Failed
Open on GitHub

fix(start-server-core): preserve live response headers

SSheraff
f2e3421a
8340Latest execution

Self-Healing CI

Classification: Environment state

We classified this as an environment_state failure because only 1 of 16 serverFn redirect combinations failed with network-level errors (fetch failed × 35, ECONNRESET), while all other serverFn redirect variants — including the same thrower/scenario with different parameters — passed successfully. This inconsistent pattern cannot be traced to specific lines in the setResponseHeaders diff and does not meet the burden of proof required for a code_change classification.

No code changes are applicable.

The failing test is redirects › serverFn › serverFn redirects to target: internal, scenario: navigate, thrower: beforeLoad, reloadDocument: false in tanstack-react-start-e2e-basic:test:e2e--vite-prerender.

Why this is not a code_change:

  1. High burden of proof not met: The code_change classification requires being able to identify exact lines in the diff that cause the failure. The errors are network-level (fetch failed × 35, ECONNRESET) and cannot be traced to specific lines in the setResponseHeaders change.

  2. Inconsistent failure pattern: The PR changes setResponseHeaders in @tanstack/start-server-core. If this broke serverFn redirects, we would expect all 16 serverFn redirect combinations to fail. Instead, only 1 specific combination fails (internal + navigate + beforeLoad + reloadDocument: false) while all 15 other serverFn redirect variants pass — including the same thrower/scenario with reloadDocument: true, and all external/direct_visit combinations.

  3. Flakiness rate is 0%: This rules out flaky_task, but combined with the network-level error pattern (fetch failed, ECONNRESET appearing earlier in the same run), it points to a transient infrastructure/environment issue rather than a deterministic code regression.

  4. No similar failures in other branches: The similar-task-failure-detector found no matching failures on other branches, which is consistent with either classification — but does not provide the required evidence to meet the code_change burden of proof.

  5. PR author validated redirect tests: The PR body explicitly states "all 37 redirect tests passed" during local validation, further reducing confidence that the diff is the root cause.