feat: SDK-native column constraints (create, update, typed read) — Layer 1 of cutting raw-urllib usage

[arorashivam96]

SDK-native column constraints — Layer 1 of reducing raw-urllib usage

The problem this fixes

In the Dataverse plugin's telemetry, a large share of traffic is raw urllib (hand-rolled HTTP) instead of the managed SDK. The single biggest driver is schema metadata on the dv-metadata skill: agents were forced to hand-build calls to EntityDefinitions/Attributes because the SDK could not express column constraints. Concretely, three operations had no SDK path:

1. Create a column with constraints (MaxLength, MinValue/MaxValue, Format, RequiredLevel, a custom DisplayName).
2. Update an existing column's constraints (requires the counter-intuitive PUT-not-PATCH plus a derived @odata.type discriminator).
3. Read type-specific fields back (needs a second request with an @odata.type URL cast).

So for a routine ask like "add a Rating 1–5 and a 2000-char Comment", the agent dropped to raw urllib, hit the metadata quirks, and looped. This PR gives all three an SDK path, so agents stay on the managed client (which carries auth, retries, and telemetry attribution) and never hand-roll the metadata Web API.

Scope: this is Layer 1 (SDK / product fixes). Layer 2 — fixing the skills (updating dv-metadata to teach the new SDK spec and stop describing the raw Web-API workaround) — is a separate follow-up in the plugin repo and depends on this PR shipping.

What's in this PR

Four changes, all backward-compatible, sync + async parity.

1. Create with constraints (closes #194)

tables.create / add_columns now accept a dict column spec alongside the existing str / Enum:

client.tables.create("cfb_CustomerFeedback", {
    "cfb_Name":    "string",                                   # unchanged
    "cfb_Rating":  {"type": "int",  "min_value": 1, "max_value": 5},
    "cfb_Comment": {"type": "memo", "max_length": 2000, "display_name": "Comment"},
})

Override keys: max_length, min_value, max_value, precision, format, required, display_name.

2. Update an existing column (closes #202)

New tables.update_column / update_columns. The SDK reads the typed attribute, applies the overrides, and issues the PUT with the right @odata.type internally — the caller never sees a verb or a discriminator:

client.tables.update_column("cfb_CustomerFeedback", "cfb_Comment",
                            {"max_length": 4000, "display_name": "Customer Comment"})

3. Typed reads in one request (closes #203)

typed=True on list_columns / get_column, backed by RetrieveMetadataChanges with property projection — type-specific fields come back in a single request, no @odata.type URL cast:

col  = client.tables.get_column("account", "emailaddress1", typed=True)   # col["MaxLength"], etc.
cols = client.tables.list_columns("account", typed=True)

4. Attribution allowlist

Add dv-overview + erp-xpp to OperationContext._ALLOWED_SKILLS (parity with the plugin's scripts/auth.py, which allows all 10). Previously get_client('dv-overview'|'erp-xpp') raised ValueError on the SDK side, and the plugin silently fell back to raw urllib for those two skills.

Compatibility

str / Enum column specs and every existing signature are unchanged; the only new parameter is a keyword-only typed=False on list_columns. No CHANGELOG / version edit (per CONTRIBUTING — maintainers handle those at release).

Testing

Unit — full suite green: 2437 passed (sync + async). New coverage: dict-spec overrides, update_column (PUT verb + @odata.type + override application + error paths), RetrieveMetadataChanges typed reads (request shape, projection, criteria), get_column(typed=True), and allowlist parity.

Live end-to-end against a Dataverse org (*.crm10.dynamics.com) — 12/12 checks pass:

Check	Result
#194 create table w/ bounded int (1–5) + 2000-char renamed memo	PASS
#203 typed read → MinValue=1, MaxValue=5, MaxLength=2000 (one request)	PASS
#203 list_columns(typed=True) returns typed fields	PASS
#202 update_column memo 2000 → 4000, re-read confirms	PASS
core CRUD (create / retrieve / update / list / delete) — no regression	PASS

For contrast, the unmodified SDK raises ValueError: Unsupported column spec type ... <class 'dict'> on the very same create — the exact gap this PR closes.

[Copilot]

Copilot review overview

🟡 Changes recommended

Typed reads silently ignore filters, missing-column errors use the wrong subcode, and async coverage and public documentation are incomplete.

Review effort: Balanced
Findings: 2 · 2

Open (4)

• Typed async path ignores supplied filter · New
• Typed sync path ignores supplied filter · New
• Update batch docstrings for dictionary constraint support · New
• Add README examples for new table APIs and typed mode · New

What changed in this PR

Adds SDK-native column constraints, typed metadata reads, column updates, and operation-context allowlist parity.

Changes:

• Supports constraint dictionaries when creating columns.
• Adds typed column reads and constraint updates for sync/async clients.
• Allows dv-overview and erp-xpp operation contexts.

File	Description
tests/​unit/​test_tables_operations.py	Updates sync delegation expectations.
tests/​unit/​test_operation_context.py	Tests added skill identifiers.
tests/​unit/​data/​test_odata_internal.py	Tests constraints, updates, and typed reads.
tests/​unit/​aio/​test_async_tables.py	Updates async delegation expectations.
src/​PowerPlatform/​Dataverse/​operations/​tables.py	Exposes new sync table APIs.
src/​PowerPlatform/​Dataverse/​data/​_odata.py	Implements sync typed reads and updates.
src/​PowerPlatform/​Dataverse/​data/​_odata_base.py	Adds shared constraint handling.
src/​PowerPlatform/​Dataverse/​core/​config.py	Expands the skill allowlist.
src/​PowerPlatform/​Dataverse/​aio/​operations/​async_tables.py	Exposes async table APIs.
src/​PowerPlatform/​Dataverse/​aio/​data/​_async_odata.py	Implements async typed reads and updates.

---

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.