vllm - 💡(How to fix) Fix macOS source checkout resolves to UnspecifiedPlatform (empty device_type) [1 pull requests]

Description: When running vLLM from a source checkout on macOS (no pip install), cpu_platform_plugin() returns None because the sys.platform.startswith("darwin") fallback is unreachable. This causes resolve_current_platform_cls_qualname() to select UnspecifiedPlatform, which has device_type = "", crashing any torch.device("") call.

Minimal Reproducible Example:

# On macOS, from a source checkout (no pip install):
from vllm.platforms import current_platform
print(current_platform.device_type)  # "" → torch.device("") crashes

Expected behavior: cpu_platform_plugin() returns "vllm.platforms.cpu.CpuPlatform" on macOS regardless of package install state.

Actual behavior: RuntimeError: Device string must not be empty

Hypothesis: vllm_version_matches_substr("cpu") re-raises PackageNotFoundError in source checkouts. The macOS fallback (sys.platform.startswith("darwin")) is inside the same try block, so the exception skips it entirely.

Environment: macOS (darwin), Python 3.13, source checkout without pip install.