frame.f_lineno can be None

[fabioz]

Bug report

As far as I know frame.f_lineno should never be None, yet, in the case below it is:

import sys
def tracefunc(frame, event, arg):
    if frame.f_lineno is None:
        raise AssertionError('frame.f_lineno is None!', frame)
    return tracefunc
sys.settrace(tracefunc)

# Apparently anything we import would give a frame.f_lineno = None during tracing.
import threading

Your environment

• CPython versions tested on: 3.11.0b3 (main, Jun 1 2022, 13:29:14) [MSC v.1932 64 bit (AMD64)]
• Operating system and architecture: Windows

[markshannon]

The line number is meaningless when calling a module during an import.
If there were a line number, it would create spurious "line" events when tracing.

If you change tracefunc to:

def tracefunc(frame, event, arg):
    if frame.f_lineno is None and event != "call":
        raise AssertionError('frame.f_lineno is None!', frame)
    return tracefunc

No assertion should be raised.

[markshannon]

We might be able to set the line number to 0, https://peps.python.org/pep-0626/#the-f-lineno-attribute, without producing a line event.

[markshannon]

As far as I know frame.f_lineno should never be None

It is possible in unusual circumstances, such as when the cycle GC invokes Python or a keyboard interrupt, but it shouldn't happen in normal code execution.

[fabioz]

As far as I know frame.f_lineno should never be None

It is possible in unusual circumstances, such as when the cycle GC invokes Python or a keyboard interrupt, but it shouldn't happen in normal code execution.

This seems a bit surprising for me... I went on looking for docs but there's surprisingly few information about this (my use case is that I was typing things with cython and it's breaking due to being None in Python 3.11 -- I had never seen a report for that previously where frame.f_lineno could be anything but an integer).

Given that the first valid line number is 1, if it's possible to set it to 0 to indicate an invalid number while keeping the same type would make things better IMHO.

[pablogsal]

Moving this back to release blocker.

[tiran]

Main and 3.11 branch are now fixed. Thank you @iritkatriel for the fix and @ambv for the backport.