Normalise selectors in plan synthesiser and verifier #3
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| @staticmethod | ||
| def _normalise_selector(raw_selector: str) -> str: | ||
| token = raw_selector.split(",", 1)[0].strip() | ||
| token = token.rstrip(")").strip() | ||
| token = token.strip("\"'") |
There was a problem hiding this comment.
The normaliser now calls raw_selector.split(",", 1) and discards everything after the first comma. Legitimate CSS selectors often contain commas inside attribute values ([aria-label="Hello, world"]) or intentionally select multiple elements (#foo, #bar). After this change _plan_compatible will normalise such selectors to only the prefix, so even when the context contains the full selector the synthesiser (and the identical logic in PlanVerifier._normalise_selector) will reject a compatible plan as missing its selector. The normalisation should respect quoted segments or only strip comma-separated arguments rather than blindly truncating.
Useful? React with 👍 / 👎.
| if "(" in token and token.endswith(")"): | ||
| prefix, arg_str = token.split("(", 1) | ||
| verb_name = prefix.strip() | ||
| args = [arg_str[:-1]] if arg_str.endswith(")") else [arg_str] | ||
| if arg_str.endswith(")"): | ||
| arg_str = arg_str[:-1] | ||
| raw_args = [part for part in arg_str.split(",") if part.strip()] | ||
|
|
||
| def _clean_argument(value: str) -> str: | ||
| cleaned = value.strip().rstrip(")").strip() | ||
| cleaned = cleaned.strip("\"'") | ||
| return cleaned | ||
|
|
||
| args = [_clean_argument(part) for part in raw_args] |
There was a problem hiding this comment.
When _handle_plan_sketch splits arg_str by "," it does not consider quoting, so a token such as Type(#field, "hello, world") produces args = ["#field", "hello", "world"] instead of keeping the comma inside the second argument. Any plan where argument values legitimately contain commas or closing parentheses will now be broken and the original string cannot be reconstructed. Using a parsing approach that honours quoted strings or balanced parentheses would avoid fragmenting arguments.
Useful? React with 👍 / 👎.
Summary
Testing
https://chatgpt.com/codex/tasks/task_e_68f6f8f5417c83279350661ed4f833e9