microsoft/dataverse-sdk-dev

Dataverse SDK Development Guide

Overview

This skill provides guidance for developers working on the PowerPlatform Dataverse Client Python SDK repository itself (not using the SDK).

Best Practices

API Design

1. Public methods in operation namespaces - New public methods go in the appropriate namespace module under src/PowerPlatform/Dataverse/operations/ (records.py, query.py, tables.py, batch.py). The client.py file exposes these via namespace properties (client.records, client.query, client.tables, client.batch). Public types and constants live in their own modules (e.g., models/metadata.py, models/batch.py, common/constants.py)
2. Every public method needs README example - Public API methods must have examples in README.md
3. Reuse existing APIs - Always check if an existing method can be used before making direct Web API calls
4. Update documentation when adding features - Keep README and SKILL files (both copies) in sync
5. Consider backwards compatibility - Avoid breaking changes
6. Internal vs public naming - Modules, files, and functions not meant to be part of the public API must use a _ prefix (e.g., _odata.py, _relationships.py). Files without the prefix (e.g., constants.py, metadata.py) are public and importable by SDK consumers

Dataverse Property Naming Rules

Dataverse uses two different naming conventions for properties. Getting this wrong causes 400 errors that are hard to debug.

| Property type | Name convention | Example | When used | |---|---|---|---| | Structural (columns) | LogicalName (always lowercase) | new_name, new_priority | $select, $filter, $orderby, record payload keys | | Navigation (relationships / lookups) | Navigation Property Name (usually SchemaName, PascalCase, case-sensitive) | new_CustomerId, new_AgentId | $expand, @odata.bind annotation keys |

Navigation property names are case-sensitive and must match the entity's $metadata. Using the logical name instead of the navigation property name results in 400 Bad Request errors.

Critical rule: The OData parser validates @odata.bind property names case-sensitively against declared navigation properties. Lowercasing [email protected] to [email protected] causes: ODataException: An undeclared property 'new_customerid' which only has property annotations...

SDK implementation:

• _lowercase_keys() lowercases all keys EXCEPT those containing @odata. (preserves navigation property casing in @odata.bind keys)
• _lowercase_list() lowercases $select and $orderby params (structural properties)
• $expand params are passed as-is (navigation properties, PascalCase)
• _convert_labels_to_ints() skips @odata. keys entirely (they are annotations, not attributes)

When adding new code that processes record dicts or builds query parameters:

• Always use _lowercase_keys() for record payloads. Never manually call .lower() on all keys
• Never lowercase $expand values or @odata.bind key prefixes
• If iterating record keys, skip keys containing @odata. when doing attribute-level operations

Code Style

6. No emojis - Do not use emoji in code, comments, or output
7. Standardize output format - Use [INFO], [WARN], [ERR], [OK] prefixes for console output
8. No noqa comments - Do not add # noqa: BLE001 or similar linter suppression comments
9. Document public APIs - Add Sphinx-style docstrings with examples for public methods
10. Define all in module files - Each module declares its own exports via __all__ (e.g., errors.py defines __all__ = ["HttpError", ...]). Package __init__.py files should not re-export or redefine another module's __all__; they use __all__ = [] to indicate no star-import exports.
11. Run black before committing - Always run python -m black <changed files> before committing. CI will reject unformatted code. Config is in pyproject.toml under [tool.black].

Docstring Type Annotations (Microsoft Learn Compatibility)

This SDK's API reference is published on Microsoft Learn. The Learn doc pipeline parses :type: and :rtype: directives differently from standard Sphinx -- every word between :class: references is treated as a separate cross-reference (<xref:word>). Using Sphinx-style :class:\list` of :class:`str`produces brokenxref:of` links on Learn.

Rules for :type: and :rtype: directives:

• Use Python bracket notation for generic types: list[str], dict[str, typing.Any], list[dict]
• Use or (without :class:) for union types: str or None, dict or list[dict]
• Use bracket nesting for complex types: collections.abc.Iterable[list[dict]]
• Use ~ prefix for SDK types to show short name: list[~PowerPlatform.Dataverse.models.record.Record]
• :class: is fine for single standalone types: :class:\str`, :class:`bool``

Never use :class:\X` of :class:`Y`or:class:`X` mapping :class:`Y` to :class:`Z`-- the wordsof, mapping, tobecome brokenxref:` links.

Correct examples:

:type data: dict or list[dict]
:rtype: list[str]
:rtype: collections.abc.Iterable[list[~PowerPlatform.Dataverse.models.record.Record]]
:type select: list[str] or None
:type columns: dict[str, typing.Any]

Wrong examples (NEVER use):

:type data: :class:`dict` or :class:`list` of :class:`dict`
:rtype: :class:`list` of :class:`str`
:type columns: :class:`dict` mapping :class:`str` to :class:`typing.Any`