MDEV-25440: Indexed CHAR columns are broken with NO_PAD collations

cmp_data(): Compare different-length CHAR fields with
the new strnncollsp_nchars function that will pad spaces if needed.

Any InnoDB ROW_FORMAT except the original one that was named
ROW_FORMAT=REDUNDANT in MySQL 5.0.3 will internally store
CHAR(n) columns as variable-length if the character encoding is
variable length. Spaces may be trimmed from the end.
For NOT NULL values, the minimum length is always n*mbminlen.
In cmp_data() we only know the lengths in bytes and we cannot
easily know the ROW_FORMAT.

is_strnncoll_compatible(): Refactored from innobase_mysql_cmp().

innobase_mysql_cmp(): Merged to cmp_whole_field().

cmp_whole_field(): Invoke strnncollsp_nchars for the DATA_MYSQL
(the CHAR type with any other collation than latin1_swedish_ci).

Reviewed by: Alexander Barkov
Tested by: Roel Roel Van de Paar

mysql-test/suite/innodb/r/no_pad.result

+CREATE TABLE t1 (a CHAR(8), id INT, PRIMARY KEY (a,id)) COLLATE utf8_nopad_bin
+ENGINE=InnoDB ROW_FORMAT=REDUNDANT;
+INSERT INTO t1 VALUES ('',1);
+ALTER TABLE t1 ROW_FORMAT=DYNAMIC;
+INSERT INTO t1 VALUES ('',2);
+ALTER TABLE t1 ROW_FORMAT=REDUNDANT;
+DROP TABLE t1;

mysql-test/suite/innodb/t/no_pad.test

+--source include/have_innodb.inc
+
+CREATE TABLE t1 (a CHAR(8), id INT, PRIMARY KEY (a,id)) COLLATE utf8_nopad_bin
+ENGINE=InnoDB ROW_FORMAT=REDUNDANT;
+
+INSERT INTO t1 VALUES ('',1);
+ALTER TABLE t1 ROW_FORMAT=DYNAMIC;
+INSERT INTO t1 VALUES ('',2);
+ALTER TABLE t1 ROW_FORMAT=REDUNDANT;
+DROP TABLE t1;

storage/innobase/rem/rem0cmp.cc

 /*****************************************************************************
 
 Copyright (c) 1994, 2019, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2020, 2021, MariaDB Corporation.
+Copyright (c) 2020, 2022, MariaDB Corporation.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -49,25 +49,11 @@ At the present, the comparison functions return 0 in the case,
 where two records disagree only in the way that one
 has more fields than the other. */
 
-/** Compare two data fields.
-@param[in] prtype precise type
-@param[in] a data field
-@param[in] a_length length of a, in bytes (not UNIV_SQL_NULL)
-@param[in] b data field
-@param[in] b_length length of b, in bytes (not UNIV_SQL_NULL)
-@return positive, 0, negative, if a is greater, equal, less than b,
-respectively */
-UNIV_INLINE
-int
-innobase_mysql_cmp(
-	ulint		prtype,
-	const byte*	a,
-	unsigned int	a_length,
-	const byte*	b,
-	unsigned int	b_length)
+#ifndef DBUG_OFF
+/** @return whether a data type is compatible with strnncoll() functions */
+static bool is_strnncoll_compatible(ulint type)
 {
-#ifdef UNIV_DEBUG
-	switch (prtype & DATA_MYSQL_TYPE_MASK) {
+  switch (type) {
   case MYSQL_TYPE_BIT:
   case MYSQL_TYPE_STRING:
   case MYSQL_TYPE_VAR_STRING:
@@ -76,22 +62,12 @@ innobase_mysql_cmp(
   case MYSQL_TYPE_BLOB:
   case MYSQL_TYPE_LONG_BLOB:
   case MYSQL_TYPE_VARCHAR:
-		break;
+    return true;
   default:
-		ut_error;
-	}
-#endif /* UNIV_DEBUG */
-
-	uint cs_num = (uint) dtype_get_charset_coll(prtype);
-
-	if (CHARSET_INFO* cs = get_charset(cs_num, MYF(MY_WME))) {
-		return(cs->coll->strnncollsp(
-			       cs, a, a_length, b, b_length));
+    return false;
   }
-
-	ib::fatal() << "Unable to find charset-collation " << cs_num;
-	return(0);
 }
+#endif /* DBUG_OFF */
 
 /*************************************************************//**
 Returns TRUE if two columns are equal for comparison purposes.
@@ -309,68 +285,52 @@ cmp_gis_field(
 @param[in] b_length length of b, in bytes (not UNIV_SQL_NULL)
 @return positive, 0, negative, if a is greater, equal, less than b,
 respectively */
-static
-int
-cmp_whole_field(
-	ulint		mtype,
-	ulint		prtype,
-	const byte*	a,
-	unsigned int	a_length,
-	const byte*	b,
-	unsigned int	b_length)
+static int cmp_whole_field(ulint mtype, ulint prtype,
+                           const byte *a, unsigned a_length,
+                           const byte *b, unsigned b_length)
 {
-	float		f_1;
-	float		f_2;
-	double		d_1;
-	double		d_2;
-
   switch (mtype) {
+  default:
+    ib::fatal() << "Unknown data type number " << mtype;
+    return 0;
   case DATA_DECIMAL:
-		return(cmp_decimal(a, a_length, b, b_length));
+    return cmp_decimal(a, a_length, b, b_length);
   case DATA_DOUBLE:
-		d_1 = mach_double_read(a);
-		d_2 = mach_double_read(b);
-
-		if (d_1 > d_2) {
-			return(1);
-		} else if (d_2 > d_1) {
-			return(-1);
+    {
+      const double af= mach_double_read(a), bf= mach_double_read(b);
+      return af > bf ? 1 : bf > af ? -1 : 0;
     }
-
-		return(0);
-
   case DATA_FLOAT:
-		f_1 = mach_float_read(a);
-		f_2 = mach_float_read(b);
-
-		if (f_1 > f_2) {
-			return(1);
-		} else if (f_2 > f_1) {
-			return(-1);
+    {
+      const float af= mach_float_read(a), bf= mach_float_read(b);
+      return af > bf ? 1 : bf > af ? -1 : 0;
     }
-
-		return(0);
   case DATA_VARCHAR:
   case DATA_CHAR:
-		return(my_charset_latin1.coll->strnncollsp(
-			       &my_charset_latin1,
-			       a, a_length, b, b_length));
+    /* latin1_swedish_ci is treated as a special case in InnoDB.
+    Because it is a fixed-length encoding (mbminlen=mbmaxlen=1),
+    non-NULL CHAR(n) values will always occupy n bytes and we
+    can invoke strnncollsp() instead of strnncollsp_nchars(). */
+    return my_charset_latin1.coll->strnncollsp(&my_charset_latin1,
+                                               a, a_length, b, b_length);
   case DATA_BLOB:
-		if (prtype & DATA_BINARY_TYPE) {
-			ib::error() << "Comparing a binary BLOB"
-				" using a character set collation!";
-			ut_ad(0);
-		}
+    ut_ad(!(prtype & DATA_BINARY_TYPE)); /* our only caller tested this */
     /* fall through */
   case DATA_VARMYSQL:
+    DBUG_ASSERT(is_strnncoll_compatible(prtype & DATA_MYSQL_TYPE_MASK));
+    if (CHARSET_INFO *cs= get_charset(dtype_get_charset_coll(prtype),
+                                      MYF(MY_WME)))
+      return cs->coll->strnncollsp(cs, a, a_length, b, b_length);
+    break;
   case DATA_MYSQL:
-		return(innobase_mysql_cmp(prtype,
-					  a, a_length, b, b_length));
-	default:
-		ib::fatal() << "Unknown data type number " << mtype;
+    DBUG_ASSERT(is_strnncoll_compatible(prtype & DATA_MYSQL_TYPE_MASK));
+    if (CHARSET_INFO *cs= get_charset(dtype_get_charset_coll(prtype),
+                                      MYF(MY_WME)))
+      return cs->coll->strnncollsp_nchars(cs, a, a_length, b, b_length,
+                                          std::max(a_length, b_length));
   }
 
-	return(0);
+  ib::fatal() << "Unable to find charset-collation for " << prtype;
 }
 
 /** Compare two data fields.