[zvonand]

Changelog category (leave one):

• Bug Fix

Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):

Fixed ALTER MODIFY COLUMN failing when converting a Nullable column to a different type with a DEFAULT (e.g. Nullable(UInt8) → String,LowCardinality(String), or Array(UInt8)). (ClickHouse#102156 by @il9ue)

Documentation entry for user-facing changes

...

CI/CD Options

Exclude tests:

• Fast test
• Integration Tests
• Stateless tests
• Stateful tests
• Unit tests
• Performance tests
• Aarch64 tests
• All with ASAN
• All with TSAN
• All with MSAN
• All with UBSAN
• All with Coverage
• All Regression
• Disable CI Cache

Regression jobs to run:

• Fast suites (mostly <1h)
• Aggregate Functions (2h)
• Alter (1.5h)
• Benchmark (30m)
• CAS (content-addressed storage; Antalya only)
• ClickHouse Keeper (1h)
• Iceberg (2h)
• LDAP (1h)
• OAuth (5m)
• Parquet (1.5h)
• RBAC (1.5h)
• SSL Server (1h)
• S3 (2h)
• S3 Export (2h)
• Swarms (30m)
• Tiered Storage (2h)

[zvonand]

Backporter: cherry-pick conflicts resolved by AI

Conflicted files:

• src/Interpreters/inplaceBlockConversions.cpp (both modified)

Commits:

• bfa54d5: cherry-pick with the conflicts committed as-is

• 635a666: resolution

• src/Interpreters/inplaceBlockConversions.cpp conflicted because upstream uses make_intrusive for AST nodes and stable-25.8 still uses std::make_shared. I kept the upstream logic in convertRequiredExpressions (ifNull(_CAST(col, 'Nullable(T)'), _CAST(default, 'T')) for targets that can be Nullable, and if(isNull(col), ...) for Array/Tuple/Map) and switched it to std::make_shared.

• The bug (_CAST(ifNull(col, default), 'T')) is present in stable-25.8, so the fix applies there.

• 03575_modify_column_null_to_default_cross_type.sql: removed Case 9. It sets allow_experimental_nullable_tuple_type, which doesn't exist in stable-25.8, and uses Nullable(Tuple(...)), which stable-25.8 rejects.

• 03575_modify_column_null_to_default_cross_type.reference: removed Case 9's two output lines to match.

[github-actions]

Workflow [PR], commit [635a666]

[zvonand]

@blau-ai

[blau-ai]

CI triage for #2709 (head 635a666cb3)

Verdict: 2 failing checks, 0 caused by this PR. 1 is an infra/network flake and 1 is the DCO check, which fails on every stable-25.8 backport. The PR check is just the aggregate gate. No code changes are needed.

Check	Failure	Classification
Stateless tests (amd_asan, distributed plan, sequential)	02789_filesystem_cache_alignment	❌ Not PR-related: outbound network timeout to GitHub
DCO	Missing Signed-off-by	❌ Not PR-related: process check, fails on sibling backports too

---

1. Stateless tests (amd_asan, distributed plan, sequential): 02789_filesystem_cache_alignment

Taken from the praktika report (Failed: 1, Passed: 653):

Code: 1000. DB::Exception: Timeout: connect timed out: 185.199.109.133:443. (POCO_EXCEPTION)
...
18. DB::WebObjectStorage::loadFiles(...)
19. DB::WebObjectStorage::tryGetFileInfo(...)
20. DB::MetadataStorageFromStaticFilesWebServer::existsFile(...)
21. DB::SystemRemoteDataPathsSource::nextFile()

• The test's own checks produced the expected output (OK/0 lines in stdout). It failed only because the query on system.remote_data_paths walks every disk the server knows about. That includes the ad-hoc type = web disk that 03355_mergetree_table_disk.sql registers earlier in the same server lifetime (endpoint = 'https://raw.githubusercontent.com/ClickHouse/web-tables-demo/...'). The connect to GitHub's IP (185.199.109.133) timed out from the runner.
• This PR only touches src/Interpreters/inplaceBlockConversions.cpp (the Nullable→non-Nullable DEFAULT conversion during ALTER MODIFY COLUMN). Neither the web object storage, the filesystem cache nor system.remote_data_paths is involved.
• Next step: safe to re-run. If you want to harden this, 02789 could filter system.remote_data_paths by disk_name, so a leftover web disk from another test can't be reached. That would go in a separate PR, not this one.

2. DCO

Neither commit (bfa54d5038 upstream merge, 635a666cb3 conflict resolution) has a Signed-off-by: trailer. The same check fails on other stable-25.8 backports, e.g. #2694, so it's the usual non-blocking state for these PRs. Next step: ignore, or have a maintainer set DCO to pass, following the usual process for backports.

---

Health check of the PR's own change

• The new test 03575_modify_column_null_to_default_cross_type passed Stateless tests (amd_asan, flaky check): 50/50 runs.
• Every other stateless suite that ran is green: Fast test, debug/ubsan/asan parallel+sequential, old analyzer + DatabaseReplicated, ParallelReplicas, AsyncInsert, distributed plan s3.
• tsan/msan/arm jobs were skipped on purpose by the CI options in the PR body (job reasons: job name includes excluded tag 'tsan'/'msan'/'arm'). So this PR has no TSAN/MSAN coverage. That's acceptable for a change limited to inplaceBlockConversions.cpp, but you should know it.
• The conflict resolution (std::make_shared instead of make_intrusive, dropping the Nullable(Tuple) case 9 that 25.8 can't express) matches the 25.8 APIs, and every build that ran succeeded.