Bug report

Bug description:

The ParseError class from the xml module does not cast the msg attribute to str when running on the pure-Python version, here is a MRE:

import xml.etree.ElementTree as etree
try:
    etree.XML("<")
except etree.ParseError as e:
    print("XMLParser:", etree.XMLParser)
    print("args:", e.args)
    print("msg:", repr(e.msg))
    print("msg type:", type(e.msg))

On regular CPython using the C-accelerator:

XMLParser: <class 'xml.etree.ElementTree.XMLParser'>
args: ('unclosed token: line 1, column 0',)
msg: 'unclosed token: line 1, column 0'
msg type: <class 'str'>

If you force the execution of the Python module:

XMLParser: <class 'xml.etree.ElementTree.XMLParser'>
args: (ExpatError('unclosed token: line 1, column 0'),)
msg: ExpatError('unclosed token: line 1, column 0')
msg type: <class 'pyexpat.ExpatError'>

The issue is this function:

    def _raiseerror(self, value):
        err = ParseError(value)
        err.code = value.code
        err.position = value.lineno, value.offset
        raise err

At all call sites, value is an exception, so effectively the first argument is always an exception. Note that ParseError is a sub-class of SyntaxError.
In the C-version, the first agument is always cast to a str.
I encountered this bug through PyPy that use the Python version of lxml.

CPython versions tested on:

CPython main branch

Operating systems tested on:

macOS

Linked PRs

• gh-158166: Make ElementTree ParseError messages strings #158168