fix(config): round-trip processors and profilers

[johnnygreco]

📋 Summary

DataDesignerConfigBuilder.from_config() did not restore processors or profilers from an existing DataDesignerConfig, so round-tripping through a builder silently dropped those settings. This PR restores those fields and adds a regression test for the behavior.

🔗 Related Issue

N/A

🔄 Changes

• Restore processor configs when building a DataDesignerConfigBuilder from an existing config.
• Restore profiler configs when building a DataDesignerConfigBuilder from an existing config.
• Add a regression test covering processor and profiler round-trip behavior.

🧪 Testing

• make test passes (not run; focused test suite below)
• Unit tests added/updated
• E2E tests added/updated (N/A)

Focused checks run:

• uv run pytest packages/data-designer-config/tests/config/test_config_builder.py -q
• uv run ruff check packages/data-designer-config/src/data_designer/config/config_builder.py packages/data-designer-config/tests/config/test_config_builder.py
• uv run ruff format --check packages/data-designer-config/src/data_designer/config/config_builder.py packages/data-designer-config/tests/config/test_config_builder.py
• git diff --check

✅ Checklist

• Follows commit message conventions
• Commits are signed off (DCO)
• Architecture docs updated (N/A)

[greptile-apps]

Greptile Summary

This PR fixes a silent data-loss bug in DataDesignerConfigBuilder.from_config() where processors and profilers present in an existing DataDesignerConfig were never restored, causing a round-trip through the builder to drop those settings entirely.

• Adds two loops in from_config to replay add_processor and add_profiler calls onto the fresh builder, correctly inheriting the upsert semantics and DropColumnsProcessor side-effects that already exist on those methods.
• Adds three regression tests: basic processor/profiler restoration, DropColumnsProcessorConfig glob-pattern side-effects (drop=True on matched columns), and a full field-coverage round-trip asserting every DataDesignerConfig field survives the builder cycle.

Confidence Score: 5/5

The change is a minimal, targeted fix — two loops added to an existing restoration method — with no side-effects on paths that did not previously touch processors or profilers.

The fix is small and correctly delegates to existing, already-tested add_processor and add_profiler methods, which handle upsert semantics and drop-column side-effects. The new tests exercise the exact scenarios described in the PR and cover the full field round-trip as a regression guard.

No files require special attention.

Important Files Changed

Filename	Overview
packages/data-designer-config/src/data_designer/config/config_builder.py	Adds processor and profiler restoration loops to from_config; fix is minimal and correct against a fresh builder instance
packages/data-designer-config/tests/config/test_config_builder.py	Adds three targeted regression tests: basic round-trip, DropColumnsProcessor side-effects, and full field coverage for DataDesignerConfig

Sequence Diagram

sequenceDiagram
    participant C as Caller
    participant F as from_config()
    participant B as DataDesignerConfigBuilder

    C->>F: BuilderConfig (with processors/profilers)
    F->>B: cls(model_configs, tool_configs)
    loop columns
        F->>B: add_column(col)
    end
    loop constraints
        F->>B: add_constraint(constraint)
    end
    opt seed_config present
        F->>B: with_seed_dataset(source, strategy)
    end
    loop processors (NEW)
        F->>B: add_processor(processor)
        Note over B: upsert by name, applies DROP side-effects
    end
    loop profilers (NEW)
        F->>B: add_profiler(profiler)
        Note over B: appends to _profilers list
    end
    F-->>C: populated builder

Loading

_{Reviews (3): Last reviewed commit: "Merge branch 'main' into johnny/fix-conf..." | Re-trigger Greptile}

[andreatgretel]

LGTM, just noting a couple of things that review flagged:

• The new test skips DropColumnsProcessor, which is the one processor type with side effects on column state. Could perhaps add a case with a glob like col_* that asserts the per-column drop flags survive the round-trip.

• Not for this PR, but from_config enumerates DataDesignerConfig fields by hand, which is what caused this bug. Might be worth a follow-up that asserts every field round-trips so the next field addition fails loudly. Claude Code flagged this one.

Smoke-tested locally with DropColumnsProcessor (incl. globs), multi-processor ordering, and YAML file round-trip - all good.

[johnnygreco]

@andreatgretel thanks for the notes. Addressed both in 28e4c293:

• Added a DropColumnsProcessorConfig round-trip case with a glob (col_*) and assertions that the restored builder preserves the expected per-column drop flags.
• Added broader DataDesignerConfig field round-trip coverage. The test derives the field set from DataDesignerConfig.model_fields, populates each current field, round-trips through from_config, and compares every restored field value so future field additions fail loudly if they are not carried through.

Focused checks are still passing locally:

• uv run pytest packages/data-designer-config/tests/config/test_config_builder.py -q
• uv run ruff check packages/data-designer-config/tests/config/test_config_builder.py
• uv run ruff format --check packages/data-designer-config/tests/config/test_config_builder.py
• git diff --check