Repository navigation
Antalya 26.8, backport of 118005: Suppress lifetime-safety-return-stack-addr-moved from clang 23, not 24 - #2828
Conversation
The suppression was gated on clang 24 on the assumption that nothing below it
emits the diagnostic. Clang 23 emits it, so a build with clang 23 fails: the
warning is on by default, `WERROR` is on by default, and
`src/Parsers/Prometheus/PrometheusQueryTree.cpp:165` trips it ten times, once per
template instantiation.
error: stack memory associated with local variable 'new_node' may be returned
later. This could be a false positive as the storage may have been moved.
[-Werror,-Wlifetime-safety-return-stack-addr-moved]
The three related `lifetime-safety` diagnostics are already gated on 23 and are
suppressed correctly; only this one is not, so the compile command carries
`-Wno-lifetime-safety-intra-tu-suggestions`,
`-Wno-lifetime-safety-cross-tu-suggestions` and
`-Wno-lifetime-safety-invalidation` but not this one.
Nothing about the code is wrong - the `unique_ptr` payload is on the heap and
outlives the local, which the comment already says. Lower the gate to match the
compiler that actually emits it, and drop the stale note about clang 24 being
Emscripten-only.
CI triage for
|
| Check | Verdict | Evidence |
|---|---|---|
| DCO | Commit e9eafcca9f has no Signed-off-by: line |
|
| Stateless tests (amd_asan_ubsan, cas s3 storage, parallel, 1/2 and 2/2) | ❌ Not PR-related: job timeout, also on other antalya-26.8 PRs |
Same ERROR on #2834 and #2827 |
Integration tests (amd_asan_ubsan, db disk, old analyzer, 1/8 and 7/8): test_backward_compatibility/test_custom_s3_disk_remote_url_allow_hosts.py |
❌ Not PR-related: fails the same way on other antalya-26.8 PRs |
Same 2 tests fail on #2834, #2824 and #2827 |
| Regression release swarms | ❌ Not PR-related: fails on other antalya-26.8 PRs |
1 module (1 failed) on #2834, #2824 and #2827 too |
PR |
Summary check | Red only because of the failures above |
1. DCO (PR-related, process only)
The cherry-picked commit keeps the upstream author (navneet83) and has no sign-off. #2827 and #2824 pass DCO, so this check is specific to this PR.
Fix (pick one):
- Have someone with push rights re-sign the commit (
git commit --amend -s) and force-push. I'm not allowed to force-push, so I haven't done this. - Or use the DCO app's remediation flow: push an empty remediation commit, or have a maintainer mark DCO as passed on the check page. That's the usual way for backports that carry upstream authorship.
2. Stateless tests (amd_asan_ubsan, cas s3 storage, parallel, 1/2 and 2/2): job timeout
Both shards were killed by the job's 3.5h limit. Neither shard got through its tests:
WARNING: Timeout exceeded [12600] for [3065]
WARNING: Job timed out: [Stateless tests (amd_asan_ubsan, cas s3 storage, parallel, 1/2)], timeout [12600], exit code [-15]
- Shard 1 reached 5169 / 6890 tests and shard 2 reached only 2656 / 6933.
- Antalya 26.8, backport of 126267: Use less s3 requests to get metadata file #2834 and MacOS compatibility #2827 fail the same way: ERROR at about 12,660s on both shards.
- Fix wrong condition for direct io on export part/ partition #2824 passed in 7.8k and 9.8k seconds, so this config is close to its time limit on this branch even when it passes.
One test also failed in shard 1 before the timeout: 05387_cas_publish_counts_replicated.
-replicated_insert ref_move 1
+replicated_insert ref_move 3
ref_move was counted 3 times instead of 1. That looks like retries under ASan load on the shared CAS disk, and it has nothing to do with a warning flag. I couldn't compare it on the sibling PRs: their shards also timed
8000
out and their reports don't list individual tests.
Next step: nothing to fix in this PR. The ASan "cas s3 storage" shards need more shards or a longer timeout on antalya-26.8. 05387_cas_publish_counts_replicated may be worth a look by whoever owns CAS (#2831 is in flight).
3. Integration: test_backward_compatibility/test_custom_s3_disk_remote_url_allow_hosts.py
Both tests fail in fixture setup: the old-version node (image clickhouse/clickhouse-server:26.8.26.12) never becomes ready. The same thing happened on the retry.
helpers/cluster.py:6383: in wait_until_port_is_ready
E Exception: Timed out while waiting for instance `node' with ip address 172.16.1.7 to start. Container status: running, ...
The same two tests fail on #2834, #2824 and #2827, which have unrelated diffs.
Next step: a branch-level issue, not for this PR. The upstream old-version image probably doesn't start with this test's config (custom S3 disk or remote_url_allow_hosts). The node's clickhouse-server.err.log in logs.tar.gz would confirm that.
4. Regression release swarms
4 of 1535 scenarios failed (1 module (1 failed)). #2834, #2824 and #2827 also fail this check with 1 module (1 failed). It's a release-binary regression suite that a warning-flag change can't affect.
Bottom line: the code is fine to merge once DCO is sorted out (re-sign the commit or have a maintainer override the check). The other red checks fail on every recent antalya-26.8 PR I checked. Re-running them won't turn them green.
Reports: result_pr.json · comparison PRs #2834, #2824, #2827
Backport of ClickHouse#118005 by @navneet83
Changelog category (leave one):
Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):
Suppress
lifetime-safety-return-stack-addr-movedfrom clang 23, not 24 (ClickHouse#118005 by @navneet83)Documentation entry for user-facing changes
Allows to use clang-23 to build ClickHouse.
CI/CD Options
Exclude tests:
Regressi 8000 on jobs to run: