Correction to replication of charsets in 4.1:

In mysqlbinlog, there was a problem with how we escaped the content of a string user variable.
To be perfect, we should have escaped with character_set_client. But this charset is unknown
to mysqlbinlog. So the simplest is to print the string in hex. This is unreadable but
100% safe with any charset (checked with Bar), no more need to bother with character_set_client.

mysql-test/r/rpl_charset.result

@@ -153,7 +153,7 @@ master-bin.000001	4413	Query	1	4413	use `test2`; SET ONE_SHOT CHARACTER_SET_CLIE
 master-bin.000001	4549	Query	1	4549	use `test2`; truncate table t1
 master-bin.000001	4602	Query	1	4602	use `test2`; SET ONE_SHOT CHARACTER_SET_CLIENT=8,COLLATION_CONNECTION=31,COLLATION_DATABASE=9,COLLATION_SERVER=64
 master-bin.000001	4738	Intvar	1	4738	INSERT_ID=1
-master-bin.000001	4766	User var	1	4766	@`a`=_cp850'Müller' COLLATE cp850_general_ci
+master-bin.000001	4766	User var	1	4766	@`a`=_cp850 0x4DFC6C6C6572 COLLATE cp850_general_ci
 master-bin.000001	4806	Query	1	4806	use `test2`; insert into t1 (b) values(collation(@a))
 master-bin.000001	4882	Query	1	4882	use `test2`; SET ONE_SHOT CHARACTER_SET_CLIENT=8,COLLATION_CONNECTION=31,COLLATION_DATABASE=9,COLLATION_SERVER=64
 master-bin.000001	5018	Query	1	5018	use `test2`; drop database test2

mysql-test/r/rpl_user_variables.result

@@ -86,11 +86,11 @@ slave-bin.000001	313	Query	1	313	use `test`; insert into t1 values (@i1), (@i2),
 slave-bin.000001	396	User var	2	396	@`r1`=12.5
 slave-bin.000001	439	User var	2	439	@`r2`=-12.5
 slave-bin.000001	482	Query	1	482	use `test`; insert into t1 values (@r1), (@r2)
-slave-bin.000001	551	User var	2	551	@`s1`=_latin1'This is a test' COLLATE latin1_swedish_ci
-slave-bin.000001	600	User var	2	600	@`s2`=_latin1'' COLLATE latin1_swedish_ci
-slave-bin.000001	635	User var	2	635	@`s3`=_latin1'abc\'def' COLLATE latin1_swedish_ci
-slave-bin.000001	677	User var	2	677	@`s4`=_latin1'abc\\def' COLLATE latin1_swedish_ci
-slave-bin.000001	719	User var	2	719	@`s5`=_latin1'abc\'def' COLLATE latin1_swedish_ci
+slave-bin.000001	551	User var	2	551	@`s1`=_latin1 0x5468697320697320612074657374 COLLATE latin1_swedish_ci
+slave-bin.000001	600	User var	2	600	@`s2`=_latin1 "" COLLATE latin1_swedish_ci
+slave-bin.000001	635	User var	2	635	@`s3`=_latin1 0x61626327646566 COLLATE latin1_swedish_ci
+slave-bin.000001	677	User var	2	677	@`s4`=_latin1 0x6162635C646566 COLLATE latin1_swedish_ci
+slave-bin.000001	719	User var	2	719	@`s5`=_latin1 0x61626327646566 COLLATE latin1_swedish_ci
 slave-bin.000001	761	Query	1	761	use `test`; insert into t1 values (@s1), (@s2), (@s3), (@s4), (@s5)
 slave-bin.000001	851	User var	2	851	@`n1`=NULL
 slave-bin.000001	877	Query	1	877	use `test`; insert into t1 values (@n1)
@@ -99,7 +99,7 @@ slave-bin.000001	965	Query	1	965	use `test`; insert into t1 values (@n2)
 slave-bin.000001	1027	Query	1	1027	use `test`; insert into t1 values (@a:=0), (@a:=@a+1), (@a:=@a+1)
 slave-bin.000001	1115	User var	2	1115	@`a`=2
 slave-bin.000001	1157	Query	1	1157	use `test`; insert into t1 values (@a+(@b:=@a+1))
-slave-bin.000001	1229	User var	2	1229	@`q`=_latin1'abc' COLLATE latin1_swedish_ci
+slave-bin.000001	1229	User var	2	1229	@`q`=_latin1 0x616263 COLLATE latin1_swedish_ci
 slave-bin.000001	1266	Query	1	1266	use `test`; insert t1 values (@q), (@q:=concat(@q, 'n1')), (@q:=concat(@q, 'n2'))
 slave-bin.000001	1370	User var	2	1370	@`a`=5
 slave-bin.000001	1412	Query	1	1412	use `test`; insert into t1 values (@a),(@a)

mysql-test/r/user_var.result

@@ -174,24 +174,24 @@ set @v=convert('abc' using ucs2);
 insert into t2 values (@v);
 show binlog events from 79;
 Log_name	Pos	Event_type	Server_id	Orig_log_pos	Info
-master-bin.000001	79	User var	1	79	@`a b`=_latin1'hello' COLLATE latin1_swedish_ci
+master-bin.000001	79	User var	1	79	@`a b`=_latin1 0x68656C6C6F COLLATE latin1_swedish_ci
 master-bin.000001	120	Query	1	120	use `test`; INSERT INTO t1 VALUES(@`a b`)
-master-bin.000001	184	User var	1	184	@`var1`=_latin1'\';aaa' COLLATE latin1_swedish_ci
+master-bin.000001	184	User var	1	184	@`var1`=_latin1 0x273B616161 COLLATE latin1_swedish_ci
 master-bin.000001	226	Query	1	226	use `test`; insert into t1 values (@var1)
 master-bin.000001	290	Query	1	290	use `test`; create table t2 (c char(30)) charset=ucs2
-master-bin.000001	366	User var	1	366	@`v`=_ucs2'\0a\0b\0c' COLLATE ucs2_general_ci
+master-bin.000001	366	User var	1	366	@`v`=_ucs2 0x006100620063 COLLATE ucs2_general_ci
 master-bin.000001	406	Query	1	406	use `test`; insert into t2 values (@v)
 /*!40019 SET @@session.max_insert_delayed_threads=0*/;
-SET @`a b`:=_latin1'hello' COLLATE latin1_swedish_ci;
+SET @`a b`:=_latin1 0x68656C6C6F COLLATE latin1_swedish_ci;
 use test;
 SET TIMESTAMP=10000;
 INSERT INTO t1 VALUES(@`a b`);
-SET @`var1`:=_latin1'\';aaa' COLLATE latin1_swedish_ci;
+SET @`var1`:=_latin1 0x273B616161 COLLATE latin1_swedish_ci;
 SET TIMESTAMP=10000;
 insert into t1 values (@var1);
 SET TIMESTAMP=10000;
 create table t2 (c char(30)) charset=ucs2;
-SET @`v`:=_ucs2'\0a\0b\0c' COLLATE ucs2_general_ci;
+SET @`v`:=_ucs2 0x006100620063 COLLATE ucs2_general_ci;
 SET TIMESTAMP=10000;
 insert into t2 values (@v);
 drop table t1, t2;

sql/log_event.cc

@@ -202,6 +202,29 @@ static inline int read_str(char * &buf, char *buf_end, char * &str,
   return 0;
 }
 
+/*
+  Transforms a string into "" or its expression in 0x... form.
+*/
+static char *str_to_hex(char *to, char *from, uint len)
+{
+  char *p= to;
+  if (len)
+  {
+    p= strmov(p, "0x");
+    for (uint i= 0; i < len; i++, p+= 2)
+    {
+      /* val[i] is char. Casting to uchar helps greatly if val[i] < 0 */
+      uint tmp= (uint) (uchar) from[i];
+      p[0]= _dig_vec_upper[tmp >> 4];
+      p[1]= _dig_vec_upper[tmp & 15];
+    }
+    *p= 0;
+  }
+  else
+    p= strmov(p, "\"\"");
+  return p; // pointer to end 0 of 'to'
+}
+
 
 /**************************************************************************
 	Log_event methods
@@ -2210,9 +2233,9 @@ void User_var_log_event::pack_info(Protocol* protocol)
       }
       else
       {
-        char *p= strxmov(buf + val_offset, "_", cs->csname, "'", NullS);
-        p+= escape_string_for_mysql(&my_charset_bin, p, val, val_len);
-        p= strxmov(p, "' COLLATE ", cs->name, NullS);
+        char *p= strxmov(buf + val_offset, "_", cs->csname, " ", NullS);
+        p= str_to_hex(p, val, val_len);
+        p= strxmov(p, " COLLATE ", cs->name, NullS);
         event_len= p-buf;
       }
       break;
@@ -2341,11 +2364,24 @@ void User_var_log_event::print(FILE* file, bool short_form, char* last_db)
       break;
     case STRING_RESULT:
     {
-      char *p;
-      if (!(p= (char *)my_alloca(2*val_len+1)))
+      /*
+        Let's express the string in hex. That's the most robust way. If we
+        print it in character form instead, we need to escape it with
+        character_set_client which we don't know (we will know it in 5.0, but
+        in 4.1 we don't know it easily when we are printing
+        User_var_log_event). Explanation why we would need to bother with
+        character_set_client (quoting Bar):
+        > Note, the parser doesn't switch to another unescaping mode after
+        > it has met a character set introducer.
+        > For example, if an SJIS client says something like:
+        > SET @a= _ucs2 \0a\0b'
+        > the string constant is still unescaped according to SJIS, not
+        > according to UCS2.
+      */
+      char *p, *q;
+      if (!(p= (char *)my_alloca(2*val_len+1+2))) // 2 hex digits per byte
         break; // no error, as we are 'void'
-      escape_string_for_mysql(&my_charset_bin, p, val, val_len);
-#if MYSQL_VERSION_ID < 50000
+      str_to_hex(p, val, val_len);
       /*
         For proper behaviour when mysqlbinlog|mysql, we need to explicitely
         specify the variable's collation. It will however cause problems when
@@ -2360,14 +2396,7 @@ void User_var_log_event::print(FILE* file, bool short_form, char* last_db)
         */
         fprintf(file, ":=???;\n");
       else
-        fprintf(file, ":=_%s'%s' COLLATE %s;\n", cs->csname, p, cs->name);
-#else
-      /*
-        In 5.0 we will have some SET CHARACTER_SET_ect automatically printed
-        for all events where it's needed.
-      */
-      fprintf(file, ":='%s';\n", p);
-#endif
+        fprintf(file, ":=_%s %s COLLATE %s;\n", cs->csname, p, cs->name);
       my_afree(p);
     }
       break;