scripts/build_wheel.py (5)

332-341: Harden ldd invocation: quote path, tolerate non-zero exit, and avoid parameter shadowing

• ldd can legitimately exit non-zero (e.g., non-dynamic or unresolved deps), which will raise due to check=True in build_run; you currently don’t handle that, so stub generation may fail prematurely.
• lib_name is re-used inside the loop, shadowing the function parameter; this is confusing and error-prone.
• Use quoting to handle spaces and avoid shell injection.

Suggested patch:

-def check_missing_libs(lib_name: str) -> list[str]:
-    result = build_run(f"ldd {lib_name}", capture_output=True, text=True)
-    missing = []
-    for line in result.stdout.splitlines():
-        if "not found" in line:
-            lib_name = line.split()[0]  # Extract the library name before "=> not found"
-            if lib_name not in missing:
-                missing.append(lib_name)
-    return missing
+def check_missing_libs(lib_path: str) -> list[str]:
+    # Use check=False so we can still parse output when ldd exits non‑zero.
+    proc = run(f'ldd "{lib_path}"', shell=True, capture_output=True, text=True, check=False)
+    missing: list[str] = []
+    for line in (proc.stdout or "").splitlines():
+        if "not found" in line:
+            soname = line.split()[0]  # e.g. "libcuda.so.1"
+            if soname not in missing:
+                missing.append(soname)
+    return missing

---

345-346: Pass an absolute library path and clarify parameter naming

Relying on the current working directory to resolve the library name is brittle. Pass the absolute path to the checked library and reflect that in the parameter name.

Apply:

-def generate_python_stubs_linux(binding_type: str, venv_python: Path,
-                                deep_ep: bool, binding_lib_name: str):
+def generate_python_stubs_linux(binding_type: str, venv_python: Path,
+                                deep_ep: bool, binding_lib_path: str):
@@
-    missing_libs = check_missing_libs(binding_lib_name)
+    missing_libs = check_missing_libs(binding_lib_path)

Also applies to: 354-355

---

807-810: Variable name misleads: this is a file, not a directory

binding_lib_dir points to a file (a .so/.pyd), not a directory. Rename for clarity.

-        binding_lib_dir = get_binding_lib(binding_type, "bindings")
-        binding_lib_file_name = binding_lib_dir.name
-        install_file(binding_lib_dir, pkg_dir)
+        binding_lib_path = get_binding_lib(binding_type, "bindings")
+        binding_lib_file_name = binding_lib_path.name
+        install_file(binding_lib_path, pkg_dir)

---

849-850: Pass absolute path into generate_python_stubs_linux to decouple from cwd

Even though you're in working_directory(pkg_dir), making the dependency explicit improves robustness.

-                        bool(deep_ep_cuda_architectures), binding_lib_file_name)
+                        bool(deep_ep_cuda_architectures), str(pkg_dir / binding_lib_file_name))

---

332-332: Minor typing note: list[str] requires Python ≥3.9

The file uses list[str] in multiple places (pre-existing), which isn’t valid in Python 3.8 without from future import annotations. If 3.8 remains a supported minimum, consider one of:

• Add at the top of the file: from future import annotations
• Or use typing.List[str] consistently

I can prepare a follow-up patch if you confirm the target Python versions in CI.

Would you like me to add from future import annotations at the top of this module to align with Python 3.8 compatibility?

Configuration used: .coderabbit.yaml
Review profile: CHILL
Plan: Pro

💡 Knowledge Base configuration:

• MCP integration is disabled by default for public repositories
• Jira integration is disabled by default for public repositories
• Linear integration is disabled by default for public repositories

You can enable these sources in your CodeRabbit configuration.