Fix crash in case of iceberg table with mixed ORC/Parquet files

[alesapin]

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 very rare crash when Iceberg table contains files of mixed format (ORC and Parquet). Fixes #88126.

Documentation entry for user-facing changes

• Documentation is written (mandatory for new features)

Before we had a single initialization of KeyCondition for the whole read. Now we switch to per-file initialization. I don't expect any performance degradation, because opening/reading file is several orders of magnitude slower than analyzing something in WHERE condition.

---

Note

Medium Risk
Touches filter initialization and Parquet page-level predicate pushdown, which can affect row/column skipping and is exercised across multiple input formats; changes are localized but impact correctness/performance of filtering paths.

Overview
Fixes a crash when reading Iceberg tables containing mixed ORC and Parquet files by making FormatFilterInfo build key_condition/PREWHERE additional columns with std::call_once semantics via initKeyConditionOnce().

Updates ORC and Parquet input formats to use this new API and removes FormatFilterInfo::opaque and the Parquet V3-specific FilterInfoExt; Parquet page-level filter pushdown now derives per-column conditions directly from key_condition and keeps them alive in Parquet::Reader.

Adds a stateless regression test (04033_iceberg_orc_parquet_v3_crash) to ensure the mixed-format Iceberg read no longer segfaults.

^{Written by Cursor Bugbot for commit 13e3d24. This will update automatically on new commits. Configure here.}

[clickhouse-gh]

Workflow [PR], commit [13e3d24]

Summary: ❌

job_name	test_name	status	info	comment
AST fuzzer (amd_ubsan)		error		IGNORED

---

AI Review

1) Summary

This PR removes cross-format mutable state from FormatFilterInfo and localizes Parquet page-filter condition state into Parquet::Reader, while keeping lazy key-condition initialization via initKeyConditionOnce. It also adds a stateless regression test for mixed ORC/Parquet Iceberg data with input_format_parquet_use_native_reader_v3 = 1. High-level verdict: the fix is correct and I found no Blocker or Major issues.

2) Missing context (if any)

• No CI report links or Praktika logs were provided in this review request.
• No benchmark artifacts were provided.

3) Findings (by severity)

No Blocker or Major findings.

4) Tests & Evidence

• Positive coverage: added tests/queries/0_stateless/04033_iceberg_orc_parquet_v3_crash.sh reproduces the mixed-format Iceberg scenario and verifies stable output.
• Negative/error-handling coverage: no explicit error-path assertions were added; this is acceptable for this regression shape.
• Additional test guidance (optional follow-up): add one variant where table format is Parquet and one file is ORC (reverse mix) to guard the symmetric path.

5) ClickHouse Compliance Checklist (Yes/No + short note)

• Data deletions logged? Yes (N/A) — no data-deletion logic changed.
• Serialization formats versioned? Yes (N/A) — no serialization format changes.
• Experimental setting gate present? Yes (N/A) — no new feature introduced.
• Settings exposed for constants/thresholds? Yes (N/A) — no new constants needing settings.
• Backward compatibility preserved? Yes — behavior is a bug fix in reader initialization/state handling.
• SettingsHistory.cpp updated for new/changed settings? Yes (N/A) — no setting changes.
• Existing tests untouched (only additions)? Yes — existing tests were not relaxed/deleted.
• Docs/user-facing notes updated? Yes (N/A) — internal bug fix; test added.
• Core-area change got extra scrutiny? Yes — format readers and filtering state handling were reviewed.

6) Performance & Safety Notes

• Safety: removing format_filter_info->opaque type-punning across formats reduces invalid-state risk.
• Performance: no hot-path algorithmic regression identified; extractSingleColumnConditions remains conditional on page_filter_push_down and key_condition.
• Benchmark status: none provided; minimal benchmark suggestion — compare SELECT count() and filtered scans on a mixed-format Iceberg dataset with V3 reader on/off.

7) User-Lens Review

The behavior is intuitive: mixed-format Iceberg reads now avoid the prior exception path and return expected rows. Error/actionability is unchanged, and the new regression test makes the fix robust against reintroduction.

8) Final Verdict

• Status: Approve

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, have a team admin enable autofix in the Cursor dashboard.}

[clickhouse-gh]

LLVM Coverage Report

Metric	Baseline	Current	Δ
Lines	83.80%	83.80%	+0.00%
Functions	23.90%	23.80%	-0.10%
Branches	76.30%	76.30%	+0.00%

PR changed lines: PR changed-lines coverage: 95.31% (61/64)
Diff coverage report
Uncovered code