fix: handle None content.parts in google_genai#5872
Closed
thakoreh wants to merge 60 commits intogetsentry:mainfrom
Closed
fix: handle None content.parts in google_genai#5872thakoreh wants to merge 60 commits intogetsentry:mainfrom
thakoreh wants to merge 60 commits intogetsentry:mainfrom
Conversation
…#5649) ## Summary - Add two GitHub Agentic Workflows (`gh aw`) that auto-generate and maintain developer-facing codebase documentation in `docs/codebase/` - **`docs-codebase-refresh`**: full regeneration of every doc page, triggered on merge to main/master or manual dispatch - **`docs-codebase-update`**: incremental update of only affected pages, triggered on merge when `sentry_sdk/**`, `MIGRATION_GUIDE.md`, or `CHANGELOG.md` change - Both workflows create PRs (never direct commits) for human review - Includes a portable `BUILD_PLAN.md` designed for reuse across other Sentry SDKs -- only the SDK Context block changes per language (will remove later) ### How This Relates to Other Sentry Docs - **[docs.sentry.io/platforms/\*](https://docs.sentry.io/)** tells users *what to do* -- setup guides, config options, API usage. - **[develop.sentry.dev/sdk/](https://develop.sentry.dev/sdk/)** tells SDK authors *what to build* -- protocol spec, envelope format, required behaviors. - **API reference** (Sphinx, TypeDoc, Javadoc, etc.) tells developers *what the API surface looks like* -- auto-generated from docstrings/annotations. Lists signatures, parameters, return types. - **`docs/codebase/*`** (this) explains *what was built and how it works* -- architecture, data flow, how modules connect, and why. Generated from full source analysis, not just docstrings. Aimed at SDK contributors and maintainers. ### Files added | File | Purpose | |------|---------| | `docs/codebase/BUILD_PLAN.md` | Portable blueprint with porting checklist | | `docs/codebase/_meta/style-guide.md` | SDK-agnostic formatting rules and page templates | | `docs/codebase/_meta/manifest.json` | Empty manifest (populated by first workflow run) | | `.github/workflows/docs-codebase-refresh.md` | Full refresh workflow source | | `.github/workflows/docs-codebase-update.md` | Incremental update workflow source | | `.github/workflows/docs-codebase-*.lock.yml` | Compiled Actions workflows | | `.gitattributes` | Marks `docs/codebase/**` as `linguist-generated` | ## Test plan - [x] `gh aw compile` produces 0 errors, 0 warnings for both workflows - [ ] Manual trigger via `gh aw run docs-codebase-refresh` generates initial docs - [ ] Verify generated pages cover all integrations in `_MIN_VERSIONS` - [ ] Push a source change to main, verify incremental workflow updates only affected pages - [ ] Both workflows create PRs (not direct commits) 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
…sentry#5654) ## Summary - Switch `docs-codebase-refresh` and `docs-codebase-update` GitHub Agentic Workflows from the default Copilot engine to `engine: claude` - Lock files recompiled via `gh aw compile` 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
…5655) Remove the agentic CI workflows and generated artifacts introduced for auto-generating codebase documentation. This reverts the work from getsentry#5649 and getsentry#5654, which added GitHub Actions workflows to generate and refresh docs using an AI agent. The generated docs files (`BUILD_PLAN.md`, `_meta/manifest.json`, `_meta/style-guide.md`) and associated action lock files are also removed. The `.gitattributes` file (which was added as part of that workflow setup) is removed as well. Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Add the `gen_ai.system` span attribute (set to `"anthropic"`) to the Anthropic integration. Other AI integrations (OpenAI, Langchain, Google GenAI, LiteLLM, Pydantic AI) already set this attribute, but it was missing from the Anthropic integration. The attribute is set in `_set_input_data()` which is called for every span (streaming/non-streaming, sync/async). Refs PY-2135 Closes getsentry#5657
Set the `gen_ai.response.id` property on spans created by the Anthropic integration. For non-streaming responses, the ID is read from `result.id` on the Message object. For streaming responses, it's captured from `event.message.id` in the `message_start` event and threaded through the iterator to be set when the stream completes. The `_collect_ai_data` function's return tuple is extended with the new `response_id` field, and `_set_output_data` accepts an optional `response_id` parameter to set on the span. Refs PY-2137 Closes getsentry#5659
Replace mocks with `httpx` types to avoid test failures when library internals change.
Use double quotes for JSON strings.
Create dedicated functions for patching synchronous and asynchronous response iterators.
…med response (getsentry#5564) Prepare for adding patches for `.stream()`, which iterate over `MessageStreamEvent`. `MessageStreamEvent` is a superset of `RawMessageStreamEvent` returned in the iterator from `create(stream=True)`, but `RawMessageStreamEvent` instances are sufficient to collect the information required for AI Client Spans.
Run post-iterator steps in a finally block so the AI Client Span is finished even if the generator does not complete.
…m()` (getsentry#5565) Patch `Messages.stream()` and `MessageStreamManager.__enter__()` to create AI Client Spans. Re-use existing code for setting attributes on AI Client Spans based on arguments to `anthropic` functions. Adapt tests that return a synchronous response stream with `create(stream=True)`.
…stream()` (getsentry#5572) Patch `AsyncMessages.stream()` and `AsyncMessageStreamManager.__enter__()` to create AI Client Spans. Adapt tests that return a asynchronous response stream with `create(stream=True)`.
…ated by `stream()` (getsentry#5665)
…#5638) Couple things going on in this PR. Bear with me, this is probably the most all over the place span first PR because the outgoing trace propagation changes make mypy complain about things elsewhere in the sdk. ### 1. Outgoing trace propagation Support getting trace propagation information from the span with `_get_traceparent`, `_get_baggage`, `_iter_headers`, etc. These mirror the old `Span` class to make integrating `StreamedSpan`s with the rest of the SDK easier (since they're used throughout), with one difference: they're explicitly private, while the corresponding `Span` methods were public. Added aliases to them so that we can use the private methods everywhere. There is definite clean up potential here once we get rid of the old spans and we no longer have to make the streaming span interface work with the existing helper scope methods. ### 2. Addressing cascading mypy issues Now that we're officially allowing `StreamedSpan`s to be set on the scope, a LOT of type hints need updating all over the SDK. In many places, I've added explicit guards against functionality that doesn't exist in span first mode. This should prevent folks from using the wrong APIs in the wrong SDK mode (tracing vs. static) as well as make mypy happy. --------- Co-authored-by: Erica Pisani <pisani.erica@gmail.com>
…cycles (getsentry#5622) Fixes getsentry#5621 and PY-2127
We'll only support continuous profiling in span first. Note: Span-first profiling is not yet supported server-side.
…522f241581534dfc89bd99ec3b1da4f6 to 6b1f51ec8af03e19087df452b426aa7e46d2b20a (getsentry#5669)
…try#5678) Add `GEN_AI_RESPONSE_FINISH_REASONS` span data to the Anthropic integration by capturing `stop_reason` from API responses. For non-streaming responses, the `stop_reason` is read directly from the `Message` result. For streaming responses, it's extracted from the `MessageDeltaEvent` delta and passed through the `_collect_ai_data` helper. This brings the Anthropic integration in line with the OpenAI integration's finish reason tracking. --------- Co-authored-by: Claude <noreply@anthropic.com>
### Description In certain scenarios, the SDK's log batcher might cause a deadlock. This happens if it's currently flushing, and during the flush, something emits a log that we try to capture and add to the (locked) batcher. With this PR, we're adding a re-entry guard to the batcher, preventing it from recursively handling log items during locked code paths like `flush()`. #### Issues Closes getsentry#5681 #### Reminders - Please add tests to validate your changes, and lint your code using `tox -e linters`. - Add GH Issue ID _&_ Linear ID (if applicable) - PR title should use [conventional commit](https://develop.sentry.dev/engineering-practices/commit-messages/#type) style (`feat:`, `fix:`, `ref:`, `meta:`) - For external contributors: [CONTRIBUTING.md](https://github.com/getsentry/sentry-python/blob/master/CONTRIBUTING.md), [Sentry SDK development docs](https://develop.sentry.dev/sdk/), [Discord community](https://discord.gg/Ww9hbqr)
### Description <!-- What changed and why? --> #### Issues <!-- * resolves: getsentry#1234 * resolves: LIN-1234 --> #### Reminders - Please add tests to validate your changes, and lint your code using `tox -e linters`. - Add GH Issue ID _&_ Linear ID (if applicable) - PR title should use [conventional commit](https://develop.sentry.dev/engineering-practices/commit-messages/#type) style (`feat:`, `fix:`, `ref:`, `meta:`) - For external contributors: [CONTRIBUTING.md](https://github.com/getsentry/sentry-python/blob/master/CONTRIBUTING.md), [Sentry SDK development docs](https://develop.sentry.dev/sdk/), [Discord community](https://discord.gg/Ww9hbqr)
This PR makes the ASGI integration work both in legacy and span streaming mode. Some features and attributes will be missing in span streaming mode for now (see the Out of scope section below). Best reviewed with whitespace ignored: https://github.com/getsentry/sentry-python/pull/5680/changes?w=1 --- A bit of a background on migrating integrations to span first. In order to support both legacy spans and span streaming, most integrations will follow the same patterns: ### API We need to use the `start_span` API from `sentry_sdk.traces` if we're in span streaming mode (`traces_lifecycle="stream"`). There are no transactions anymore. Top-level spans will also be started via the `start_span` API in span streaming mode. ### Setting data on spans If an integration sets data on a span (via `span.set_data`, `span.set_tag` etc.), it should use `span.set_attribute` when span streaming is enabled. The attributes that we set need to be in Sentry conventions. This is deliberately not the case for most quick ports of integrations like this one and will follow in [a future step](getsentry#5152). ### Trace propagation If an integration sits at a service boundary and is capable of propagating incoming trace information (like WSGI/ASGI or Celery), in span first mode we need to switch from the old style `with continue_trace(...) as transaction:` to the [new style `continue_trace()` and `new_trace()`](https://sentry-docs-git-ivana-span-first-migration-guide.sentry.dev/platforms/python/migration/span-first/#trace-propagation) (not context managers). ### `start_span` arguments You can pass things like `op`, `origin`, `source` to the old `start_span` API. With the new API, this is no longer possible, and the individual properties need to be set as attributes directly. ### Out of scope For now, the following is out of scope and will follow in the future: - Making sure all attributes are correct and that they're set in Sentry conventions: getsentry#5152 - Migrating event processors (as streaming spans are not events, event processors are not run on them, meaning some data will not be set yet): getsentry#5152
…etsentry#5683) Consolidate span finishing logic in a new `_StreamSpanContext` context manager. Forward exception info from `_StreamSpanContext.__exit__()` to `Span.__exit__()`.
Update our test matrix with new releases of integrated frameworks and libraries. ## How it works - Scan PyPI for all supported releases of all frameworks we have a dedicated test suite for. - Pick a representative sample of releases to run our test suite against. We always test the latest and oldest supported version. - Update [tox.ini](https://github.com/getsentry/sentry-python/blob/master/tox.ini) with the new releases. ## Action required - If CI passes on this PR, it's safe to approve and merge. It means our integrations can handle new versions of frameworks that got pulled in. - If CI doesn't pass on this PR, this points to an incompatibility of either our integration or our test setup with a new version of a framework. - Check what the failures look like and either fix them, or update the [test config](https://github.com/getsentry/sentry-python/blob/master/scripts/populate_tox/config.py) and rerun [scripts/generate-test-files.sh](https://github.com/getsentry/sentry-python/blob/master/scripts/generate-test-files.sh). See [scripts/populate_tox/README.md](https://github.com/getsentry/sentry-python/blob/master/scripts/populate_tox/README.md) for what configuration options are available. _____________________ _🤖 This PR was automatically created using [a GitHub action](https://github.com/getsentry/sentry-python/blob/master/.github/workflows/update-tox.yml)._ --------- Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com> Co-authored-by: Ivana Kellyer <ivana.kellyer@sentry.io>
Reusing the `toxgen/update` branch can get annoying, especially since it needs manual babysitting (empty commit to trigger CI). Let's use distinct names instead. ## Summary - Instead of always reusing the same `toxgen/update` branch, the "Update test matrix" workflow now creates a date-based branch (e.g. `toxgen/update-03-19`). - Same-day reruns force-push to the same branch, so they overwrite rather than fail. - The PR-closing logic now finds and closes all open PRs from any `toxgen/` branch, not just the current one. ## Test plan - [ ] Trigger the "Update test matrix" workflow manually and verify it creates a branch like `toxgen/update-MM-DD` - [ ] Trigger it again on the same day and verify it overwrites the branch and closes the previous PR - [ ] Verify old toxgen PRs from different branches get closed Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…#5706) ## Summary - For every auto-generated integration, adds a tox environment that aliases the highest tested version (e.g. `tox -e py3.14-httpx-latest` = `tox -e py3.14-httpx-v0.28.1`) - Makes it easy to run tests against the latest version without looking up exact version strings ## Notes - The new latest env is NOT run in CI - The new latest env points to the highest non-prerelease env --------- Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…sentry#5714) Add the experimental `suppress_asgi_chained_exceptions` option. The option defaults to `True` when unset, preserving the current behavior.
## Summary - Pin all GitHub Actions references in `.github/` workflow files to full-length commit SHAs Generated by `devenv pin_gha`. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
…etsentry#5796) Update LangChain integration to use the new `gen_ai.generate_text` operation for LLM call spans, aligning with OpenTelemetry semantic conventions. **Changes:** - Changed LLM span operation from `gen_ai.pipeline` to `gen_ai.generate_text` - Updated span naming to include the model identifier: `generate_text {model}` instead of generic "Langchain LLM call" - Removed unnecessary docstring from callback method - Updated test assertions to validate the new operation and naming convention This change improves observability by using more specific operation types that accurately reflect the semantic nature of LLM generation calls, and includes the model identifier for better span context. Related to SDK-669. Replaces some of the changes introduced in getsentry#5705 (this is being broken down into 2 parts due to upcoming changes in the langchain test suite) Co-authored-by: Claude Haiku 4.5 <noreply@anthropic.com>
Replace mocks with `httpx` types to avoid test failures when library internals change.
…5726) Add test for LangChain v1.0 functionality using sample Responses API output reused from `openai-agents` tests.
Add test for LangChain v1.0 functionality using sample Responses API output with a tool call request reused from `openai-agents` tests.
Replace test with manual hook calls with a test that uses the library in a way that the same hooks are invoked.
Normalize multiline SQL query whitespace in the asyncpg integration so that span descriptions contain single-line queries with collapsed whitespace. asyncpg passes raw multiline SQL strings as span descriptions. This makes it difficult for users to match queries in `before_send_transaction` callbacks — they'd need to account for newlines and varying indentation instead of writing simple substring checks like `"SELECT id, name FROM users" in desc`. Fixes PY-2255 and getsentry#5850 --------- Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add a GitHub Actions workflow that automatically converts non-draft PRs to draft on open/reopen, and document the draft PR requirement in CONTRIBUTING.md. **Motivation:** Our [code submission standard](https://develop.sentry.dev/sdk/getting-started/standards/code-submission/#pull-requests) says PRs must start as drafts, but this wasn't enforced. Non-draft PRs trigger review notifications prematurely and signal readiness when the author may not be done. This workflow nudges contributors to open as draft by automatically converting non-draft PRs and posting a comment explaining the policy. **What's included:** - `.github/workflows/enforce-draft-pr.yml` — triggers on `pull_request_target` (opened, reopened), uses the GraphQL `convertPullRequestToDraft` mutation. Includes error handling and comment deduplication on reopen. - `CONTRIBUTING.md` — new "Pull Requests" section documenting the draft requirement and linking to the code submission standard. **Limitations:** GitHub has no pre-creation hook, so the initial CI run and review notifications still fire before the conversion. The value is in training contributors and ensuring draft state before a maintainer picks up the PR. No exemptions — applies to everyone (maintainers, internal, external contributors). --------- Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
hasattr passes when content.parts is None, but iterating over it crashes with TypeError. Switch to a truthiness check instead. Fixes getsentry#5854
Contributor
Semver Impact of This PR🟢 Patch (bug fixes) 📋 Changelog PreviewThis is how your changes will appear in the changelog. Bug Fixes 🐛
🤖 This preview updates automatically when you update the PR. |
Contributor
|
Resolved by bcfb788 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
8 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes #5854
Gemini sometimes returns candidates where content.parts exists but
is None. hasattr lets this through and we crash on iteration.
Switched to a truthiness check and added a regression test.