MDEV-11030 Assertion `precision > 0' failed in decimal_bin_size

Fixing Item::decimal_precision() to return at least one digit.
This fixes the problem reported in MDEV.

Also, fixing Item_func_signed::fix_length_and_dec() to reserve
space for at least one digit (plus one character for an optional sign).
This is needed to have CONVERT(expr,SIGNED) and CONVERT(expr,UNSIGNED)
create correct string fields when they appear in string context, e.g.:
  CREATE TABLE t1 AS SELECT CONCAT(CONVERT('',SIGNED));

Author: Alexander Barkov

mysql-test/r/cast.result

@@ -583,7 +583,7 @@ show create table t1;
 Table	Create Table
 t1	CREATE TABLE `t1` (
   `cast(1 as unsigned)` int(1) unsigned NOT NULL,
-  `cast(1 as signed)` int(1) NOT NULL,
+  `cast(1 as signed)` int(2) NOT NULL,
   `cast(1 as double(5,2))` double(5,2) DEFAULT NULL,
   `cast(1 as decimal(5,3))` decimal(5,3) NOT NULL,
   `cast("A" as binary)` varbinary(1) NOT NULL,
@@ -819,3 +819,74 @@ utf8_bin
 select collation(cast("a" as char(10) binary ascii));
 collation(cast("a" as char(10) binary ascii))
 latin1_bin
+#
+# MDEV-11030 Assertion `precision > 0' failed in decimal_bin_size
+#
+SELECT * FROM (SELECT IFNULL(CONVERT(NULL, UNSIGNED), NULL)) sq;
+IFNULL(CONVERT(NULL, UNSIGNED), NULL)
+NULL
+CREATE TABLE t1 AS SELECT IFNULL(CONVERT(NULL, UNSIGNED), NULL);
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `IFNULL(CONVERT(NULL, UNSIGNED), NULL)` decimal(1,0) DEFAULT NULL
+) ENGINE=MyISAM DEFAULT CHARSET=latin1
+DROP TABLE t1;
+CREATE TABLE t1 AS SELECT COALESCE(CONVERT(NULL, UNSIGNED), NULL);
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `COALESCE(CONVERT(NULL, UNSIGNED), NULL)` decimal(1,0) DEFAULT NULL
+) ENGINE=MyISAM DEFAULT CHARSET=latin1
+DROP TABLE t1;
+CREATE TABLE t1 AS SELECT CASE WHEN TRUE THEN CONVERT(NULL, UNSIGNED) ELSE NULL END;
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `CASE WHEN TRUE THEN CONVERT(NULL, UNSIGNED) ELSE NULL END` decimal(1,0) DEFAULT NULL
+) ENGINE=MyISAM DEFAULT CHARSET=latin1
+DROP TABLE t1;
+CREATE TABLE t1 AS SELECT IFNULL(CONVERT(NULL,SIGNED),CONVERT(NULL,UNSIGNED)) AS a;
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `a` decimal(1,0) DEFAULT NULL
+) ENGINE=MyISAM DEFAULT CHARSET=latin1
+DROP TABLE t1;
+CREATE TABLE t1 AS SELECT
+-1,
+CONVERT(NULL,SIGNED),
+CONCAT(CONVERT(NULL,SIGNED)),
+1,
+CONVERT(NULL,UNSIGNED),
+CONCAT(CONVERT(NULL,UNSIGNED));
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `-1` int(2) NOT NULL,
+  `CONVERT(NULL,SIGNED)` int(2) DEFAULT NULL,
+  `CONCAT(CONVERT(NULL,SIGNED))` varchar(2) DEFAULT NULL,
+  `1` int(1) NOT NULL,
+  `CONVERT(NULL,UNSIGNED)` int(1) unsigned DEFAULT NULL,
+  `CONCAT(CONVERT(NULL,UNSIGNED))` varchar(1) DEFAULT NULL
+) ENGINE=MyISAM DEFAULT CHARSET=latin1
+DROP TABLE t1;
+CREATE TABLE t1 AS SELECT
+CONVERT('',SIGNED),
+CONCAT(CONVERT('',SIGNED)),
+CONVERT('',UNSIGNED),
+CONCAT(CONVERT('',UNSIGNED));
+Warnings:
+Warning	1292	Truncated incorrect INTEGER value: ''
+Warning	1292	Truncated incorrect INTEGER value: ''
+Warning	1292	Truncated incorrect INTEGER value: ''
+Warning	1292	Truncated incorrect INTEGER value: ''
+SHOW CREATE TABLE t1;
+Table	Create Table
+t1	CREATE TABLE `t1` (
+  `CONVERT('',SIGNED)` int(2) NOT NULL,
+  `CONCAT(CONVERT('',SIGNED))` varchar(2) NOT NULL,
+  `CONVERT('',UNSIGNED)` int(1) unsigned NOT NULL,
+  `CONCAT(CONVERT('',UNSIGNED))` varchar(1) NOT NULL
+) ENGINE=MyISAM DEFAULT CHARSET=latin1
+DROP TABLE t1;

mysql-test/t/cast.test

@@ -472,3 +472,42 @@ select collation(cast("a" as char(10) ascii binary));
 select collation(cast("a" as char(10) binary charset utf8));
 select collation(cast("a" as char(10) binary ascii));
 
+--echo #
+--echo # MDEV-11030 Assertion `precision > 0' failed in decimal_bin_size
+--echo #
+
+SELECT * FROM (SELECT IFNULL(CONVERT(NULL, UNSIGNED), NULL)) sq;
+
+CREATE TABLE t1 AS SELECT IFNULL(CONVERT(NULL, UNSIGNED), NULL);
+SHOW CREATE TABLE t1;
+DROP TABLE t1;
+
+CREATE TABLE t1 AS SELECT COALESCE(CONVERT(NULL, UNSIGNED), NULL);
+SHOW CREATE TABLE t1;
+DROP TABLE t1;
+
+CREATE TABLE t1 AS SELECT CASE WHEN TRUE THEN CONVERT(NULL, UNSIGNED) ELSE NULL END;
+SHOW CREATE TABLE t1;
+DROP TABLE t1;
+
+CREATE TABLE t1 AS SELECT IFNULL(CONVERT(NULL,SIGNED),CONVERT(NULL,UNSIGNED)) AS a;
+SHOW CREATE TABLE t1;
+DROP TABLE t1;
+
+CREATE TABLE t1 AS SELECT
+  -1,
+  CONVERT(NULL,SIGNED),
+  CONCAT(CONVERT(NULL,SIGNED)),
+  1,
+  CONVERT(NULL,UNSIGNED),
+  CONCAT(CONVERT(NULL,UNSIGNED));
+SHOW CREATE TABLE t1;
+DROP TABLE t1;
+
+CREATE TABLE t1 AS SELECT
+  CONVERT('',SIGNED),
+  CONCAT(CONVERT('',SIGNED)),
+  CONVERT('',UNSIGNED),
+  CONCAT(CONVERT('',UNSIGNED));
+SHOW CREATE TABLE t1;
+DROP TABLE t1;

sql/item.cc

@@ -533,7 +533,14 @@ uint Item::decimal_precision() const
                                      unsigned_flag);
     return MY_MIN(prec, DECIMAL_MAX_PRECISION);
   }
-  return MY_MIN(max_char_length(), DECIMAL_MAX_PRECISION);
+  uint res= max_char_length();
+  /*
+    Return at least one decimal digit, even if Item::max_char_length()
+    returned  0. This is important to avoid attempts to create fields of types
+    INT(0) or DECIMAL(0,0) when converting NULL or empty strings to INT/DECIMAL:
+      CREATE TABLE t1 AS SELECT CONVERT(NULL,SIGNED) AS a;
+  */
+  return res ? MY_MIN(res, DECIMAL_MAX_PRECISION) : 1;
 }
 
 

sql/item_func.h

@@ -626,8 +626,15 @@ class Item_func_signed :public Item_int_func
   longlong val_int_from_str(int *error);
   void fix_length_and_dec()
   {
-    fix_char_length(MY_MIN(args[0]->max_char_length(),
-                           MY_INT64_NUM_DECIMAL_DIGITS));
+    uint32 char_length= MY_MIN(args[0]->max_char_length(),
+                               MY_INT64_NUM_DECIMAL_DIGITS);
+    /*
+      args[0]->max_char_length() can return 0.
+      Reserve max_length to fit at least one character for one digit,
+      plus one character for the sign (if signed).
+    */
+    set_if_bigger(char_length, 1 + (unsigned_flag ? 0 : 1));
+    fix_char_length(char_length);
   }
   virtual void print(String *str, enum_query_type query_type);
   uint decimal_precision() const { return args[0]->decimal_precision(); }