fix(pydantic-ai): Use first-class hooks when available#5947
fix(pydantic-ai): Use first-class hooks when available#5947alexander-alderman-webb wants to merge 34 commits intomasterfrom
Conversation
…lying on SDK internals
…on the request object
Semver Impact of This PR🟢 Patch (bug fixes) 📋 Changelog PreviewThis is how your changes will appear in the changelog. New Features ✨
Bug Fixes 🐛Anthropic
Other
Internal Changes 🔧
🤖 This preview updates automatically when you update the PR. |
Codecov Results 📊✅ 13 passed | Total: 13 | Pass Rate: 100% | Execution Time: 11.23s All tests are passing successfully. ❌ Patch coverage is 4.23%. Project has 14925 uncovered lines. Files with missing lines (2)
Generated by Codecov Action |
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
Autofix Details
Bugbot Autofix prepared a fix for the issue found in the latest run.
- ✅ Fixed: Streaming path sets input messages before model execution
- Moved
_set_input_messagesin the streaming wrapper to execute after the original stream context finishes so instruction mutations done during execution are captured.
- Moved
Or push these changes by commenting:
@cursor push 6153a56a6a
Preview (6153a56a6a)
diff --git a/sentry_sdk/integrations/pydantic_ai/patches/graph_nodes.py b/sentry_sdk/integrations/pydantic_ai/patches/graph_nodes.py
--- a/sentry_sdk/integrations/pydantic_ai/patches/graph_nodes.py
+++ b/sentry_sdk/integrations/pydantic_ai/patches/graph_nodes.py
@@ -91,13 +91,13 @@
# Create chat span for streaming request
with ai_client_span(None, model, model_settings) as span:
- if messages:
- _set_input_messages(span, messages)
-
# Call the original stream method
async with original_stream_method(self, ctx) as stream:
yield stream
+ if messages:
+ _set_input_messages(span, messages)
+
# After streaming completes, update span with response data
# The ModelRequestNode stores the final response in _result
model_response = NoneThis Bugbot Autofix run was free. To enable autofix for future PRs, go to the Cursor dashboard.
![sentry[bot]](https://avatars.githubusercontent.com/in/12637?s=60&v=4){kind=small}
alexander-alderman-webb
changed the base branch from
webb/pydantic-ai/remove-test
to
master
![sentry-warden[bot]](https://avatars.githubusercontent.com/in/2755838?s=60&v=4){kind=small}
alexander-alderman-webb
changed the title
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 2 potential issues.
❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
Reviewed by Cursor Bugbot for commit ffcb5f3. Configure here.
| if PydanticAIIntegration.are_request_hooks_available: | ||
| metadata = kwargs.get("metadata") | ||
| if not metadata: | ||
| kwargs["metadata"] = {"_sentry_span": None} |
There was a problem hiding this comment.
Metadata check inconsistency overwrites user-provided empty dicts
Low Severity
The run wrappers use if not metadata: which treats any falsy value (including {}) as missing, while patched_init in register_hooks correctly uses if metadata is None:. A user who explicitly passes metadata={} to run() or run_stream() will have their metadata silently replaced with {"_sentry_span": None}. Using if metadata is None: would be consistent with the patched_init check and correctly distinguish "not provided" from "provided as empty dict".
Additional Locations (1)
Reviewed by Cursor Bugbot for commit ffcb5f3. Configure here.
|
|
||
| return original_init(self, *args, **kwargs) | ||
|
|
||
| Agent.__init__ = patched_init |
There was a problem hiding this comment.
Pre-existing agents silently lose chat span instrumentation
Medium Severity
register_hooks patches Agent.__init__ to inject hooks into capabilities, but agents created before sentry_init() never go through patched_init and thus lack hooks. Those agents will still get invoke_agent and execute_tool spans (patched at class level), but will silently produce no gen_ai.chat spans. This is a regression from the monkey-patching fallback path, which patches ModelRequestNode/Model at the class level and works for all agent instances regardless of creation time.
Reviewed by Cursor Bugbot for commit ffcb5f3. Configure here.
2 participants
Description
Resolve tox failure by using the
pydantic-aihooks when available. See the failing run in #5945See changes to instruction setting in https://github.com/pydantic/pydantic-ai/pull/4123/changes#diff-71730070455237accebd709ba7fe41ce60edbd238af02e7d86a208cf8f10ab12. Per-run system instructions are now added inside
ModelRequestNode.runandModelRequestNode.stream, and are therefore not available before the functions run.Issues
Reminders
tox -e linters.feat:,fix:,ref:,meta:)