[vaibhavsrv]

Why this matters

In ods/scripts/assign_gpus.py, the topology file is opened via open(args.topology) without an explicit encoding parameter. On Python's default behavior, open() uses the locale-preferred encoding — cp1252 on Windows, ascii in minimal containers. When the topology JSON file contains non-ASCII characters (e.g. GPU model names with trademark symbols or non-Latin characters), json.load() encounters a UnicodeDecodeError that is not caught by the narrow except FileNotFoundError block, producing an unhandled traceback rather than a clean error exit. Additionally, PermissionError and IsADirectoryError — both subclasses of OSError — also escape the narrow FileNotFoundError handler and crash with a raw Python traceback.

This patch adds encoding="utf-8" to the open() call and widens the exception clause from except FileNotFoundError to except OSError, ensuring all file access failures produce a clean error message and a non-zero exit code consistent with the script's existing error reporting style.

Validation

• Repro baseline: Wrote a topology file containing a non-ASCII GPU name (RTX™ 4090) as UTF-8; confirmed that open(args.topology) without encoding raised UnicodeDecodeError in locales where the default is not UTF-8. Confirmed that except FileNotFoundError did not catch permission failures.
• Post-fix: Verified ods/tests/test_assign_gpus_topology_encoding.py passes with exit code 0; missing and invalid-JSON topology files produce clean ERROR messages on stderr with non-zero exit codes.
• GPU assignment suites: 3 passed. New-test syntax and diff checks pass; new regressions wired into Linux CI.

Overlap check

• fix(gpu): respect dotenv override precedence and export prefix in env reader #5877 (vaibhavsrv): Addressed dotenv override precedence and export prefix in GPU env reader (gpu.py); zero overlap in assign_gpus.py topology loading.
• fix(model-store): specify UTF-8 encoding in registry read_text to prevent UnicodeDecodeError #5881 (vaibhavsrv): Addressed UTF-8 encoding in register-model-store.py; this addresses the same class of encoding bug in a separate script.
• Probe the live model on the real llama-server host port #5860 (0xacee): Addressed LLM port fallback in build-installation-context.py; zero overlap.

Risk / AI disclosure

AI-assisted investigation, implementation and CLI regressions. This strengthens portable JSON reading and OSError handling in GPU topology loading; GPU assignment logic, driver interaction, and capacity decisions are not altered. Independent human review and platform/runtime qualification remain gates. No running configuration, deployment or upstream merge changed.

Follow-up integration evidence

Composed with #5883 at 9a254c687 without conflicts. Production and test diffs passed together; adjacent GPU assignment capacity and spare capacity contracts remain intact.
Backlog composition was local-only (production/test diffs, excluding workflow/Makefile wiring); it is not an upstream merge or independent human approval. Declared live-review gates remain open.

[Lightheartdevs]

Thanks for this contribution. public-beta was promoted into main on 2026-09-24 and no longer receives changes, so we're closing pull requests that target it. This isn't a judgment on the change itself. If it's still needed, please rebase onto main and open a focused PR. See #7253 for details and the contribution policy.