We don't believe this PR caused these 9 failures. The diff only touches AST node-copying/provenance bookkeeping in router-utils and router-plugin, none of the failing projects were modified by this PR, and the errors themselves (Babel source-map config validation, RSC virtual-runtime module resolution, and an Rsbuild parser diagnostic wording mismatch) match what the PR author already documented as identical, pre-existing failures on the unchanged #8504 baseline.
Self-Healing CI
Classification: Environment state
An empty commit was applied to the branch to trigger a new CI pipeline execution to resolve the environment issue.
This was not classified as a code change for the following reasons:
Diff scope: The PR's only production changes are in packages/router-utils/src/ast.ts (fuses a structuredClone call and a separate provenance-tracking walk into one iterative copy that records originalNodes/copiedNodes together) and packages/router-plugin/src/core/code-splitter/compilers.ts (consumes the returned copiedNodes map instead of rebuilding it with a second walk). Neither file touches Babel configuration, rsbuild/webpack loader wiring, source-map generation, or the RSC virtual-runtime plugin machinery.
Project scope: None of the 9 failing tasks' projects (tanstack-solid-start-e2e-basic, tanstack-vue-start-e2e-basic, tanstack-react-start-e2e-rsc, tanstack-react-start-e2e-rsc-rsbuild, tanstack-solid-start-e2e-deferred-hydration, tanstack-react-start-e2e-rsbuild, tanstack-react-start-e2e-rsc-deferred-hydration, tanstack-react-start-e2e-rsc-query) appear in touched_projects (root, tanstack-solid-start-e2e-start-manifest, @tanstack/router-plugin, @tanstack/router-utils), and touched_projects is not '*'.
Error content: Reading the full terminal output for every failing task shows exactly three unrelated error signatures, none of which reference any changed function, symbol, or logic path in ast.ts or compilers.ts:
- Three tasks fail with "Error: .inputSourceMap must be a boolean, object, or undefined" thrown from @babel/core's loadPartialConfigAsync inside @rsbuild/plugin-babel's babel-loader — a Babel/Rsbuild loader configuration validation error, not an AST-shape problem.
- Three build tasks and two e2e/runtime tasks fail with "ENOENT ... open '.../node_modules/.virtual/virtual_tanstack-rsc-runtime.js'" or its runtime form "argument 'path' must be a string ... Received '\x00virtual:tanstack-rsc-runtime'" — a Vite/Rsbuild virtual-module resolution problem in the RSC plugin's runtime wiring, unrelated to node copying.
- One task fails a Playwright snapshot assertion on the exact wording of a syntax-error overlay message — a parser/loader diagnostic-formatting difference, not a functional AST regression.
Corroboration: A similar-task-failure-detector subagent was dispatched to check branch 8522, but mcp__utilities__similar_task_failure_outputs returned no data for that branch, so no direct tool-verified comparison was possible. That gap is offset by the PR body itself, which states — with a link to the actual CI run — that this branch's initial CI run reported these same nine failing tasks with identical errors to the exact #8504 baseline run (the unmodified branch this PR stacks on), and that the fixture's emitted client files are byte-identical between the two builds. That documented comparison points the same direction as the diff and error-signature analysis above.
Because none of the three error categories trace to any changed line, symbol, or logic path in the diff, and none occur in a project the PR modified, the high burden of proof required for code_change (exact lines in the diff causing the failure) is not met. The failures are also deterministic, reproducible loader/plugin-configuration and virtual-module-resolution issues rather than test-only timing or race conditions, so flaky_task doesn't fit either — leaving environment_state as the correct classification, with no code change proposed.