feat(lineage): add Body Changes view mode to filter config-only changes

[even-wei]

PR checklist

• Ensure you have added or ran the appropriate tests for your PR.
• DCO signed

What type of PR is this?

Feature

What this PR does / why we need it:

Adds a new "Body Changes" mode to the lineage view Mode dropdown. This mode uses state:modified.body, state:modified.macros, and state:modified.contract selectors to filter out config-only YAML changes (tags, descriptions, deprecation dates, warehouse, materialization, etc.) while keeping data-impacting changes visible.

This reduces noise for users whose PRs touch YAML config without changing SQL logic — a pain point raised by customers (see Slack thread in #customer-super).

Which issue(s) this PR fixes:

Resolves DRC-3047

Special notes for your reviewer:

• The existing "Changed Models" (1+state:modified+) and "All" modes remain unchanged
• No defaults are changed — changed_models is still the default for UI and MCP
• The recce summary command is not changed in this PR; it already excludes config-only changes via checksum comparison. A follow-up issue can add full selector parity if needed
• The MCP impact_analysis tool already uses this same selector pattern (state:modified.body+ state:modified.macros+ state:modified.contract+)
• Pre-existing selector test failures (DuckDB config issue) are unrelated

Does this PR introduce a user-facing change?:

Yes — new "Body Changes" option in the lineage view Mode dropdown that filters to only SQL body, macro, and contract changes.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
recce/adapter/base.py	45.58% <ø> (ø)
recce/adapter/dbt_adapter/__init__.py	79.17% <100.00%> (+0.03%)	⬆️
recce/mcp_server.py	87.98% <ø> (ø)
recce/server.py	70.77% <100.00%> (ø)
recce/tasks/core.py	79.72% <ø> (ø)
recce/tasks/lineage.py	100.00% <100.00%> (ø)
recce/tasks/rowcount.py	90.04% <100.00%> (ø)
recce/tasks/schema.py	46.51% <100.00%> (ø)
tests/adapter/dbt_adapter/test_selector.py	100.00% <100.00%> (ø)
tests/tasks/test_lineage.py	100.00% <ø> (ø)
... and 2 more

... and 4 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[even-wei]

Code Review: PR #1268

Reviewer: Claude Code Review (adversarial)
Files reviewed: 21 (16 source/non-test, 5 test)
Categories: API endpoints, Business logic, Frontend, Tests
Passes run: A, B, C, D, E, F, G, H
Note: PR is in Draft status — review proceeded at user request.

Validation Results

Pass A — Correctness & Logic — FAIL

ISSUE — LineageView.tsx does not filter body_changes correctly when used by saved Lineage Diff checks

js/packages/ui/src/components/lineage/LineageView.tsx:204-210:

} else if (
  viewOptions?.view_mode === "changed_models" ||
  viewOptions?.view_mode === "body_changes"
) {
  // Changed models or body changes (filtering done server-side via selector)
  selectedNodeIds = lineageGraph.modifiedSet;
}

lineageGraph.modifiedSet is computed in js/packages/ui/src/contexts/lineage/utils.ts:111-134 as every node with any change_status (body changes and config-only changes). The inline comment "filtering done server-side via selector" is true only for LineageViewOss.tsx:514-538, which actually calls the /api/select endpoint. LineageView.tsx never makes that call — it's a presentation component used by LineageDiffView.tsx (saved Lineage Diff checks).

Consequence: a user who saves a Lineage Diff check while in Body Changes mode will see the same node set as a Changed Models check (config-only YAML edits included), defeating the feature's purpose for that surface. The selector value is preserved on the wire, but the rendered graph isn't.

Either:

• Drop body_changes from this branch and let selectedNodeIds remain undefined, then rely on a server-resolved node list, or
• Add a body-change indicator to each node in MergedLineageResponse and build a separate bodyChangedSet in buildLineageGraph for this branch to use, or
• Confirm that saved checks always carry viewOptions.node_ids (line 201-203 short-circuits the branch). If that's the contract, the comment must say so and the new branch is unnecessary.

NOTE — Empty-state guard missing for body_changes in js/packages/ui/src/components/lineage/LineageViewOss.tsx:1378:

if (viewMode === "changed_models" && !lineageGraph.modifiedSet.length) {
  return <LineageViewNoChanges ... />;
}

For body_changes, this empty-state path is never taken. The narrower selector (state:modified.body|.macros|.contract) can return an empty set on a PR that only edits YAML config — even when modifiedSet is non-empty — so the user gets a blank canvas instead of the friendly "no changes" message. Considering the use case is precisely to triage config-only PRs, this is a UX gap, not a defect.

Pass B — Security — PASS

No new endpoints, no new authn/authz boundaries, no PII surfaces. The selector strings are static (not user-controlled).

Pass C — Cross-Reference Consistency — PASS

The Literal["all", "changed_models", "body_changes"] widening is fanned out coherently across every Python type carrier (recce/adapter/base.py, recce/server.py:SelectNodesInput, recce/tasks/{core,lineage,rowcount,schema}.py) and every TypeScript carrier (api/lineagecheck.ts, api/schemacheck.ts, api/select.ts, api/types/run.ts, LineageView.tsx, LineageDiffView.tsx, SchemaDiffView.tsx). MCP tool schemas in recce/mcp_server.py:241,362 updated in both lineage_diff and schema_diff handlers. LineageViewTopBarOss.tsx correctly left untouched — it's a thin context wrapper around the modified core.

Verified: existing LineageViewTopBar.tsx consumers don't rely on view_mode being a 2-valued union for narrowing.

Pass D — Error Handling — PASS

LineageViewOss.tsx:514-538 already wraps the /api/select call with a try/except that falls back to view_mode: "all". If the underlying dbt selector state:modified.body is unsupported on a customer's dbt version, the user sees the full graph instead of an exception. Acceptable.

Pass E — Test Coverage — PASS with one gap

• tests/adapter/dbt_adapter/test_selector.py:193-195 exercises select_nodes(view_mode="body_changes") end-to-end through the dbt adapter — confirms wiring.
• js/src/components/lineage/__tests__/LineageViewTopBar.test.tsx adds 45 lines of tests (verified locally: 56 tests passed) covering: label rendering for body_changes, dropdown radio option, click-to-select dispatching view_mode: "body_changes".
• tests/tasks/{test_lineage,test_row_count,test_schema}.py validators pass body_changes through the schema validator.

Gap: No test exercises the case the feature is for — a project where some models have body changes and others have config-only changes — to assert that select_nodes(view_mode="body_changes") returns the body-changed subset and excludes the config-only subset. The existing test (test_select_with_pacakage_mode_include_exclude) only happens to compare equal because the fixture's only modification is a body change ("same result here since the change is a body change"). A negative case (config-only model excluded) would catch any future regression in dbt's state:modified.* semantics.

Pass F — Diff-Specific Checks — PASS

• The if view_mode and view_mode == "changed_models": → if view_mode == "changed_models": cleanup is safe (None == "changed_models" is False in Python).
• All callers of the widened literals already accept Optional[str] at the JSON boundary; no breaking change to saved check files.
• MCP tool descriptions updated to advertise the new option.

Pass G — Performance — PASS

The selector compiles once per select_nodes call. Three additional parse_difference calls in the new branch are negligible compared with the existing state:modified+ graph compilation.

Pass H — Async/Concurrency — N/A

No async surface change.

Verification Results

Check	Result
make flake8	clean
uv run pytest tests/adapter/dbt_adapter/test_selector.py tests/tasks/test_lineage.py tests/tasks/test_row_count.py tests/tasks/test_schema.py	25 passed
pnpm vitest run src/components/lineage/__tests__/LineageViewTopBar.test.tsx	56 passed
pnpm lint	clean
pnpm type:check	3 errors in packages/ui/src/utils/csv/excel.{ts,test.ts} — pre-existing on main, not introduced by this PR
pnpm test (full)	3688 passed, 17 failed — all 17 failures are in pre-existing excel.test.ts, unrelated to this PR

Verdict: NO-GO

Issues

1. LineageView.tsx:204-210 — body_changes not actually filtered when rendered through LineageDiffView. Saved Lineage Diff checks created in Body Changes mode will display the same superset of nodes as Changed Models. Either special-case the branch to a body-change subset, or remove body_changes from the branch and rely on viewOptions.node_ids (clarify which contract is intended).

Notes

1. LineageViewOss.tsx:1378 — extend the empty-state guard to viewMode === "body_changes" so a config-only PR shows the friendly "no changes" placeholder instead of a blank canvas.
2. Test gap in tests/adapter/dbt_adapter/test_selector.py — existing body_changes assertion is degenerate (same result as changed_models because the fixture's only mod is a body change). Add a fixture variant with a config-only mod and assert it is excluded by body_changes.

What I could not verify

• Whether the saved Lineage Diff check's stored viewOptions typically includes node_ids — that would short-circuit the branch in LineageView.tsx:201-203 and make the Issue moot. Confirming this requires reading addLineageDiffCheck plumbing or a manual save-and-replay test in a real environment.
• That state:modified.body / .macros / .contract are universally available in the minimum-supported dbt version — I trust the existing fallback at LineageViewOss.tsx:525-538 covers this, but didn't enumerate the dbt-version test matrix.

What I looked for and did not find

• SQL/command/template injection: none (selector strings are static literals).
• Schema drift between Python Literal and TypeScript union: clean.
• Stale callers of select_nodes / SelectNodesInput / LineageDiffParams not updated: all paths consistent.
• Mock drift in LineageViewTopBar.test.tsx: passes locally; mocks correctly target @datarecce/ui/contexts and @datarecce/ui/components/run, and use the real LineageViewTopBarOss (which delegates to the modified LineageViewTopBar).
• N+1 / unbounded query introduced by selector composition: none.

---

PR is in draft. Mark ready-for-review only after addressing the LineageView filtering branch (or documenting why node_ids always short-circuits it).