Use a current sanitizer runtime on the Forgejo runner #617

Closed
opened 2026-10-20 16:46:22 +02:00 by erikinkinen · 0 comments
Owner

Goal

Run the full ASan/UBSan suite reliably on the Forgejo runner with a current runtime, while preserving the ordinary GCC build and all instrumentation.

Evidence

During #614 / PR #616, sanitized GCC 12 builds linked correctly but the remote tests stalled at startup. Runs 23501/23504 stopped reporting progress at the ID test. Bounded diagnostic runs 23509/23512 showed repeated AddressSanitizer:DEADLYSIGNAL, including in the trivial smoke executable. The remote host reports Linux 6.14.0-1-pve. These failed/stalled runs were retained and then cancelled to stop the output loop.

All 45 tests pass locally under Clang 22 ASan/UBSan, ordinary GCC 16 Release, and GCC 12 ASan/UBSan inside the same node:20-bookworm image. This points to a runner/runtime interaction rather than establishing an AES memory error. The pattern is consistent with the older-runtime/high-ASLR incompatibility discussed in google/sanitizers #1614; exact host ASLR settings still require read-only confirmation. LLVM reports the ASan fix in version 17.0.0.

Steps

  • Preserve the failed-run evidence and environment comparison, and select an official container/toolchain with a modern sanitizer runtime.
  • Add a separate full sanitized CI job with finite time limits, keeping the existing ordinary GCC job and all sanitizer instrumentation enabled.
  • Validate all existing tests and legacy goldens in the selected container locally and on both Forgejo push/PR checks; document the supported sanitizer commands.

Done

Both ordinary and sanitized suites pass with recorded compiler/runtime identities. No test, sanitizer, memory check or host address-randomization setting is disabled to obtain a pass.

## Goal Run the full ASan/UBSan suite reliably on the Forgejo runner with a current runtime, while preserving the ordinary GCC build and all instrumentation. ## Evidence During #614 / PR #616, sanitized GCC 12 builds linked correctly but the remote tests stalled at startup. Runs 23501/23504 stopped reporting progress at the ID test. Bounded diagnostic runs 23509/23512 showed repeated `AddressSanitizer:DEADLYSIGNAL`, including in the trivial smoke executable. The remote host reports Linux 6.14.0-1-pve. These failed/stalled runs were retained and then cancelled to stop the output loop. All 45 tests pass locally under Clang 22 ASan/UBSan, ordinary GCC 16 Release, and GCC 12 ASan/UBSan inside the same node:20-bookworm image. This points to a runner/runtime interaction rather than establishing an AES memory error. The pattern is consistent with the older-runtime/high-ASLR incompatibility discussed in [google/sanitizers #1614](https://github.com/google/sanitizers/issues/1614); exact host ASLR settings still require read-only confirmation. LLVM reports the ASan fix in version 17.0.0. ## Steps - [x] Preserve the failed-run evidence and environment comparison, and select an official container/toolchain with a modern sanitizer runtime. - [x] Add a separate full sanitized CI job with finite time limits, keeping the existing ordinary GCC job and all sanitizer instrumentation enabled. - [x] Validate all existing tests and legacy goldens in the selected container locally and on both Forgejo push/PR checks; document the supported sanitizer commands. ## Done Both ordinary and sanitized suites pass with recorded compiler/runtime identities. No test, sanitizer, memory check or host address-randomization setting is disabled to obtain a pass.
erikinkinen added this to the Phase 2 milestone 2026-10-20 16:46:22 +02:00
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
erikinkinen/AES#617
No description provided.