[Bug] [Feature] Fix issue #1039 by exposing And, Not regex operators from llguidance
#1043
Conversation
guidance/library/_json.py
Outdated
| names = set(properties.keys()) | set(required) | ||
| if len(names) > 0: | ||
| # If there are any properties, we need to disallow them as additionalProperties | ||
| key_grammar = as_regular_grammar( | ||
| And([ | ||
| lexeme(r'"([^"\\]|\\["\\/bfnrt]|\\u[0-9a-fA-F]{4})*"'), | ||
| Not(lexeme('"(' + '|'.join(quote_regex(name) for name in names) + ')"')), | ||
| ]), | ||
| lexeme = True, | ||
| ) | ||
| else: | ||
| key_grammar = _gen_json_string() |
There was a problem hiding this comment.
Beyond the addition of new types and a bit of reorganization of _grammar.py, here's the meat of the bug fix.
There was a problem hiding this comment.
You could put the quotes outside the whole thing and use Or over String. Maybe a bit cleaner
| def test_out_of_order_non_required_properties_not_validated_as_additionalProperties(self): | ||
| schema = { | ||
| "type": "object", | ||
| "properties": {"a": {"const": "foo"}, "b": {"const": "bar"}}, | ||
| "required": ["b"], | ||
| } | ||
| test_string = '{"b": "bar", "a": "BAD"}' | ||
| grammar = gen_json(schema=schema) | ||
| assert grammar.match(test_string) is None |
There was a problem hiding this comment.
Test for issue #1039 which fails on main
|
Codecov ReportAttention: Patch coverage is
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.
Additional details and impacted files@@ Coverage Diff @@
## main #1043 +/- ##
==========================================
- Coverage 71.39% 63.38% -8.02%
==========================================
Files 63 63
Lines 4670 4708 +38
==========================================
- Hits 3334 2984 -350
- Misses 1336 1724 +388 ☔ View full report in Codecov by Sentry. |
Fixes issue #1039 by ensuring that the keys for JSON properties validated against
additionalPropertiesaren't allowed to match any of the keys for the properties themselves.I'm still a bit unsure about the API of
AndandNot, as well as what the right type hierarchy is (e.g. should they be subclasses ofRegularGrammar?), but I am comfortable with punting on those questions a bit as they are very much not part of the "public" API yet.