Repository navigation
traceback formatting raises FileNotFoundError with empty sys.path #159693
Description
Bug report
On CPython 3.15.0, formatting a ModuleNotFoundError raises a second exception when sys.path contains no existing directory.
import sys
import traceback
import importlib.resources.readers # Load dependencies before clearing sys.path.
sys.path[:] = []
error = ModuleNotFoundError(
"No module named 'example_missing'",
name="example_missing",
)
print(traceback.format_exception_only(error))Expected: Expected: A formatted ModuleNotFoundError. (["ModuleNotFoundError: No module named 'example_missing'\n"])
Actual: exception raised (FileNotFoundError: MultiplexedPath must contain at least one path)
The same condition formats successfully on CPython 3.14.8. A path containing only nonexistent directories also triggers the failure; adding an existing directory prevents it.
In _find_incompatible_extension_module() (https://github.com/python/cpython/blob/v3.15.0/Lib/traceback.py#L2057), filtering sys.path can leave no arguments for MultiplexedPath, whose exception escapes the formatter.
Found on Linux ARM64 using a uv-managed CPython 3.15.0 build, and reproduced on macOS
- Linux: Python 3.15.0 (main, Oct 14 2026, 01:07:17) [Clang 23.1.3 ]
- macOS: Python 3.15.0 (main, Oct 9 2026, 15:11:47) [Clang 22.1.3 ]
CPython versions tested on:
3.15
Operating systems tested on:
Linux
Activity
picnixz commentedon Oct 19, 2026
I would personally think it's weird to have no
sys.pathat all but the fix should be fairly simple so it's fine to fix this as well. Did you encounter this in a legitimate session btw?heikkitoivonen commentedon Oct 19, 2026
I was adding support for 3.15.0 on my https://pythoncomplexity.com/ site, and Claude/Codex found this when working on unit tests.
picnixz commentedon Oct 19, 2026
Yeah we don't wan't to do defensive programming actually. I honestly can't imagine a situation where you would empty your path for a long time. However, what I can expect is:
I don't know if
format_exception_onlyis called and if it raises, then is ignored in the interpreter somewhere and rewritten in C only just in case, but if we useformat_exception_onlyimplicitly when raising exceptions, then I would accept a fix. If we don't usetracebackfor formatting exceptions in the REPL or elsewhere, I would rather reject the fix.General rule: please don't report issues found by LLMs if they are not realistic in general (hence my question about a real situatio happening).