Fix projection matching regression from removeTrivialWrappers in appendExpression

[amosbird]

Summary

PR #88798 added removeTrivialWrappers inside QueryDAG::appendExpression to strip materialize/identity wrappers, enabling aggregate projection matching through views. However, calling it during incremental DAG merging (per expression step) breaks the name chain between steps: early stripping changes output names (e.g. materialize(val) → val) before the next mergeInplace expects them, causing three different failures depending on query shape:

• NOT_FOUND_COLUMN_IN_BLOCK: materialize(<col>) on UNION ALL view + projection + WHERE on aggregated column (regression 26.2 → 26.4) #104117 — NOT_FOUND_COLUMN_IN_BLOCK: materialize(val) on UNION ALL view + projection + WHERE on aggregated column
• LOGICAL_ERROR: Block structure mismatch between MergeSortingTransform and WindowTransform with projection + window + WHERE on non-key column when query_plan_remove_unused_columns=0 (regression in 26.4) #104235 — LOGICAL_ERROR: Block structure mismatch between MergeSortingTransform and WindowTransform with projection + window function + WHERE on non-key column (query_plan_remove_unused_columns = 0)
• Block structure mismatch (AMBIGUOUS_COLUMN_NAME) when projection is picked and SELECT alias collides with source column name (regression from #88798) #104256 — AMBIGUOUS_COLUMN_NAME when SELECT alias collides with source column name + projection (query_plan_remove_unused_columns = 0)

The fix moves removeTrivialWrappers from appendExpression (called per-step) to the end of build (after the full DAG is assembled and filter nodes are wired into outputs). This preserves the original fix for #32753 (projection matching through views) while avoiding the name resolution breakage during incremental merging.

Note on #104256 empty results: The empty-result behavior when SELECT toString(x) AS x ... WHERE x IN ... is a pre-existing analyzer semantic — alias shadows the source column name in WHERE regardless of projections. This is controlled by prefer_column_name_to_alias (default 0 = prefer alias). The regression from #88798 was only the exception (AMBIGUOUS_COLUMN_NAME), not the empty result.

Closes #104117
Closes #104235
Closes #104256

Changelog category (leave one):

• Bug Fix (user-visible misbehavior in an official stable release)

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

Fix exceptions (NOT_FOUND_COLUMN_IN_BLOCK, LOGICAL_ERROR, AMBIGUOUS_COLUMN_NAME) when using projections with UNION ALL views, window functions, or alias-column name collisions. Regression from #88798.

Documentation entry for user-facing changes

• Documentation is not required

[clickhouse-gh]

Workflow [PR], commit [9139e41]

Summary: ✅

---

AI Review

Summary

This PR moves removeTrivialWrappers from per-step QueryDAG::appendExpression to the end of QueryDAG::build, which matches the stated intent to avoid breaking incremental DAG merge/name resolution while keeping wrapper cleanup before projection analysis. The change is small, targeted, and accompanied by a new stateless regression test that covers the three reported exceptions; I did not find correctness, safety, or compatibility issues in the patch itself.

Missing context

• ⚠️ No CI logs/results were available in this review context, so CI health could not be independently validated here.

ClickHouse Rules

Item	Status	Notes
Deletion logging	➖
Serialization versioning	➖
Core-area scrutiny	✅
No test removal	✅
Experimental gate	➖
No magic constants	✅
Backward compatibility	✅
SettingsChangesHistory.cpp	➖
PR metadata quality	✅
Safe rollout	✅
Compilation time	✅
No large/binary files	✅

Final Verdict

Status: ✅ Approve

[clickhouse-gh]

LLVM Coverage Report

Metric	Baseline	Current	Δ
Lines	84.10%	84.10%	+0.00%
Functions	91.10%	91.10%	+0.00%
Branches	76.60%	76.60%	+0.00%

Changed lines: 100.00% (7/7) | lost baseline coverage: 1 line(s) · Uncovered code

Full report · Diff report