Wanlin Du

This commit **introduces new capabilities** for URI path templating and enhances querystring serialization within the `google.genai` library's interaction layer. A new utility module, `google/genai/_interactions/_utils/_path.py`, provides the `path_template` function for robust URI construction, including proper quoting of path segments and query parts. Concurrently, the `google/genai/_interactions/_qs.py` module receives an **enhancement to querystring serialization**, specifically adding support for the 'indices' array format within the `_stringify_item` method. These changes collectively improve the library's ability to handle complex URI structures and query parameters.

This commit **refactors** the `model_dump` utility within the **`google.genai._interactions._compat`** module to enhance **Pydantic version compatibility**. It introduces a `TypedDict` for `model_dump` keyword arguments and modifies the `model_dump` function to conditionally pass the `by_alias` argument based on the detected Pydantic version. This **internal compatibility improvement** ensures correct serialization behavior for Pydantic models across different environments, preventing potential issues related to argument handling. This **maintenance** work enhances the robustness of data interactions within the GenAI library.

This commit **refactors** the type definition for `DynamicAgentConfigParam` within the **`google.genai._interactions.types` module** of the Generative AI library. It simplifies the declaration by directly using `TypedDict` with `extra_items=object`, eliminating the previous `TypeAlias` and `Union` with `Dict`. This is an **internal maintenance** change aimed at improving the clarity and robustness of the type system. The **refactoring** has **no functional impact** on the library's runtime behavior or external API, primarily affecting how the type is declared and understood by type checkers.

This commit introduces a **new capability** to enable **server-side Managed Compute Platform (MCP)** by adding an `mcp_servers` field to the `Tool` class within the `google.genai` library. This feature ensures that when server-side MCP is configured for a tool, **automatic function calling (AFC) is explicitly disabled** for that tool, preventing conflicting execution mechanisms. The change involves updating internal tool conversion logic (`_Tool_to_mldev`, `_Tool_to_vertex`), enhancing the detection of AFC-incompatible tools in `find_afc_incompatible_tool_indexes`, and introducing new types like `McpServer`. This significantly impacts how tools are defined and processed, providing more control over tool execution environments, with comprehensive test coverage for chat and `generate_content` interactions.

This commit provides a **documentation update** for the `background` parameter across the **`google.genai._interactions`** module. It clarifies that this parameter is **input-only** by updating its docstring in various interaction methods within `google/genai/_interactions/resources/interactions.py` and in parameter definition classes like `BaseCreateModelInteractionParams` and `BaseCreateAgentInteractionParams` in `google/genai/_interactions/types/interaction_create_params.py`. This **maintenance** change improves the accuracy and clarity of the API documentation, guiding developers on the intended usage of the `background` parameter.

This commit **regenerates the project documentation** for version 1.59.0, incorporating new API details. It specifically **adds documentation** for `aiohttpClient` within `HttpOptions` and introduces new definitions and fields for `VoiceActivity` and `VoiceActivityType`. This **documentation update** ensures that users have access to the latest information regarding new HTTP client configurations and voice activity features. The changes span `genai.html`, `genindex.html`, `index.html`, and `modules.html`, providing comprehensive coverage of these new elements.

This commit **regenerates the project documentation** for version 1.58.0, primarily focusing on the **`genai` module**. It **adds comprehensive documentation** for the new `AsyncGeminiNextGenAPIClientAdapter` and `GeminiNextGenAPIClientAdapter` classes and their methods. This **documentation update** ensures that the `genindex.html`, `index.html`, and `modules.html` files are correctly updated to reflect and link to these new API client adapters. The change provides users with up-to-date information on the latest Gemini API integrations, improving discoverability and understanding of these new client functionalities.

This commit performs a **maintenance update** to the `google.genai` library's **streaming function call tests**. It modifies `google/genai/tests/models/test_function_call_streaming.py` to utilize the `gemini-3-pro-preview` model for validating **function calling examples**. Additionally, it includes a **refactoring** of the internal test message structures to align with the updated model usage. This ensures the **test suite** remains current and accurately validates the streaming function call functionality against the latest preview model, improving the relevance and reliability of the tests.

This commit **adds comprehensive tests** for the **streaming function call feature** within the `google.genai` library. It specifically validates the behavior when `stream_function_call_arguments` is enabled, covering diverse parameter types, conversational history, and asynchronous execution. This **maintenance chore** significantly improves the reliability and robustness of the **streaming function call functionality**, ensuring its correct operation across various language implementations.

This commit introduces a **new capability** to **support streaming of function call arguments** across the `google.genai` library. It defines new types like `PartialArg` and `stream_function_call_arguments` in `types.py` and updates various converters in `_live_converters.py`, `_tokens_converters.py`, `batches.py`, `caches.py`, and `models.py` to properly handle and serialize these streaming arguments. This enables the feature for **model generation**, **batch operations**, and **caching**, allowing models to stream partial function call arguments as they are generated. Additionally, a new utility function, `raise_error_for_afc_incompatible_config`, was added to validate configurations for automatic function calling when streaming is involved. This enhancement significantly improves the flexibility and responsiveness of function calling interactions with generative AI models.

This commit **improves test reliability** by ensuring that default base URLs are reset to `None` after each test case in `google/genai/tests/client/test_client_initialization.py`. This **maintenance** change prevents global state leakage between tests, which could lead to flakiness or incorrect test results for the **`google.genai.client` module's initialization logic**. The cleanup ensures that subsequent tests start with a clean slate, making the **testing infrastructure** more robust and predictable.

This commit **fixes a test** within the **`google.genai` client testing suite** by ensuring that the `GEMINI_API_KEY` environment variable is explicitly deleted when building a test client designed to operate without environment variables. This **maintenance fix** prevents unintended API key detection during tests, thereby improving the reliability and accuracy of **client configuration tests** in `test_client_requests.py`. The change specifically affects the `build_test_client_no_env_vars` helper, guaranteeing proper isolation for tests verifying client initialization logic.

This commit introduces a **fix** to the **`google.genai` library** by ensuring **Automatic Function Calling (AFC)** is correctly handled when incompatible tools are provided. It adds a new utility function, `find_afc_incompatible_tool_indexes`, to identify such tools within `google/genai/_extra_utils.py`. The core **content generation methods** like `generate_content` and `generate_content_stream` in `google/genai/models.py` are updated to utilize this utility, **disabling AFC with a warning** if incompatible tools are detected. This change improves the robustness of the **`genai` library's tool-use capabilities**, preventing unexpected behavior for users attempting to leverage AFC with unsupported tool configurations. New unit tests were also added to validate this behavior.

This commit **refactors** the `google.genai` library by introducing a new internal field, `_async_httpx_client_args`, within the `BaseApiClient` class in `google/genai/_api_client.py`. This addition allows for the storage of specific arguments intended for the underlying asynchronous `httpx` client, enabling more granular and flexible configuration of HTTP requests. The change primarily affects the **internal API client configuration**, providing a mechanism for future enhancements or custom client behavior. This work is a **refactoring** effort to improve the configurability of the **API client subsystem**. Additionally, a minor **chore** updates the **version number** in `google/genai/version.py`.

This commit introduces a **new capability** to the **Google GenAI client library**, enabling users to specify **custom HTTP headers** for file upload requests. It modifies the **file upload mechanisms** within `google/genai/_api_client.py` by adding a helper function, `_update_headers_with_append_keys`, to intelligently merge user-provided headers while preserving critical `user-agent` and `x-goog-api-client` information. The **`Files.upload` and `AsyncFiles.async_upload` methods** in `google/genai/files.py` are updated to accept these custom headers, providing greater flexibility for advanced use cases like custom authentication or tracing. This enhancement is thoroughly validated with new unit tests covering both the header merging logic and the end-to-end upload process.

This commit provides a **bug fix** by **enabling the `id` field** for `FunctionCall` objects within the **Google Generative AI SDK's Vertex AI integration**. Previously, attempts to convert `FunctionCall` objects containing an `id` field to or from the Vertex AI format would raise a `ValueError`. This change modifies the conversion logic in `_live_converters.py`, `caches.py`, and `models.py` by removing the restriction and adding proper assignment within functions like `_FunctionCall_to_vertex` and `_FunctionCall_from_vertex`. As a result, `FunctionCall` objects can now correctly retain and transmit their `id` when interacting with the **Vertex AI platform**, improving data fidelity for tool calls.

This commit introduces **new converter functions** for the `FunctionCall` class, enabling robust serialization and deserialization of function call objects. These **new capabilities** are integrated across core `google.genai` modules, including `_live_converters.py`, `_tokens_converters.py`, `batches.py`, `caches.py`, and `models.py`. Specifically, existing `_Part` converters are updated to utilize these new `FunctionCall` converters for both MLDev and Vertex AI formats. This ensures consistent and correct handling of `FunctionCall` data throughout the library, improving interoperability with various Google AI platforms and internal `genai` features.

This commit introduces a **new feature** to the **`google.genai` library**, enabling users to retrieve the **full raw HTTP response body** from `generate_content` and `async_generate_content` calls. This capability is activated by setting a new `should_return_http_response` field in `GenerateContentConfig` within the **`types` module**. To support this, the **`_replay_api_client`** receives a **fix** to correctly parse and re-serialize raw HTTP responses for comparison, ensuring robust testing. This enhancement provides greater transparency and control for developers needing access to low-level HTTP details, with new tests verifying the functionality.

This commit introduces a **new capability** by adding the **response body** for the `generateContent` API. This enhancement ensures that the full output from the **content generation service** is now properly structured and returned, significantly improving the utility for clients consuming the generated data. It directly impacts the **AI model interaction** module, providing a complete and usable response. This allows downstream systems to properly process and utilize the generated content, making the `generateContent` function more robust and functional.

This commit **introduces a new capability** allowing users to retrieve the full HTTP response, including the raw body, when interacting with the **`generate_content`** and file creation APIs. A new `should_return_http_response` option has been added to the **`GenerateContentConfig`** in `google.genai.types` and implemented in the **`google.genai.models`** module's `generate_content` and `async_generate_content` methods. Similarly, the `create` and `async_create` methods in **`google.genai.files`** now also support returning the complete HTTP response. This **feature enhancement** provides greater transparency and control, enabling developers to inspect the full network response for advanced debugging or custom handling, with internal updates to `_replay_api_client` ensuring correct response verification.