Python: fix CosmosHistoryProvider saves code interpreter tool calls chunk by chunk (#5793)

[hanhan761]

Summary

Fix #5793: During streaming, code interpreter tool calls are received as multiple content items with the same \call_id\ but different \sequence_number\ values. Before persisting to Cosmos DB, consecutive \code_interpreter_tool_call\ chunks with matching \call_id\ are now merged into a single \Content\ item with the complete, aggregated text.

Changes

\python/packages/azure-cosmos/agent_framework_azure_cosmos/_history_provider.py\

• Added _merge_code_interpreter_chunks()\ helper function to merge a list of \code_interpreter_tool_call\ chunks into one \Content\ item, concatenating the text inputs and merging \�dditional_properties.
• Added \CosmosHistoryProvider._aggregate_code_interpreter_calls()\ static method that scans \contents\ for consecutive \code_interpreter_tool_call\ items with the same \call_id\ and merges them.
• Updated \save_messages()\ to run contents through the aggregation before serializing to Cosmos DB.

\python/packages/azure-cosmos/tests/test_cosmos_history_provider.py\

Added \TestCodeInterpreterAggregation\ class with tests:

• \ est_merge_code_interpreter_chunks: verifies chunk text concatenation
• \ est_aggregate_code_interpreter_calls_merges_consecutive_chunks: verifies full aggregation pipeline with mixed content types
• \ est_aggregate_preserves_independent_tool_calls: verifies different \call_id\ values are not merged
• \ est_save_messages_calls_aggregation: verifies \save_messages\ calls aggregation and saves aggregated content

Issue

Fixes #5793

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

Adds aggregation logic to CosmosHistoryProvider so that streamed code_interpreter_tool_call chunks (which arrive as multiple Content items sharing the same call_id) are merged into a single content item before being persisted to Cosmos DB.

Changes:

• Introduces _merge_code_interpreter_chunks helper and CosmosHistoryProvider._aggregate_code_interpreter_calls static method.
• Invokes aggregation inside save_messages prior to serializing each message.
• Adds unit tests covering merging, non-merging of distinct call_ids, and end-to-end behavior through save_messages.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 4 comments.

File	Description
python/packages/azure-cosmos/agent_framework_azure_cosmos/_history_provider.py	Implements chunk merging helper and aggregation hook in save_messages.
python/packages/azure-cosmos/tests/test_cosmos_history_provider.py	Adds TestCodeInterpreterAggregation test class verifying merge logic.

[hanhan761]

All 4 review comments have been addressed in commit 2ae1e66:

1. save_messages no longer mutates caller's Message — now builds the message dict directly from \message.to_dict()\ with aggregated contents, leaving the original \Message\ untouched.
2. Non-text inputs are preserved — _merge_code_interpreter_chunks\ now collects non-text inputs (e.g. images) and appends them to \merged_inputs\ alongside the concatenated text.
3. \sequence_number\ is dropped from merged properties — the per-chunk \sequence_number\ key is explicitly skipped when merging \�dditional_properties, since the aggregated item no longer represents a single chunk.
4. Tests converted to async style — imports moved to module level, @pytest.mark.asyncio\ used instead of \�syncio.run, added new test cases for immutability, non-text inputs, and sequence_number behavior.