Skip to content

ci: make label-selected ASAN AI TAP shards reliable and complete #6098

Description

@renecannao

Context

PR #6083 validates the label-selected AddressSanitizer path for the central TAP build and its fan-out. All four central ASAN build variants succeed and the broad TAP fan-out passes, but CI-ai-g1 and CI-ai-g2 expose existing GenAI/MCP test and plugin-lifecycle problems.

The failing AI artifacts contain no ASAN or LeakSanitizer diagnostics. This tracker separates those findings from the ASAN workflow change so #6083 can remain focused.

This work intentionally leaves CI-unit-tests-asan-coverage unchanged.

Acceptance criteria

  • A trivial v3.0 PR with the ci:asan label completes CI-ai-g1 and CI-ai-g2 unless a genuine sanitizer defect is found.
  • All 22 tests declared by each AI group execute, or are explicitly reassigned to a documented workflow.
  • MCP/GenAI variable initialization and runtime projection do not depend on test order.
  • MCP endpoint tests use the plugin's explicit authentication contract.
  • Stale shell tests, paths, helpers, and runtime dependencies are either modernized or removed from the groups.
  • Vector database tests use deterministic correctness checks under instrumentation.

Work items

Related

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    GHarea/testAutomated testing infrastructurestatus/confirmedIssue has been confirmed/reproducedtype/testTest improvement, comment, change, or request

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions