gh-129598: allow multi stmts for ast single with ';'

[tom-pytel]

Had to add a bit more than just a visitor node due to possibility of block statements, so the following are all handled correctly:

$ ./python
Python 3.14.0a4+ (heads/astsingle:9ba281d871, Feb  3 2025, 09:15:49) [GCC 11.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from ast import *

>>> print(unparse(parse('i = 1; "expr"; raise Exception', mode='exec')))
i = 1
'expr'
raise Exception

>>> print(unparse(parse('i = 1; "expr"; raise Exception', mode='single')))
i = 1; 'expr'; raise Exception

>>> print(unparse(parse('if i:\n "expr"\nelse:\n raise Exception', mode='single')))
if i:
    'expr'
else:
    raise Exception

>>> print(unparse(parse('@decorator\ndef func():\n "docstring"\n i = 1; "expr"; raise Exception', mode='single')))
@decorator
def func():
    """docstring"""
    i = 1
    'expr'
    raise Exception

• Issue: ast.parse(..., mode='single') parses of multiple statements which are then not unparsed #129598

[picnixz]

Can we have tests please?

[picnixz]

We might want to address the cleanup parts in a separate PR as it's a bit orthogonal. If possible I'd like @JelleZijlstra's opinion on whether we should worry about the state of the visitor after an error or not, and whether we should worry about whether it's reusable or not.

[picnixz]

Another question is whether the mapping precedences should be cleared in visit() as well.

[tom-pytel]

whether we should worry about whether it's reusable or not.

I don't think so. Its not exported from ast module, recreated on every call to unparse() and its not even really a class, just a wrapper around some very simple state.

[picnixz]

Except for my question on the precedences mapping, I think it's fine.

[tom-pytel]

Except for my question on the precedences mapping, I think it's fine.

Sorry, missed that, what is the question?

Ah, NM, its up there. Probably for someone else do decide and another PR?