ElementTree: invalid paths raise KeyError instead of SyntaxError, and C and Python disagree on them #161808

Bug report

Bug description

Some malformed paths make find(), findall() and findtext() raise a bare KeyError from ElementPath, while other malformed paths raise SyntaxError("invalid path"). Which paths reach that code also depends on the implementation:

import xml.etree.ElementTree as ET

root = ET.Element("root")
root.find("@")     # KeyError: '@'   (C accelerator and pure Python)
root.find("x=y")   # pure Python: KeyError: '='   C accelerator: None
root.find(")")     # pure Python: KeyError: ')'   C accelerator: None
root.find("]")     # pure Python: KeyError: ']'   C accelerator: None

(The pure-Python results can be reproduced by blocking the accelerator: sys.modules["_elementtree"] = None before importing xml.etree.ElementTree.)

Cause. ElementPath.iterfind() looks up each token with ops[token[0]]. Tokens that start no supported step (@, =, (, ), ], !=, ::) are not in ops, so the lookup raises KeyError. The C accelerator only hands a path to ElementPath when its checkpath() finds one of / * [ @ .; any other path is compared to tags literally. So x=y, ) and ] never reach the lookup in C, but always do in the pure-Python implementation.

Expected. SyntaxError("invalid path") for every path ElementPath cannot parse, as it already raises for other malformed paths. For example, replace the lookup with ops.get(token[0]) and raise SyntaxError when it is missing.

Whether the C fast path should also send such paths to ElementPath, so that both implementations agree, is a separate choice. It would change results for existing code that looks up literal tags such as "a b" or "a:b", so I am only proposing the SyntaxError change.

I'm happy to open a PR.

CPython versions tested on:

CPython main branch, 3.14, 3.13, 3.12

Operating systems tested on:

Linux

Linked PRs

• gh-161808: Raise SyntaxError for unsupported ElementPath tokens #161810