schema: review-report-v2
target: CommitFest 6482 — CREATE TABLE LIKE INCLUDING COMMENTS copies table-level comment (uncommitted staged diff)
head: 4d5fdb21ddda46972d7703562c5ad4076feae8b8
base: 4d5fdb21ddda46972d7703562c5ad4076feae8b8
diff_sha: d9b8108fc514bf0326770c6bf9a3058abd033fd6e7eb04d93712042f0fbf713f
verdict: needs-review
blocker_counts:
  verified: 0
  unverified: 0
model: glm-5.2:cloud
source: postgresql-master
slug: commitfest-6482-review

CommitFest 6482 Review — CREATE TABLE LIKE INCLUDING COMMENTS copies table-level comment

Patch summary

CREATE TABLE ... LIKE ... INCLUDING COMMENTS previously copied comments on columns, check constraints, not-null constraints, indexes, and extended statistics, but not the comment on the source relation itself. This patch adds that copy.

Changed files (repo-relative):

• src/backend/parser/parse_utilcmd.c — implementation
• src/test/regress/expected/create_table_like.out — regression test expected
• src/test/regress/sql/create_table_like.sql — regression test SQL

Changed symbols: transformTableLikeClause, expandTableLikeClause

What the patch does:

1. Adds CREATE_TABLE_LIKE_COMMENTS to the deferral-option mask in transformTableLikeClause (parse_utilcmd.c:1311-1317) so expandTableLikeClause is driven when INCLUDING COMMENTS is requested.
2. Adds a new block in expandTableLikeClause (parse_utilcmd.c:1631-1652) that calls GetComment(RelationGetRelid(relation), RelationRelationId, 0) on the LIKE source and, if non-NULL, builds a CommentStmt with objtype = OBJECT_TABLE and object = list_make2(schemaname, relname) (or list_make1(relname)), appended to result.
3. Adds a regression test: COMMENT ON TABLE ctlt3 + SELECT obj_description('ctlt13_like'::regclass, 'pg_class').

Provenance / model routes

All seats used model glm-5.2:cloud (user-fixed override). Route policy: user-fixed-model.

Security review summary

No privilege/role escalation, SECURITY DEFINER/search_path, catalog write race/TOCTOU, SQL injection, crypto/TLS, or C memory-safety vulnerability is introduced by this diff. The CommentStmt targets the newly-created child relation (owned by the invoker); CommentObject re-checks ownership. The parent comment is read via GetComment under AccessShareLock (consistent with existing column/constraint/index/stat comment paths). No string SQL is built — the object name is a structured List of String nodes. The char *comment local is reused safely within the same memory context (not a leak/UAF under the PG memory-context filter).

Findings

CommentStmt *stmt = makeNode(CommentStmt);
stmt->objtype = get_relkind_objtype(childrel->rd_rel->relkind);
if (heapRel->schemaname)
    stmt->object = (Node *) list_make2(makeString(heapRel->schemaname),
                                      makeString(heapRel->relname));
else
    stmt->object = (Node *) list_make1(makeString(heapRel->relname));

Verification Summary

• Verified: 5 (F1, F2, F4, F5, COV-1)
• Cannot-verify: 1 (F3 — design decision pending)
• Refuted: 1 (F6 — benign)
• Blockers: 0 verified / 0 unverified

Repair Validation Summary

All candidate repairs were validated in a disposable worktree copy only; the reviewed worktree was not modified. Build/test commands executed: make -s -j4 (exit 0), make -C src/test/regress check-tests TESTS="cf6482_like_comments create_table_like" (All tests passed), make -C doc -s check (exit 0).

Changed-Code Coverage

Coverage requirement: required (changed instrumentable implementation in src/backend/parser/parse_utilcmd.c).

Lane: host-disposable (passed). Fresh disposable gcc 15.2.0 gcov build at frozen HEAD with the staged feature diff applied; --enable-coverage (-fprofile-arcs -ftest-coverage, -O0 -g); make -j4 exit 0; pg_regress create_table_like → ok 1 - create_table_like.

Counter epoch: fresh (all .gcda deleted before test run; .gcno from build). No integrity errors (no stamp/merge/profiling errors).

Test commands executed:

make -C src/test/regress check-tests TESTS=create_table_like

Coverage totals

Covered (10/11 lines): 1311 (mask if, 39 hits), 1312 (new COMMENTS bit), 1634 (if options & COMMENTS, 25 hits), 1636 (GetComment, 13 hits), 1638 (if comment != NULL, 13 hits), 1640 (makeNode, 1 hit), 1642 (objtype = OBJECT_TABLE, 1 hit), 1643 (if schemaname, 1 hit — line covered, branch 1 uncovered), 1644 (list_make2, 1 hit), 1648 (stmt->comment, 1 hit), 1650 (lappend, 1 hit).

Uncovered line

Line 1647 (expandTableLikeClause, src/backend/parser/parse_utilcmd.c):

else
    stmt->object = (Node *)list_make1(makeString(heapRel->relname));

• Coverability: coverable-now
• Reason: Else branch of if (heapRel->schemaname) — reached when the LIKE child's RangeVar has no explicit schema name (temp-table target or unqualified CREATE TABLE). A temp-instance probe (CREATE TEMP TABLE cf6482_temp (LIKE cf6482_src INCLUDING COMMENTS)) drove line 1647 from 0 to 1 hit, confirming coverability.
• Test example: cf6482_temp case in cf6482_like_comments.

Uncovered branch

Line 1643 branch 1 (false outcome of if (heapRel->schemaname)):

• Count: 0
• Coverability: coverable-now
• Reason: Taken when heapRel->schemaname == NULL (temp-table / unqualified target). Covered by the same cf6482_temp test case.

Touched functions

Non-instrumented changes

None. The diff touches only the implementation .c (instrumented) and test .sql/.out (test files, not counted as executable implementation).

Residual limitations

• Measurement is qualified as "under the executed test set" (create_table_like only); not a global suite-coverage figure.
• No cannot-measure and no infeasible-with-reason waivers apply.

pgsql-hackers thread ledger

Status: reconciled Ledger report slug: commitfest-6482-review-handoff-patch-reviewer

Thread: "COMMENTS are not being copied in CREATE TABLE LIKE" (Thread-ID 636704), Jim Jones (author), Feb-Jul 2026.

Message chronology:

1. e08cb97f-...@uni-muenster.de (2026-02-12, patch) — Original bug report + v1 patch.
2. CAHGQGwE...@mail.gmail.com (2026-02-13, review) — Fujii Masao: multi-LIKE semantics question.
3. 5888209c-...@uni-muenster.de (2026-02-13, author-reply) — Author proposes newline-separated concatenation; v2 patch.
4. ceccea29-...@uni-muenster.de (2026-02-26, patch) — Rebase (v3).
5. CAL-Pge4e...@mail.gmail.com (2026-03-24, review) — Carlos: tests OK, requests docs update, proposes concatenation wording.
6. fda18ba2-...@uni-muenster.de (2026-03-25, author-reply) — Author confirms docs intent; asks reviewers to weigh in.
7. tencent_D0A4135E...@qq.com (2026-07-18, review) — ZizhuanLiu: suggests skip-on-multiple.
8. 88b16128-...@uni-muenster.de (2026-07-23, author-reply) — Author disagrees with blanket skip; open to withdrawing if concatenation impractical.

Per-finding dispositions:

• F1: new (not previously raised on-list)
• F2: already-raised (Carlos 2026-03-24)
• F3: unreconciled (actively debated, no consensus)
• F4: new
• F5: new
• F6: new
• COV-1: new

Aggregated Test Plan

The proposed cf6482_like_comments regression test (proposal-only, not applied to the reviewed worktree) exercises:

• Plain-table target (cf6482_plain)
• INCLUDING ALL with a commented source (cf6482_all)
• Partitioned-table target (cf6482_part)
• Temp-table target (cf6482_temp) — covers COV-1 line 1647
• Foreign-table target (cf6482_ft) — exposes/covers F1 after the fix
• Foreign-table INCLUDING ALL (cf6482_ft_all)
• Multi-LIKE last-wins (cf6482_multi) — pins current F3 behavior

Validated via make -C src/test/regress check-tests TESTS=cf6482_like_comments (All tests passed) against the patched disposable build.

Residual risk

• F1 is a functional regression that must be fixed before committer-ready (foreign-table LIKE target fails).
• F3 multi-LIKE semantics is unresolved on-list; the staged diff's silent last-wins may not match the committer's chosen contract.
• F2 docs gap should be addressed before commit.
• F4/F5/COV-1 are low-severity nits/test gaps addressed by the proposed repair patch.

Deterministic verdict

needs-review

No BLOCKERs. F1 is a correctness MAJOR (functional regression on foreign-table target). F3 is an unreconciled live thread objection (multi-LIKE semantics). Per the harness verdict rules: any correctness MAJOR or live thread objection => needs-review.

Blocker split: 0 verified / 0 unverified.

Maintainer CC

Author: Jim Jones (jim.jones@uni-muenster.de) Reviewer: Carlos, Fujii Masao (masao.fujii@oss.nttdata.com)

Per-commit rollup

Single uncommitted staged diff (no commit history); all changes are in the working-tree index of the review worktree. Diff SHA: d9b8108fc514bf0326770c6bf9a3058abd033fd6e7eb04d93712042f0fbf713f.

Verified-clean section

No findings were refuted as actual defects requiring repair except F6 (deferral-mask change, refuted as benign). The security review found no security vulnerabilities.

---

*Review conducted by the pg-review-harness (model glm-5.2:cloud). Artifact validated with validate_review_artifact.py (sha256 edbcb304...), exit 0. Reviewed worktree was read-only; all candidate repairs were validated in disposable scratch only.*