Shuowei Li

This commit delivers a **bug fix** to the **SQL generation logic** within the `bigframes` `ibis` backend. It primarily **prevents invalid SQL generation** by modifying the `visit_Aggregate` method to insert a `SELECT 1` placeholder when aggregate operations are performed on empty selections, ensuring robust query construction. Additionally, it **enhances BigQuery's `visit_Cast` method** to correctly handle casting to NULL types and specific scenarios involving NULL literals for struct and array types. A new system test has been added to verify the proper generation of the `SELECT 1` fallback.

This commit provides a **fix** by performing **documentation maintenance** and improving **doctest stability** across several **BigFrames modules**. It specifically addresses expected output changes, likely stemming from an underlying fix for BigQuery log suppression, by updating doctest examples in `packages/bigframes/bigframes/operations/ai.py`, `packages/bigframes/bigframes/session/__init__.py`, and `packages/bigframes/third_party/bigframes_vendored/pandas/io/gbq.py`. These adjustments, which include adding print statements and ellipsis, are crucial for preventing brittle tests and ensuring the reliability and accuracy of **documentation examples**, especially within the `gbq.py` module.

This commit delivers a **bug fix** for the **BigFrames data manipulation core**, resolving crashes that occurred when performing **`melt` operations** on **empty DataFrames** or those with empty MultiIndex columns. The `melt` implementation in `bigframes/core/blocks.py` was updated to correctly retrieve value labels and include explicit handling for empty input IDs in `unpivot` and empty pandas indexes in `_pd_index_to_array_value`. This ensures the robustness of `melt` when processing sparse or empty datasets, preventing unexpected application failures. New system tests were added to verify that `melt` correctly handles empty MultiIndex columns and preserves their structure, enhancing the reliability of **BigFrames** for complex data transformations.

This commit introduces a **new feature** to enable **full round-trip persistence for multimodal reference columns** in BigFrames. It specifically allows **saving multimodal metadata descriptions** when writing data to BigQuery via the `.to_gbq()` method. The **BigQuery data type conversion logic** in `bigframes/dtypes.py` is enhanced to correctly handle object reference types and their associated description tags, and the `BigQueryCachingExecutor` is updated to process these columns during schema updates. This ensures that complex multimodal data structures and their metadata are accurately preserved when persisted to and retrieved from BigQuery.

This commit performs **maintenance** on the **BigFrames ML system tests** to improve their resilience against evolving BigQuery ML service outputs. It **updates assertions** in tests for **ARIMA_PLUS forecasting** and **LLM (Gemini) scoring** functionalities, which were failing due to new columns being appended to `ML.EVALUATE` results. Specifically, `pd.testing.assert_frame_equal` calls are modified to compare only the expected columns, such as `mean_absolute_error` and `mean_squared_error`. This change ensures that core metrics are still verified while making the test suite robust to future backward-compatible schema additions from BigQuery ML, preventing unnecessary test failures.

This commit **fixes a bug** in the `bigframes.pandas.DataFrame.describe()` method that caused errors when attempting to compute summary statistics on unsupported BigQuery complex types like `OBJ_REF_DTYPE` and `JSON` columns. It **modifies the `describe` method** and its internal aggregation logic to restrict calculations for these specific types to only a basic `count()`, skipping unhashable or mathematically incompatible metrics. This **enhancement** improves the robustness of the **`bigframes.pandas` API**, allowing `describe()` to execute successfully on DataFrames containing these complex types and also correctly handle empty column sets. New test cases have been added to validate the proper behavior for these previously problematic scenarios.

This commit introduces a **new capability** by adding the `bpd.options.display.render_mode` configuration option, providing a clearer and more flexible way to control how **BigFrames DataFrame and Series objects** are visualized in interactive environments. This new setting, which supports `html`, `plaintext`, and `anywidget` modes, refactors the previous `repr_mode` concept and impacts the core **rendering logic** within `packages/bigframes/bigframes/display/html.py`. Backward compatibility for `display.repr_mode = "anywidget"` is maintained, and the change is thoroughly validated with **new unit tests** in `packages/bigframes/tests/unit/display/test_render_mode.py` and updates to existing notebooks and system tests. This enhancement improves the **user experience** by offering precise control over output formats, making BigFrames more adaptable to various interactive and text-only environments.

This commit **updates the BigFrames multimodal notebook example** (`multimodal_dataframe.ipynb`) to demonstrate advanced image modifications. It **refactors** the image transformation logic within the example to utilize **custom BigQuery Python UDFs** that leverage the `opencv` library. Specifically, the notebook now showcases functions like `image_blur` and `apply_transformation` for image processing. This **documentation and example update** also **removes the deprecated `display_blob` function** and refreshes the Gemini prediction section, providing a more current and robust illustration of image processing within BigQuery using Python UDFs.

This commit **refactors** the **BigQuery DataFrames multimodal capabilities documentation** by updating the `multimodal_dataframe.ipynb` notebook. It replaces reliance on an **internal `.blob` accessor** with **public APIs** and new helper functions such as `get_metadata`, `get_content_type`, and `display_blob`. This **documentation update** ensures that examples provided to users utilize **stable and supported interfaces** of the BigQuery DataFrames library. The change improves the long-term maintainability and accuracy of the multimodal data handling examples, guiding users towards recommended practices.

This commit introduces **new functionality** by adding custom **BigQuery Python UDFs** for **PDF text extraction** and **PDF chunking**. These UDFs, named `pdf_extract` and `pdf_chunk`, leverage the `pypdf` library to enable direct API usage for processing PDF documents within BigQuery. The `multimodal_dataframe.ipynb` notebook in `packages/bigframes/notebooks/multimodal/` has been updated to demonstrate the practical application of these new capabilities. This **feature enhancement** provides users with clear examples and direct methods for integrating advanced PDF content analysis into their multimodal data workflows.

This commit **updates the multimodal notebook** (`multimodal_dataframe.ipynb`) to leverage **public BigFrames APIs** for retrieving runtime JSON strings. It refactors the notebook's implementation by replacing internal operations with stable methods like `bigframes.bigquery.obj` and `bigframes.bigquery.to_json_string`. This **documentation update** also introduces a new helper function, `get_runtime_json_str`, enhancing the example's robustness and maintainability. The change ensures the **multimodal examples** within the `bigframes` package demonstrate best practices by relying on supported public interfaces.

This commit **updates the documentation examples** within the `multimodal_dataframe.ipynb` notebook. It specifically modifies the **audio transcription examples** to directly utilize the `bigframes.bigquery.ai.generate` API. This **documentation update** moves away from the previously used `blob.audio_transcribe` convenience function, providing users with a more explicit demonstration of the underlying API for multimodal operations. The change ensures the **notebook's examples reflect current best practices** for interacting with the BigQuery AI capabilities, affecting the **BigFrames multimodal features** documentation.

This commit **enhances the `multimodal_dataframe.ipynb` notebook** by adding a new section demonstrating **EXIF metadata extraction** from images. It introduces a practical example of implementing a **custom BigQuery Python UDF** (`extract_exif`) that leverages external libraries like `pillow` and `requests` to parse EXIF tags from image URLs. This **new feature** provides users with a clear pattern for handling image metadata and integrating custom Python logic efficiently within their **BigFrames workflows** for BigQuery DataFrames.

This commit initiates the **release process for `bigframes` version `2.35.0`**, updating the version metadata across the `bigframes` package and its vendored components. This **maintenance** task incorporates a significant set of **new features** and **bug fixes** into the library. Notable additions include `bigframes.pandas.col` with operators, `bigquery.load_data`, `bigquery.ai.generate_text`, `generate_embedding`, and an IPython cell magic, alongside improvements to progress output and Anywidget integration. Merging this pull request will automatically trigger the official `bigframes` `2.35.0` release, making these enhancements available to users.

This commit **fixes** excessive console noise and significantly improves the **user experience** when using the **`anywidget` display mode** by suppressing various warnings and cleaning up progress output. Specifically, it suppresses `JSONDtypeWarning` and `FutureWarning` within `bigframes/display/anywidget.py` and `bigframes/display/html.py` during display bundle generation, and also removes extraneous progress bar output. Further **maintenance** includes adding general `FutureWarning` filters in `bigframes/__init__.py` for Google Cloud libraries and refining warning handling in `bigframes/pandas/io/api.py`. This **enhancement** ensures a much cleaner and less cluttered console for interactive users of the `anywidget` display.

This commit introduces a **new capability** to the **Anywidget display mode** by automatically disabling progress bars and job logging during interactive operations. Specifically, it wraps methods such as `get_anywidget_bundle`, `_initial_load`, and `_set_table_html` within `packages/bigframes/bigframes/display/anywidget.py` using `option_context("display.progress_bar", None)`. This **improves user experience** by eliminating visual clutter from progress bars during initial widget loading, pagination, and sorting. The change ensures a **cleaner and seamless interactive data exploration experience** within notebook environments.

fix: exlcude gcsfs 2026.2.0 (#2445)

This commit **enhances the user experience** in **BigQuery DataFrames' interactive Anywidget display mode** by **disabling progress bars** and job logging. This **feature improvement** prevents visual clutter during initial widget loading and subsequent interactions like pagination and sorting within the `TableWidget`. By wrapping calls such as `repr_mimebundle`, `_initial_load`, and `_set_table_html` with a display option context, the change ensures a **cleaner and more seamless notebook interface** for interactive data exploration.

This commit **temporarily disables all system tests related to blob functionality** within the `bigframes` package. It's a **maintenance** change that adds a module-level `pytest.mark.skip` marker to several test files, including `test_function.py`, `test_io.py`, `test_properties.py`, and `test_urls.py`. This ensures that these specific **blob tests** are skipped during test execution, likely to address a known, temporary issue or instability. The scope of this change is limited to preventing these tests from running without altering their underlying code.

This commit is a **release preparation** chore for the **`bigframes` library**, advancing its version to `2.34.0`. It updates the version metadata in `packages/bigframes/bigframes/version.py` and `packages/bigframes/third_party/bigframes_vendored/version.py`, along with the Librarian state. This release introduces **new features** such as `bigquery.ml.generate_embedding`, `bigquery.ml.generate_text`, and `bigquery.create_external_table` methods, alongside a new `sql_compiler` option and deprecation warnings for `.blob` accessor. Additionally, it includes a **bug fix** addressing broken job URLs, providing significant enhancements and stability improvements for `bigframes` users.