25.8: Fix Nullable(X)->Y cross-type ALTER MODIFY COLUMN (follow-up to #84770)

[il9ue]

Changelog category (leave one):

• Bug Fix

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

Fix ALTER MODIFY COLUMN failure when converting nullable columns to a different non-nullable type (e.g. Nullable(UInt8) to String). Previously, cross-type nullable conversions would fail with NO_COMMON_TYPE error. Follow-up to upstream ClickHouse#84770.

CI/CD Options

Exclude tests:

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

Regression jobs to run:

• Fast suites (mostly <1h)
• Aggregate Functions (2h)
• Alter (1.5h)
• Benchmark (30m)
• ClickHouse Keeper (1h)
• Iceberg (2h)
• LDAP (1h)
• Parquet (1.5h)
• RBAC (1.5h)
• SSL Server (1h)
• S3 (2h)
• Tiered Storage (2h)

[il9ue]

Context

This is a follow-up fix for the backport of upstream ClickHouse/ClickHouse#84770 ("Handle NULLs in ALTER MODIFY COLUMN"), which was backported to Altinity 24.8 via #1370 and to 25.8 via #1344.

Problem

PR ClickHouse#84770 correctly handles Nullable(X) → X conversions (same base type), but fails for Nullable(X) → Y where X != Y (cross-type conversion).

Reproducer:

CREATE TABLE XXX (x Nullable(UInt8), y String) ENGINE=MergeTree ORDER BY tuple();
INSERT INTO XXX (y) VALUES (1);
ALTER TABLE XXX MODIFY COLUMN x String DEFAULT '';
-- FAILS: NO_COMMON_TYPE error

This was reported by a customer on Altinity Stable 24.8.14. The same failure is reproducible on upstream v26.1.6.6 (ClickHouse Playground), while v26.2.5.45 succeeds.

Root Cause

convertRequiredExpressions() in inplaceBlockConversions.cpp generates:

_CAST(ifNull(column, default_value), 'TargetType')

When converting Nullable(UInt8) → String, ifNull(Nullable(UInt8), String_default) internally calls getLeastSupertype({UInt8, String}), which throws NO_COMMON_TYPE because UInt8 and String have no common supertype.

Why cherry-picking the upstream fix (PR ClickHouse#96790) is not sufficient

Upstream resolved this in ClickHouse/ClickHouse#96790 by modifying ifNull.cpp to use getLeastSupertypeOrVariant() when use_variant_as_common_type = true. In v26.2+, this setting defaults to true, so the ifNull call succeeds by returning Variant(UInt8, String).

However, on Altinity 24.8 and 25.8, use_variant_as_common_type defaults to false. Cherry-picking ClickHouse#96790 alone would NOT fix the bug on these branches — the code path would still call getLeastSupertype() and throw.

Fix

Instead of modifying ifNull behavior, this PR fixes the expression generation in convertRequiredExpressions() directly.

Before (broken for cross-type):

_CAST(ifNull(col, default), 'String')

After (works for all type combinations):

ifNull(_CAST(col, 'Nullable(String)'), _CAST(default, 'String'))

This works because:

1. _CAST(Nullable(UInt8), 'Nullable(String)') → Nullable(String) — standard type cast, always succeeds
2. ifNull(Nullable(String), String) → String — trivial, same base types
3. No dependency on use_variant_as_common_type or any other settings
4. Same-type conversions (Nullable(X) → X) continue to work correctly

Scope

• Files changed: src/Interpreters/inplaceBlockConversions.cpp (~10 lines in convertRequiredExpressions())
• Tests added: Cross-type test case in 03575_modify_column_null_to_default.sql
• Risk: Low — only affects the nullable-to-non-nullable code path when base types differ

Branches

This fix applies to both Altinity branches that have PR ClickHouse#84770:

• 24.8.14 ← separate PR (cherry-pick of same commit)
• 25.8.16 ← this PR

Altinity 25.3 does not have ClickHouse#84770 and is not affected.

[github-actions]

Workflow [PR], commit [c613cbf]

[il9ue]

Updated: moved test to a separate file to fix CI failures

The -- { echoOff } directive didn't suppress SQL echo as expected because the original test (03575_modify_column_null_to_default.sql) runs in -- { echoOn } mode throughout. Appended test lines were being echoed to stdout, causing a mismatch with the .reference file.

What changed in this amend:

• Removed our additions from the existing 03575_modify_column_null_to_default.sql / .reference (restored to original)
• Created a new dedicated test file: 03575_modify_column_null_to_default_cross_type.sql / .reference
• Added no-random-settings, no-random-merge-tree-settings tags to avoid randomized settings interfering with the test
• The C++ fix in inplaceBlockConversions.cpp is unchanged

Other CI failures are pre-existing and unrelated to this PR:

• 02477_single_value_data_string_regression — memory limit exceeded (flaky, environment-dependent)
• RabbitMQ integration tests — all failing (infrastructure issue)
• Integration tests timeout / Docker image errors — CI runner issues
• Grype scan — CVE scan finding pre-existing vulnerabilities in base image
• DCO sign-off — now fixed with git commit --amend -s