Select handler returns wrong representative byte for GROUP BY/DISTINCT on CHAR(1) latin1 (with columnstore_innodb_queries_use_mcs=ON)

Summary

GROUP BY / DISTINCT on a CHAR(1) latin1 column returns the wrong representative byte through the ColumnStore Select Handler. Counts (collation-aware) are correct, the byte shown to the client is not.

Repro (standalone)

CREATE TABLE t (c1 CHAR(1) CHARACTER SET latin1 COLLATE latin1_swedish_ci) ENGINE=ColumnStore;

SET NAMES utf8;

INSERT INTO t VALUES ('a'),('A'),('ä'),('Ä'),('ã'),('Ã');

SELECT c1, HEX(c1), COUNT(*) FROM t GROUP BY c1 ORDER BY c1;

Storage is correct (SELECT c1, HEX(c1) returns 61/41/E4/C4/E3/C3 either way). Only the GROUP BY representative emitted by the handler is wrong.

In the full CI basic suite the symptom is amplified to 4 distinct groups with garbled bytes — same root cause.

Root cause (suspected)

dbcon/mysql/ha_mcs_select_handler*.cpp GROUP BY result-row materialiser: for a fixed-width latin1 CHAR column, the representative byte is taken from the original (utf8mb3) AST literal instead of the latin1 byte from the row.

Workaround

SET SESSION columnstore_select_handler=OFF;

Trigger / status

Pre-existing bug. Was exposed in CI by columnstore commit 09783833b (2026-04-17, which set innodb_queries_use_mcs=ON globally via run_mtr.sh). That approach was reverted in MCOL-6340 / PR #3989, which scopes the variable back to the future suite via suite.opt. As a result:

• basic.ctype_cmp_char1_latin1_swedish_ci is not disabled and passes in CI under the standard suite configuration.
• The bug is still reproducible — either via the standalone repro above, or by re-enabling innodb_queries_use_mcs=ON globally and re-running the MTR test.

Affected MTR test

columnstore/basic.ctype_cmp_char1_latin1_swedish_ci (artifacts attached: ctype_artifacts.tgz).