MDEV-10743 LDML: a new syntax to reuse sort order from another 8bit s…

…imple collation

mysql-test/r/ctype_ldml.result

@@ -454,8 +454,12 @@ select "foo" = "foo " collate latin1_test;
 The following tests check that two-byte collation IDs work
 select * from information_schema.collations where id>256 and is_compiled<>'Yes' order by id;
 COLLATION_NAME	CHARACTER_SET_NAME	ID	IS_DEFAULT	IS_COMPILED	SORTLEN
+ascii2_bin2	ascii2	319			1
 ascii2_general_ci	ascii2	320	Yes		1
 ascii2_bin	ascii2	321			1
+ascii2_general_inherited_ci	ascii2	322			1
+ascii2_general_inherited2_ci	ascii2	323			1
+ascii2_badly_inherited_ci	ascii2	324			1
 utf8mb4_test_ci	utf8mb4	326			8
 utf16_test_ci	utf16	327			8
 utf8mb4_test_400_ci	utf8mb4	328			8
@@ -1185,20 +1189,62 @@ ch	w	ducet
 DROP TABLE t1;
 #
 # Testing that the MY_CS_PUREASCII flag is set properly
+# Comparison between ascii2 and latin1 should not give "illegal collation error"
 #
 CREATE TABLE t1 (a VARCHAR(10) CHARACTER SET ascii2, b VARCHAR(10) CHARACTER SET latin1);
 INSERT INTO t1 VALUES ('a','a'),('b','b');
 SELECT * FROM t1 WHERE a=b;
 a	b
 a	a
 b	b
+ALTER TABLE t1 MODIFY a VARCHAR(10) CHARACTER SET ascii2 COLLATE ascii2_bin2;
+SELECT * FROM t1 WHERE a=b;
+a	b
+a	a
+b	b
 ALTER TABLE t1 MODIFY a VARCHAR(10) CHARACTER SET ascii2 COLLATE ascii2_bin;
 SELECT * FROM t1 WHERE a=b;
 a	b
 a	a
 b	b
+ALTER TABLE t1 MODIFY a VARCHAR(10) CHARACTER SET ascii2 COLLATE ascii2_general_inherited_ci;
+SELECT * FROM t1 WHERE a=b;
+a	b
+a	a
+b	b
+ALTER TABLE t1 MODIFY a VARCHAR(10) CHARACTER SET ascii2 COLLATE ascii2_general_inherited2_ci;
+SELECT * FROM t1 WHERE a=b;
+a	b
+a	a
+b	b
 DROP TABLE t1;
 #
+# Testing that in case of two binary collations
+# "BINARY" in a column definition uses the collation with the least id
+#
+CREATE TABLE t1 (a VARCHAR(10) CHARACTER SET ascii2 BINARY);
+INSERT INTO t1 VALUES ('test');
+SELECT COLLATION(a) FROM t1;
+COLLATION(a)
+ascii2_bin2
+DROP TABLE t1;
+#
+# Testing mixing of two binary collations of the same character set
+#
+CREATE TABLE t1 (
+a VARCHAR(10) CHARACTER SET ascii2 COLLATE ascii2_bin,
+b VARCHAR(10) CHARACTER SET ascii2 COLLATE ascii2_bin2
+);
+INSERT INTO t1 VALUES ('a','a');
+SELECT * FROM t1 WHERE a=b;
+ERROR HY000: Illegal mix of collations (ascii2_bin,IMPLICIT) and (ascii2_bin2,IMPLICIT) for operation '='
+DROP TABLE t1;
+#
+# Testing bad collation inheritance
+#
+CREATE TABLE t1 (a VARCHAR(10) CHARACTER SET ascii2 COLLATE ascii2_badly_inherited_ci);
+ERROR HY000: Unknown collation: 'ascii2_badly_inherited_ci'
+#
 # Testing that the MY_CS_CSSORT flag is set properly
 #
 CREATE TABLE t1 (a VARCHAR(10) CHARACTER SET latin1 COLLATE latin1_test);

mysql-test/std_data/ldml/Index.xml

@@ -319,8 +319,29 @@
   </charset>
 
   <charset name="ascii2">
+    <!--
+      Notes:
+      - ascii2 has two collations with "binary" flag.
+        ctype_ldml.test makes sure that
+          CREATE TABLE t1 (a VARCHAR(10) CHARACTER SET ascii2 BINARY);
+        uses ascii2_bin2, which is the collation with the least ID.
+      - ascii2_general_inherited_ci inherits sort order in ascii2.xml
+      - ascii2_genegal_inherited2_ci inherits sort order directly in this file.
+    -->
+    <collation name="ascii2_bin2" id="319" flag="binary"/>
     <collation name="ascii2_general_ci"	id="320" flag="primary"/>
     <collation name="ascii2_bin" id="321" flag="binary"/>
+    <collation name="ascii2_general_inherited_ci" id="322"/>
+    <collation name="ascii2_general_inherited2_ci" id="323">
+      <rules>
+        <import source="ascii2_general_ci"/>
+      </rules>
+    </collation>
+    <collation name="ascii2_badly_inherited_ci" id="324">
+      <rules>
+        <import source="ascii2_non_existing_ci"/>
+      </rules>
+    </collation>
   </charset>
 
   <charset name="latin1">

mysql-test/std_data/ldml/ascii2.xml

@@ -116,6 +116,12 @@
 
 <collation name="ascii2_bin"	flag="binary"/>
 
+<collation name="ascii2_general_inherited_ci">
+  <rules>
+    <import source="ascii2_general_ci"/>
+  </rules>
+</collation>
+
 </charset>
 
 </charsets>

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

@@ -390,8 +390,12 @@ select "foo" = "foo " collate latin1_test;
 The following tests check that two-byte collation IDs work
 select * from information_schema.collations where id>256 and is_compiled<>'Yes' order by id;
 COLLATION_NAME	CHARACTER_SET_NAME	ID	IS_DEFAULT	IS_COMPILED	SORTLEN
+ascii2_bin2	ascii2	319			1
 ascii2_general_ci	ascii2	320	Yes		1
 ascii2_bin	ascii2	321			1
+ascii2_general_inherited_ci	ascii2	322			1
+ascii2_general_inherited2_ci	ascii2	323			1
+ascii2_badly_inherited_ci	ascii2	324			1
 utf8mb4_test_ci	utf8mb4	326			8
 utf16_test_ci	utf16	327			8
 utf8mb4_test_400_ci	utf8mb4	328			8

mysql-test/t/ctype_ldml.test

@@ -413,16 +413,48 @@ DROP TABLE t1;
 
 --echo #
 --echo # Testing that the MY_CS_PUREASCII flag is set properly
+--echo # Comparison between ascii2 and latin1 should not give "illegal collation error"
 --echo #
 CREATE TABLE t1 (a VARCHAR(10) CHARACTER SET ascii2, b VARCHAR(10) CHARACTER SET latin1);
 INSERT INTO t1 VALUES ('a','a'),('b','b');
-# should not give "illegal collation" error
+SELECT * FROM t1 WHERE a=b;
+ALTER TABLE t1 MODIFY a VARCHAR(10) CHARACTER SET ascii2 COLLATE ascii2_bin2;
 SELECT * FROM t1 WHERE a=b;
 ALTER TABLE t1 MODIFY a VARCHAR(10) CHARACTER SET ascii2 COLLATE ascii2_bin;
-# should not give "illegal collation" error
+SELECT * FROM t1 WHERE a=b;
+ALTER TABLE t1 MODIFY a VARCHAR(10) CHARACTER SET ascii2 COLLATE ascii2_general_inherited_ci;
+SELECT * FROM t1 WHERE a=b;
+ALTER TABLE t1 MODIFY a VARCHAR(10) CHARACTER SET ascii2 COLLATE ascii2_general_inherited2_ci;
+SELECT * FROM t1 WHERE a=b;
+DROP TABLE t1;
+
+--echo #
+--echo # Testing that in case of two binary collations
+--echo # "BINARY" in a column definition uses the collation with the least id
+--echo #
+CREATE TABLE t1 (a VARCHAR(10) CHARACTER SET ascii2 BINARY);
+INSERT INTO t1 VALUES ('test');
+SELECT COLLATION(a) FROM t1;
+DROP TABLE t1;
+
+
+--echo #
+--echo # Testing mixing of two binary collations of the same character set
+--echo #
+CREATE TABLE t1 (
+ a VARCHAR(10) CHARACTER SET ascii2 COLLATE ascii2_bin,
+ b VARCHAR(10) CHARACTER SET ascii2 COLLATE ascii2_bin2
+);
+INSERT INTO t1 VALUES ('a','a');
+--error ER_CANT_AGGREGATE_2COLLATIONS
 SELECT * FROM t1 WHERE a=b;
 DROP TABLE t1;
 
+--echo #
+--echo # Testing bad collation inheritance
+--echo #
+--error ER_UNKNOWN_COLLATION
+CREATE TABLE t1 (a VARCHAR(10) CHARACTER SET ascii2 COLLATE ascii2_badly_inherited_ci);
 
 --echo #
 --echo # Testing that the MY_CS_CSSORT flag is set properly

mysys/charset.c

@@ -197,13 +197,55 @@ static int cs_copy_data(struct charset_info_st *to, CHARSET_INFO *from)
 }
 
 
+static my_bool simple_8bit_charset_data_is_full(CHARSET_INFO *cs)
+{
+  return cs->ctype && cs->to_upper && cs->to_lower && cs->tab_to_uni;
+}
+
+
+/**
+  Inherit missing 8bit charset data from another collation.
+  Arrays pointed by refcs must be in the permanent memory already,
+  e.g. static memory, or allocated by my_once_xxx().
+*/
+static void
+inherit_charset_data(struct charset_info_st *cs, CHARSET_INFO *refcs)
+{
+  if (!cs->to_upper)
+    cs->to_upper= refcs->to_upper;
+  if (!cs->to_lower)
+    cs->to_lower= refcs->to_lower;
+  if (!cs->ctype)
+    cs->ctype= refcs->ctype;
+  if (!cs->tab_to_uni)
+    cs->tab_to_uni= refcs->tab_to_uni;
+}
+
+
+static my_bool simple_8bit_collation_data_is_full(CHARSET_INFO *cs)
+{
+  return cs->sort_order || (cs->state & MY_CS_BINSORT);
+}
+
+
+/**
+  Inherit 8bit simple collation data from another collation.
+  refcs->sort_order must be in the permanent memory already,
+  e.g. static memory, or allocated by my_once_xxx().
+*/
+static void
+inherit_collation_data(struct charset_info_st *cs, CHARSET_INFO *refcs)
+{
+  if (!simple_8bit_collation_data_is_full(cs))
+    cs->sort_order= refcs->sort_order;
+}
+
 
 static my_bool simple_cs_is_full(CHARSET_INFO *cs)
 {
-  return ((cs->csname && cs->tab_to_uni && cs->ctype && cs->to_upper &&
-	   cs->to_lower) &&
-	  (cs->number && cs->name &&
-	  (cs->sort_order || (cs->state & MY_CS_BINSORT) )));
+  return  cs->number && cs->csname && cs->name &&
+          simple_8bit_charset_data_is_full(cs) &&
+          (simple_8bit_collation_data_is_full(cs) || cs->tailoring);
 }
 
 
@@ -336,7 +378,7 @@ static int add_collation(struct charset_info_st *cs)
     cs->name= NULL;
     cs->state= 0;
     cs->sort_order= NULL;
-    cs->state= 0;
+    cs->tailoring= NULL;
   }
   return MY_XML_OK;
 }
@@ -631,6 +673,39 @@ const char *get_charset_name(uint charset_number)
 }
 
 
+static CHARSET_INFO *inheritance_source_by_id(CHARSET_INFO *cs, uint refid)
+{
+  CHARSET_INFO *refcs;
+  return refid && refid != cs->number &&
+         (refcs= all_charsets[refid]) &&
+         (refcs->state & MY_CS_AVAILABLE) ? refcs : NULL;
+}
+
+
+static CHARSET_INFO *find_collation_data_inheritance_source(CHARSET_INFO *cs)
+{
+  const char *beg, *end;
+  if (cs->tailoring &&
+      !strncmp(cs->tailoring, "[import ", 8) &&
+      (end= strchr(cs->tailoring + 8, ']')) &&
+      (beg= cs->tailoring + 8) + MY_CS_NAME_SIZE > end)
+  {
+    char name[MY_CS_NAME_SIZE + 1];
+    memcpy(name, beg, end - beg);
+    name[end - beg]= '\0';
+    return inheritance_source_by_id(cs, get_collation_number(name));
+  }
+  return NULL;
+}
+
+
+static CHARSET_INFO *find_charset_data_inheritance_source(CHARSET_INFO *cs)
+{
+  uint refid= get_charset_number_internal(cs->csname, MY_CS_PRIMARY);
+  return inheritance_source_by_id(cs, refid);
+}
+
+
 static CHARSET_INFO *
 get_internal_charset(MY_CHARSET_LOADER *loader, uint cs_number, myf flags)
 {
@@ -665,6 +740,19 @@ get_internal_charset(MY_CHARSET_LOADER *loader, uint cs_number, myf flags)
     {
       if (!(cs->state & MY_CS_READY))
       {
+        if (!simple_8bit_charset_data_is_full(cs))
+        {
+          CHARSET_INFO *refcs= find_charset_data_inheritance_source(cs);
+          if (refcs)
+            inherit_charset_data(cs, refcs);
+        }
+        if (!simple_8bit_collation_data_is_full(cs))
+        {
+          CHARSET_INFO *refcl= find_collation_data_inheritance_source(cs);
+          if (refcl)
+            inherit_collation_data(cs, refcl);
+        }
+
         if ((cs->cset->init && cs->cset->init(cs, loader)) ||
             (cs->coll->init && cs->coll->init(cs, loader)))
         {

sql/item.cc

@@ -2078,6 +2078,9 @@ bool DTCollation::aggregate(const DTCollation &dt, uint flags)
         set(0, DERIVATION_NONE, 0);
         return 1;
       }
+      if (collation->state & MY_CS_BINSORT &&
+          dt.collation->state & MY_CS_BINSORT)
+        return 1;
       if (collation->state & MY_CS_BINSORT)
         return 0;
       if (dt.collation->state & MY_CS_BINSORT)