/llama_cpp.py

@@ -246,7 +246,7 @@ def _wsl_system_rocm_lib_dirs() -> "list[str]":
         with open("/proc/version", encoding = "utf-8", errors = "replace") as fh:
             if "microsoft" not in fh.read().lower():
                 return []
-    except OSError:
+    except (OSError, UnicodeDecodeError):
         return []
     out: "list[str]" = []
     for d in ("/opt/rocm/lib", "/opt/rocm/lib64"):
@@ -573,7 +573,7 @@ def _load_swa_cache() -> dict:
                 _SWA_CACHE = json.load(f)
                 if not isinstance(_SWA_CACHE, dict):
                     _SWA_CACHE = {}
-        except (FileNotFoundError, json.JSONDecodeError, OSError):
+        except (FileNotFoundError, json.JSONDecodeError, OSError, UnicodeDecodeError):
             _SWA_CACHE = {}
         return _SWA_CACHE
 
@@ -586,7 +586,7 @@ def _save_swa_cache(cache: dict) -> None:
         with open(tmp, "w", encoding = "utf-8") as f:
             json.dump(cache, f, indent = 2, sort_keys = True)
         tmp.replace(path)
-    except OSError:
+    except (OSError, UnicodeDecodeError):
         pass
 
 
@@ -5138,7 +5138,7 @@ class LlamaCppBackend:
             self._llama_log_path = log_dir / f"diffusion-{int(time.time())}-port-{self._port}.log"
             self._llama_log_fh = open(self._llama_log_path, "w", encoding = "utf-8", buffering = 1)
             logger.info(f"diffusion runner stdout/stderr -> {self._llama_log_path}")
-        except OSError as e:
+        except (OSError, UnicodeDecodeError) as e:
             logger.debug(f"Could not open diffusion runner log file: {e}")
 
         # The shim (and its visual server) die with this backend process, so a
@@ -6355,7 +6355,7 @@ class LlamaCppBackend:
                 buffering = 1,
             )
             logger.info(f"llama-server stdout/stderr -> {self._llama_log_path}")
-        except OSError as e:
+        except (OSError, UnicodeDecodeError) as e:
             # Best-effort; never block the load on logging.
             logger.debug(f"Could not open llama-server log file: {e}")
             self._llama_log_path = None
@@ -8302,7 +8302,7 @@ class LlamaCppBackend:
                                 buffering = 1,
                             )
                             logger.info(f"llama-server stdout/stderr -> {self._llama_log_path}")
-                        except OSError as e:
+                        except (OSError, UnicodeDecodeError) as e:
                             # Best-effort; never block the load on logging.
                             logger.debug(f"Could not open llama-server log file: {e}")
                             self._llama_log_path = None

studio/backend/hub/services/models/ollama.py

@@ -216,7 +216,7 @@ def _ollama_model_info_from_manifest(
 
     try:
         manifest = json.loads(tag_file.read_text(encoding = "utf-8"))
-    except (json.JSONDecodeError, OSError) as e:
+    except (json.JSONDecodeError, OSError, UnicodeDecodeError) as e:
         logger.debug("Skipping unreadable/invalid Ollama manifest %s: %s", tag_file, e)
         return None
 
@@ -231,7 +231,7 @@ def _ollama_model_info_from_manifest(
                 cfg = json.loads(config_blob.read_text(encoding = "utf-8"))
                 model_type = cfg.get("model_type", "")
                 file_type = cfg.get("file_type", "")
-            except (json.JSONDecodeError, OSError) as e:
+            except (json.JSONDecodeError, OSError, UnicodeDecodeError) as e:
                 logger.debug("Could not parse Ollama config blob %s: %s", config_blob, e)
 
     layers = manifest.get("layers") or []

studio/backend/hub/utils/download_registry.py

@@ -463,7 +463,7 @@ def _read_marker_value(marker: Path) -> Optional[str]:
         if not marker.exists():
             return None
         value = marker.read_text(encoding = "utf-8").strip()
-    except OSError:
+    except (OSError, UnicodeDecodeError):
         return None
     return value if value in VALID_TRANSPORTS else None
 

studio/backend/main.py

@@ -362,7 +362,7 @@ def get_unsloth_version() -> str:
         for line in version_file.read_text(encoding = "utf-8").splitlines():
             if line.startswith("__version__ = "):
                 return line.split("=", 1)[1].strip().strip('"').strip("'")
-    except OSError:
+    except (OSError, UnicodeDecodeError):
         pass
     return "dev"
 

studio/backend/plugins/data-designer-unstructured-seed/src/data_designer_unstructured_seed/impl.py

@@ -29,7 +29,7 @@ class UnstructuredSeedReader(SeedReader[UnstructuredSeedSource]):
                 try:
                     meta = json_mod.loads(meta_path.read_text(encoding = "utf-8"))
                     orig_name = meta.get("original_filename", path_obj.name)
-                except (json_mod.JSONDecodeError, OSError):
+                except (json_mod.JSONDecodeError, OSError, UnicodeDecodeError):
                     pass
             file_entries.append((path_obj, orig_name))
 

studio/backend/routes/models.py

@@ -723,7 +723,7 @@ def _scan_ollama_dir(ollama_dir: Path, limit: Optional[int] = None) -> List[Loca
 
             try:
                 manifest = json.loads(tag_file.read_text(encoding = "utf-8"))
-            except (json.JSONDecodeError, OSError) as e:
+            except (json.JSONDecodeError, OSError, UnicodeDecodeError) as e:
                 logger.debug(
                     "Skipping unreadable/invalid Ollama manifest %s: %s",
                     tag_file,
@@ -741,7 +741,7 @@ def _scan_ollama_dir(ollama_dir: Path, limit: Optional[int] = None) -> List[Loca
                         cfg = json.loads(config_blob.read_text(encoding = "utf-8"))
                         model_type = cfg.get("model_type", "")
                         file_type = cfg.get("file_type", "")
-                    except (json.JSONDecodeError, OSError) as e:
+                    except (json.JSONDecodeError, OSError, UnicodeDecodeError) as e:
                         logger.debug(
                             "Could not parse Ollama config blob %s: %s",
                             config_blob,

studio/backend/run.py

@@ -786,7 +786,7 @@ def _remove_pid_file():
             stored = _PID_FILE.read_text(encoding = "utf-8").strip()
             if stored == str(os.getpid()):
                 _PID_FILE.unlink(missing_ok = True)
-    except OSError:
+    except (OSError, UnicodeDecodeError):
         pass
 
 
@@ -934,7 +934,7 @@ def _iter_frontend_fallback_candidates() -> "list[Path]":
             for finder in sp.glob("__editable___*_finder.py"):
                 try:
                     src = finder.read_text(encoding = "utf-8")
-                except OSError:
+                except (OSError, UnicodeDecodeError):
                     continue
                 # Tolerate single/multi-line dict literals; [^}]* rejects nested
                 # dicts, which the setuptools editable template never emits.

studio/backend/utils/hardware/hardware.py

@@ -903,7 +903,7 @@ def _rocm_kfd_gpu_pci_ids() -> list[str]:
                             props[parts[0]] = int(parts[1])
                         except ValueError:
                             continue
-        except OSError:
+        except (OSError, UnicodeDecodeError):
             return []  # unreadable node could be a GPU: fail closed, don't shift
         if props.get("simd_count", 0) <= 0:
             continue  # CPU node, not a GPU

studio/backend/utils/paths/storage_roots.py

@@ -126,7 +126,7 @@ def _xdg_user_dir(key: str) -> Path | None:
     config = Path.home() / ".config" / "user-dirs.dirs"
     try:
         lines = config.read_text(encoding = "utf-8").splitlines()
-    except OSError:
+    except (OSError, UnicodeDecodeError):
         return None
     prefix = f"{key}="
     for line in lines:

studio/backend/utils/utils.py

@@ -114,7 +114,7 @@ def hf_cache_snapshot_dir(model_name: str) -> Optional[Path]:
                 snapshot = repo_dir / "snapshots" / commit
                 if snapshot.is_dir():
                     return snapshot
-            except OSError:
+            except (OSError, UnicodeDecodeError):
                 continue
     return None
 

studio/install_llama_prebuilt.py

@@ -2508,7 +2508,7 @@ def detect_host() -> HostInfo:
                         if _vf.read().strip().lower() == "0x8086":
                             has_intel_gpu = True
                             break
-                except OSError:
+                except (OSError, UnicodeDecodeError):
                     continue
         elif is_windows:
             # Registry first (in-process; see windows_intel_gpu_in_registry).
@@ -4264,7 +4264,7 @@ def free_local_port() -> int:
 def read_log_excerpt(log_path: Path, *, max_lines: int = 60) -> str:
     try:
         content = log_path.read_text(encoding = "utf-8", errors = "replace")
-    except FileNotFoundError:
+    except (FileNotFoundError, UnicodeDecodeError):
         return ""
     return "\n".join(content.splitlines()[-max_lines:])
 
@@ -4680,7 +4680,7 @@ def _wsl_system_rocm_lib_dirs() -> list[str]:
         with open("/proc/version", encoding = "utf-8", errors = "replace") as fh:
             if "microsoft" not in fh.read().lower():
                 return []
-    except OSError:
+    except (OSError, UnicodeDecodeError):
         return []
     out: list[str] = []
     for d in ("/opt/rocm/lib", "/opt/rocm/lib64"):

studio/install_node_prebuilt.py

@@ -535,7 +535,9 @@ def install_lock(lock_path: Path) -> Iterator[None]:
                 break
             except FileExistsError:
                 try:
-                    raw = lock_path.read_text(encoding = "utf-8").strip()
+                    # errors="replace" so an undecodable lock reaches the int()
+                    # below and is treated as a stale PID, not retried forever.
+                    raw = lock_path.read_text(encoding = "utf-8", errors = "replace").strip()
                 except FileNotFoundError:
                     continue
                 stale = False
@@ -669,7 +671,7 @@ def load_metadata(install_dir: Path) -> dict | None:
         return None
     try:
         data = json.loads(path.read_text(encoding = "utf-8"))
-    except (json.JSONDecodeError, OSError):
+    except (json.JSONDecodeError, OSError, UnicodeDecodeError):
         return None
     return data if isinstance(data, dict) else None
 

studio/install_python_stack.py

@@ -776,7 +776,7 @@ def _linux_amd_gfx_from_cpuinfo() -> "str | None":
     """Infer gfx arch from /proc/cpuinfo on integrated AMD APUs (Strix Halo/Point)."""
     try:
         text = Path("/proc/cpuinfo").read_text(encoding = "utf-8", errors = "replace")
-    except OSError:
+    except (OSError, UnicodeDecodeError):
         return None
     if re.search(r"Ryzen AI Max|Radeon 80[0-9][05]S|Strix Halo", text, re.IGNORECASE):
         return "gfx1151"
@@ -828,7 +828,7 @@ def _is_wsl() -> bool:
     try:
         with open("/proc/version", encoding = "utf-8", errors = "replace") as fh:
             return "microsoft" in fh.read().lower()
-    except OSError:
+    except (OSError, UnicodeDecodeError):
         return False
 
 
@@ -856,7 +856,7 @@ def _linux_amd_display_device_present() -> bool:
                     continue
                 if (dev / "class").read_text(encoding = "utf-8").strip().startswith("0x03"):
                     return True
-            except OSError:
+            except (OSError, UnicodeDecodeError):
                 continue
     except OSError:
         pass
@@ -1069,7 +1069,7 @@ def _has_rocm_gpu() -> bool:
                     try:
                         with open(gpu_id_path, encoding = "utf-8") as fh:
                             gpu_id = fh.read().strip()
-                    except OSError:
+                    except (OSError, UnicodeDecodeError):
                         continue
                     if not gpu_id or gpu_id == "0":  # gpu_id 0 = CPU node
                         continue
@@ -1081,7 +1081,7 @@ def _has_rocm_gpu() -> bool:
                     try:
                         with open(props_path, encoding = "utf-8") as fh:
                             props = fh.read()
-                    except OSError:
+                    except (OSError, UnicodeDecodeError):
                         continue  # can't confirm vendor -- skip
                     if not re.search(r"\bvendor_id\s+4098\b", props):
                         continue

studio/prebuilt_core.py

@@ -1078,7 +1078,9 @@ def install_lock(lock_path: Path) -> Iterator[None]:
             except FileExistsError:
                 stale = False
                 try:
-                    raw = lock_path.read_text(encoding = "utf-8").strip()
+                    # errors="replace" so an undecodable lock reaches the int()
+                    # below and is treated as a corrupt PID, not retried forever.
+                    raw = lock_path.read_text(encoding = "utf-8", errors = "replace").strip()
                 except FileNotFoundError:
                     # Lock vanished between our open and read -- retry
                     continue
@@ -2070,7 +2072,7 @@ def load_prebuilt_metadata(ops: ModuleOps, install_dir: Path) -> dict[str, Any]
         return None
     try:
         payload = json.loads(path.read_text(encoding = "utf-8"))
-    except (json.JSONDecodeError, OSError):
+    except (json.JSONDecodeError, OSError, UnicodeDecodeError):
         return None
     return payload if isinstance(payload, dict) else None
 

unsloth_cli/_inference.py

@@ -100,7 +100,7 @@ def _json_rank_count_from_env(name: str) -> Optional[int]:
         else:
             with open(value, "r", encoding = "utf-8") as f:
                 data = json.load(f)
-    except (OSError, json.JSONDecodeError):
+    except (json.JSONDecodeError, OSError, UnicodeDecodeError):
         return None
     if isinstance(data, list):
         return len(data)

unsloth_cli/commands/studio.py

@@ -335,7 +335,7 @@ def _iter_editable_studio_source_roots(venv_dir: Path):
             for finder in sp.glob("__editable___*_finder.py"):
                 try:
                     src = finder.read_text(encoding = "utf-8")
-                except OSError:
+                except (OSError, UnicodeDecodeError):
                     continue
                 # Tolerate single- or multi-line dict literals; [^}]* still
                 # rejects nested dicts, which the setuptools template never