feat(datafabric-tool): ground relationship-field joins in text-to-SQL prompt [DS-8791]#962
feat(datafabric-tool): ground relationship-field joins in text-to-SQL prompt [DS-8791]#962milind-jain-uipath wants to merge 6 commits into
Conversation
… prompt Surface entity relationship (foreign-key) fields to the Data Fabric SQL sub-graph so it can join related entities. A relationship field stores the related record's Id; the prompt now instructs the model to join on related.Id = parent.<relField> and project explicit related columns. - FieldSchema carries the related entity's SQL table, join key (Id), and reference field; foreign-key fields are tagged "fk"; add is_relationship. - build_entity_context populates these from the SDK field metadata (reference_entity, reference_field, field_display_type). - The rendered schema adds a per-entity "Relationships" subsection with the join expression, gated to related entities present in the set. - The v1 prompt adds a RELATIONSHIP FIELDS section: LEFT JOIN for optional relationships, INNER JOIN when the related record must exist or is filtered. - SQL_CONSTRAINTS permits LEFT JOIN only for relationship/foreign-key joins on Id; general joins remain INNER-only. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
milind-jain-uipath
changed the title
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
Pull request overview
This PR improves the Data Fabric text-to-SQL prompting by explicitly grounding relationship/foreign-key fields as joins (instead of treating them like human-readable attributes), and relaxes join constraints to allow LEFT JOIN specifically for those relationship joins.
Changes:
- Enriches
FieldSchemawith reference/relationship metadata (related table + representative field) and tags FK fields asfk. - Updates the prompt builder to render a per-entity Relationships subsection with a concrete join expression, only when the related entity is present in the entity set.
- Expands the v1 prompt + SQL constraints to document relationship-field semantics and when
LEFT JOINvsINNER JOINis appropriate.
Reviewed changes
Copilot reviewed 5 out of 5 changed files in this pull request and generated 1 comment.
Show a summary per file
| File | Description |
|---|---|
tests/agent/tools/test_datafabric_prompt_builder.py |
Adds tests covering relationship join rendering and prompt documentation updates. |
src/uipath_langchain/agent/tools/datafabric_tool/prompts/v1.py |
Documents relationship-field semantics and join intent guidance. |
src/uipath_langchain/agent/tools/datafabric_tool/models.py |
Extends schema model with relationship metadata and fk display tagging. |
src/uipath_langchain/agent/tools/datafabric_tool/datafabric_prompts.py |
Updates SQL constraints to allow LEFT JOIN only for FK/relationship joins on Id. |
src/uipath_langchain/agent/tools/datafabric_tool/datafabric_prompt_builder.py |
Populates reference metadata from SDK fields and renders relationship join hints in the schema context. |
| type_name = field.sql_type.name if field.sql_type else "unknown" | ||
| ref_entity_table: str | None = None | ||
| ref_field_name: str | None = None | ||
| if field.is_foreign_key or field.field_display_type == "Relationship": | ||
| ref_entity = getattr(field, "reference_entity", None) |
![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: 8487c17f6a
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".
| - LEFT JOIN is allowed ONLY for relationship (foreign-key) joins on the related | ||
| entity's Id (see "Relationship fields" guidance) — use it for optional | ||
| relationships to keep parent rows |
There was a problem hiding this comment.
When callers render this builder with prompt_version="v0", the legacy SQL_EXPERT_SYSTEM_PROMPT from this same module is emitted before these relaxed constraints and still lists LEFT JOIN under unsupported join constructs. That leaves the v0 prompt simultaneously forbidding LEFT JOINs and allowing/recommending them for optional relationship fields, so optional relationship questions in that compatibility mode can still be steered away from the only join type the new relationship section is trying to ground. Update the legacy prompt or gate the new relationship/LEFT JOIN guidance by prompt version.
Useful? React with 👍 / 👎.
- Drop the redundant "never SELECT parent.*" aside (SQL_CONSTRAINTS already forbids SELECT *). - Tie join-type choice to the relationship field's required flag: a required field -> INNER JOIN is safe (related record always exists); optional -> LEFT JOIN to keep parent rows where it is unset. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Detect a relationship as (is_foreign_key OR fieldDisplayType == "Relationship") and use that single condition both to set is_foreign_key on the FieldSchema and to extract the reference target, so a Relationship-typed field without the is_foreign_key flag is still tagged fk and rendered in the Relationships section. Read field_display_type via getattr for safety. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ionship-join-grounding # Conflicts: # src/uipath_langchain/agent/tools/datafabric_tool/datafabric_prompt_builder.py # src/uipath_langchain/agent/tools/datafabric_tool/models.py # tests/agent/tools/test_datafabric_prompt_builder.py
|
What
Grounds the Data Fabric SQL sub-graph to join related (foreign-key) entities.
A relationship field stores the related record's Id, not its attributes. The prompt now surfaces each entity's relationships and instructs the model to join on
related.Id = parent.<relField>, projecting explicit related columns.Changes
models.py:FieldSchemacarriesref_entity_table,ref_join_key(Id), andref_field_name; foreign-key fields are taggedfk; addsis_relationship.datafabric_prompt_builder.py: populates the reference fields from SDK field metadata (reference_entity,reference_field,field_display_type); renders a per-entity "Relationships" subsection with the join expression, gated to related entities present in the set.prompts/v1.py: adds aRELATIONSHIP FIELDSsection — join onrelated.Id = parent.<relField>, project explicit related columns, and choose the join type by intent: LEFT JOIN for optional relationships (keep parent rows), INNER JOIN when the relationship field isrequired, the related record must exist, or you filter on the related entity's columns.datafabric_prompts.py:SQL_CONSTRAINTSpermits LEFT JOIN only for relationship/foreign-key joins onId; general joins remain INNER-only.Dependency
Depends on the related-entity auto-registration in UiPath/Agents#5725: the related entity must be present in the agent's entity set (and thus routable) for the join to resolve at query time. This PR grounds the SQL; that PR makes the related entity queryable.
Testing
pytest tests/agent/toolspasses.ruff check,ruff format --check, andmypyclean onsrc/uipath_langchain/agent.🤖 Generated with Claude Code