fix: raise clear error when all records are dropped during generation

[nabinchha]

Summary

• When every record fails during generation (e.g. LLM or image model errors), the empty dataset previously reached the profiler, which raised a misleading DatasetProfilerConfigurationError about missing columns.
• Both preview() and create() now check for an empty dataset before profiling and raise a DataDesignerGenerationError with an actionable message instead.
• Added tests for both code paths to verify the correct error is raised.

Closes #382

Test plan

• test_create_raises_generation_error_when_dataset_is_empty — verifies create() raises DataDesignerGenerationError when all records are dropped
• test_preview_raises_generation_error_when_dataset_is_empty — verifies preview() raises DataDesignerGenerationError when all records are dropped
• Existing tests pass (make test)

closes #382

[greptile-apps]

Greptile Summary

This PR fixes a poor user experience where all records being dropped during generation (due to LLM or image model errors) caused an empty dataset to propagate to the profiler, which then raised a misleading DatasetProfilerConfigurationError about missing columns. The fix adds explicit empty-dataset checks in both create() and preview(), surfacing a clear DataDesignerGenerationError instead.

Key changes:

• create(): A new try/except block wraps load_dataset_with_dropped_columns() to catch storage-level exceptions (e.g. no parquet files written) as DataDesignerGenerationError; a subsequent len == 0 guard catches the defensive case of an empty-but-loadable dataset.
• preview(): An early len(processed_dataset) == 0 guard after process_preview() raises DataDesignerGenerationError before reaching the profiler; the message correctly attributes the cause to "generation or processing failures" to cover both row-filter processors and LLM failures.
• The now-redundant len(processed_dataset) > 0 condition is removed from the success-log check in preview(), since that point is only reachable with a non-empty dataset.
• Three new tests cover the create() load-failure, create() empty-DataFrame, and preview() empty-DataFrame paths.

Confidence Score: 4/5

• This PR is safe to merge; it improves error clarity without changing any happy-path logic.
• The changes are focused and targeted — they add early-exit guards without touching the generation or profiling pipelines. The except Exception wrapper for load_dataset_with_dropped_columns() is intentionally broad (to catch any storage-level failure) and follows the same pattern used elsewhere in the file. The three new tests correctly exercise the added code paths. The stub_sampler_only_config_builder fixture does not invoke LLM calls, so the tests that run the real build pipeline with num_records=1 are stable. No regressions are expected in the existing test suite.
• No files require special attention.

Important Files Changed

Filename	Overview
packages/data-designer/src/data_designer/interface/data_designer.py	Adds early empty-dataset guards in both create() and preview() with dedicated try/except for load_dataset_with_dropped_columns(); removes now-redundant len(processed_dataset) > 0 guard from the success-log condition.
packages/data-designer/tests/interface/test_data_designer.py	Adds three new error-path tests: empty DataFrame from load_dataset_with_dropped_columns, FileNotFoundError from same, and empty DataFrame from process_preview; tests for create() run the real sampler build pipeline before hitting the patch.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[create / preview called] --> B[builder.build / build_preview + process_preview]
    B -- Exception --> C[raise DataDesignerGenerationError\n'Error generating dataset']
    B -- Success --> D{create or preview?}

    D -- create --> E[load_dataset_with_dropped_columns]
    E -- Exception --> F[raise DataDesignerGenerationError\n'Failed to load generated dataset']
    E -- Empty DataFrame --> G[raise DataDesignerGenerationError\n'Dataset is empty']
    E -- Non-empty DataFrame --> H[profiler.profile_dataset]

    D -- preview --> I{len processed_dataset == 0?}
    I -- Yes --> J[raise DataDesignerGenerationError\n'Dataset is empty']
    I -- No --> H

    H -- Exception --> K[raise DataDesignerProfilingError]
    H -- Success --> L[Return Results]

Loading

_{Last reviewed commit: e686dbb}

[andreatgretel]

packages/data-designer/src/data_designer/interface/data_designer.py:298

if (
    len(processed_dataset) > 0
    and isinstance(analysis, DatasetProfilerResults)
    and len(analysis.column_statistics) > 0
):

nit: I think len(processed_dataset) > 0 is always true here now that the guard at line 275 raises on empty datasets. Might be nice to simplify the condition?

[andreatgretel]

Just a couple of nits, nothing blocking. Thanks for addressing the create() case too!

[nabinchha]

Just a couple of nits, nothing blocking. Thanks for addressing the create() case too!

cleaned up in e686dbb