Python 3.10: Parenthesized context managers

[gfokkema]

Besides match case, python 3.9 alpha6 / 3.10 also added parenthesized imports and context-managers. Parenthesized imports works no problem in jedi, but parenthesized context managers lead to a syntax error:

>>> import jedi
>>> source = '''
... with (
...     open('/dev/stdout', 'w') as a,
...     open('/dev/stdout', 'w') as b,
... ):
...     a.write('a')
...     b.write('b')
... '''
>>> script = jedi.Script(source, path='example.py')
>>> [print(x, x.get_message()) for x in script.get_syntax_errors()]
<SyntaxError from=(3, 29) to=(3, 31)> SyntaxError: invalid syntax
<SyntaxError from=(4, 29) to=(4, 31)> SyntaxError: invalid syntax
<SyntaxError from=(5, 0) to=(5, 1)> SyntaxError: invalid syntax
<SyntaxError from=(6, 0) to=(6, 4)> IndentationError: unexpected indent

See:

• https://docs.python.org/3/whatsnew/3.10.html#parenthesized-context-managers

Probably relates to:

• davidhalter/jedi#1830

[davidhalter]

Moved to parso. As you said this is a more general issue with the PEG grammar (parso is only able to parse LL at the moment).

[isidentical]

@davidhalter if we agree on some sort of a verification process (or use the existing syntax error generation), we could simply add a rule like this:

namedexpr_test: asexpr_test [':=' asexpr_test]
asexpr_test: test ['as' test]

Which in theory (and also in practice, currently used by black) allows parenthesized context managers. Then if this is used anywhere outside of a with-statement's context expression, we would raise an error. This would be another inconsistency in the grammar that would be eliminated in a later stage, just like the LHS of many expressions.

Not sure if you think this would worth the effort without the full pattern matching stuff though.

[davidhalter]

Not sure if you think this would worth the effort without the full pattern matching stuff though.

Yeah, that's mostly what I'm worried about as well. It's a good idea, but full pattern matching is probably the more pressing issue.