Harden Malloy compatibility boundaries - #334
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.
Reviewed commit: 620bbacde0
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".
| target_model: str | None = Field( | ||
| default=None, | ||
| description=( | ||
| "Canonical related model when the relationship name is a role alias. " | ||
| "When omitted, name remains the target model for backwards compatibility." | ||
| ), |
There was a problem hiding this comment.
When an imported Malloy join uses an alias such as buyer is customers, name is now buyer while the graph only contains the canonical customers model. Existing exporters still resolve targets through relationship.name; for example, Cube uses resolved_models.get(relationship.name) in sidemantic/adapters/cube.py:1795, and TMDL uses graph.models.get(rel.name) in sidemantic/adapters/tmdl.py:1834, so CLI conversions silently omit these joins. Update downstream exporters to use related_model while preserving the role where supported, or explicitly block the conversion where aliases cannot be represented.
AGENTS.md reference: AGENTS.md:L7-L9
Useful? React with 👍 / 👎.
| Relationship( | ||
| name=name, | ||
| target_model=target_name if aliased_reference else None, | ||
| type=rel_type, | ||
| foreign_key=self._collapse_join_keys(foreign_keys), |
There was a problem hiding this comment.
For directory/CLI loading, _infer_relationships runs after Malloy parsing but checks whether a relationship already targets a model using r.name == target (sidemantic/loaders.py:1613-1618). With join_one: buyer is customers on customer_id = buyer.id, this relationship has name='buyer' and target_model='customers', so a customer_id dimension causes inference to add a second canonical customers relationship and a reverse edge. This exposes a join path absent from the Malloy source and can introduce ambiguity; inference should compare r.related_model for role-aliased relationships.
AGENTS.md reference: AGENTS.md:L7-L8
Useful? React with 👍 / 👎.
1 participant
Makes Malloy imports fail closed instead of silently changing unsupported semantics. Adds invariant source filters, intrinsic schema exposure, exact and role-aliased joins, module visibility, typed dialect-aware expressions, bounded one-stage query mapping, structured fidelity diagnostics, and transactional loader behavior.\n\nAdds focused Python coverage and an opt-in official Malloy differential harness, and rewrites the compatibility documentation around the actual supported boundary.