fix(sdk): decode SSE error body before formatting in stream client

[CAN-Aidoo]

Summary

Decode the bytes body in SSEClient._handle_sse_response before interpolating it into the error message, using .decode(errors=\"replace\") to stay safe on non-UTF-8 responses. Drops the acknowledging # TODO, the # type: ignore[str-bytes-safe], and the # noqa: E501.

Motivation

response.aread() returns bytes, so the previous f-string rendered the body as b'{\"error\":...}' — leading b, surrounding quotes, and \xNN escapes for any non-ASCII byte. The success path two lines below already does response_text.decode(); this PR reuses that pattern for the error path. The existing # TODO: Fix bytes formatting - should decode error_text or use !r in the file called this out.

Test plan

• New tests/evaluator/test_stream_client.py covers the happy path (readable decoded body in the raised exception, no b' artifact) and non-UTF-8 input (does not raise on decode).
• uv run ruff check passes on the modified files.

Summary by CodeRabbit

• Bug Fixes

  • Improved error messages for failed streaming responses so response bodies are shown as readable text instead of raw bytes.
  • Ensured invalid text in error responses no longer breaks error handling; the HTTP status code is still included.

• Tests

  • Added coverage for error-response handling, including valid text bodies and invalid UTF-8 cases.

[CLAassistant]

All committers have signed the CLA.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: f61520e1-8779-4c5b-b3bd-40d281f54e4b

📥 Commits

Reviewing files that changed from the base of the PR and between 93429cf and e1a073c.

📒 Files selected for processing (2)

• packages/traceloop-sdk/tests/evaluator/test_stream_client.py
• packages/traceloop-sdk/traceloop/sdk/evaluator/stream_client.py

---

📝 Walkthrough

Walkthrough

The SSE error-handling path in SSEClient._handle_sse_response now decodes the error response body bytes to a string using errors="replace" before constructing the error message, replacing prior bytes-based formatting. New async tests validate decoded exception messages for both valid and invalid UTF-8 error bodies.

Changes

SSE Error Body Decoding

Layer / File(s)	Summary
Decode error body in SSEClient packages/traceloop-sdk/traceloop/sdk/evaluator/stream_client.py	Error body bytes are decoded to a string with errors="replace" before being used in the constructed error message, removing the prior bytes-based approach and its TODO comment.
Tests for decoded error messages packages/traceloop-sdk/tests/evaluator/test_stream_client.py	Adds a _FakeResponse helper and two tests verifying that valid UTF-8 JSON error bodies are decoded into exception messages and invalid UTF-8 bodies still raise exceptions containing the status code.

Estimated code review effort: 1 (Trivial) | ~5 minutes

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the main change: decoding SSE error bodies before formatting them in the stream client.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}