InnoDB: Fixed bugs in the padding and trimming of trailing spaces

that affected the UCS2 character set. (Bug #7350)


innobase/data/data0type.c:
  Added dtype_get_charset_coll_noninline()
innobase/include/data0type.h:
  Added dtype_get_charset_coll_noninline()
innobase/include/row0mysql.h:
  Added charset field to mysql_row_templ_struct.
innobase/include/row0mysql.ic:
  row_mysql_store_col_in_innobase_format():
  When removing trailing spaces, treat the UCS2 character set properly.
innobase/rem/rem0cmp.c:
  cmp_whole_field(): Do not remove trailing 0x20 bytes, as
  innobase_mysql_cmp() implicitly pads the strings with trailing
  spaces as necessary.
innobase/row/row0sel.c:
  row_sel_field_store_in_mysql_format(): Do not pad with 0x20 bytes.
  row_sel_store_mysql_rec(): Pad VARCHARs with trailing spaces
  (0x20, or 0x0020 in UCS2).
sql/ha_innodb.cc:
  build_template(): Initialize templ->charset

1 files changed, 62 insertions, 6 deletions

diff --git a/innobase/row/row0sel.c b/innobase/row/row0sel.c
index ce76f48e7a7..61ba0b53172 100644
--- a/innobase/row/row0sel.c
+++ b/innobase/row/row0sel.c
@@ -2204,9 +2204,6 @@ row_sel_field_store_in_mysql_format(
 		dest = row_mysql_store_var_len(dest, len);
 		ut_memcpy(dest, data, len);
 
-		/* Pad with trailing spaces */
-		memset(dest + len, ' ', col_len - len); 
-
 		/* ut_ad(col_len >= len + 2); No real var implemented in
 		MySQL yet! */
 		
@@ -2334,7 +2331,45 @@ row_sel_store_mysql_rec(
 				mysql_rec + templ->mysql_col_offset,
 				templ->mysql_col_len, data, len,
 				templ->type, templ->is_unsigned);
-				
+
+			if (templ->type == DATA_VARCHAR
+					|| templ->type == DATA_VARMYSQL
+					|| templ->type == DATA_BINARY) {
+				/* Pad with trailing spaces */
+				data = mysql_rec + templ->mysql_col_offset;
+
+				/* Handle UCS2 strings differently.  As no new
+				collations will be introduced in 4.1, we
+				hardcode the charset-collation codes here.
+				5.0 will use a different approach. */
+				if (templ->charset == 35
+						|| templ->charset == 90
+						|| (templ->charset >= 128
+						&& templ->charset <= 144)) {
+					/* space=0x0020 */
+					ulint	col_len = templ->mysql_col_len;
+
+					ut_a(!(col_len & 1));
+					if (len & 1) {
+						/* A 0x20 has been stripped
+						from the column.
+						Pad it back. */
+						goto pad_0x20;
+					}
+					/* Pad the rest of the string
+					with 0x0020 */
+					while (len < col_len) {
+						data[len++] = 0x00;
+					pad_0x20:
+						data[len++] = 0x20;
+					}
+				} else {
+					/* space=0x20 */
+					memset(data + len, 0x20,
+						templ->mysql_col_len - len);
+				}
+			}
+
 			/* Cleanup */
 			if (extern_field_heap) {
  				mem_heap_free(extern_field_heap);
@@ -2368,8 +2403,29 @@ row_sel_store_mysql_rec(
 				pad_char = '\0';
 			}
 
-			memset(mysql_rec + templ->mysql_col_offset, pad_char,
-							templ->mysql_col_len);
+			/* Handle UCS2 strings differently.  As no new
+			collations will be introduced in 4.1,
+			we hardcode the charset-collation codes here.
+			5.0 will use a different approach. */
+			if (templ->charset == 35
+					|| templ->charset == 90
+					|| (templ->charset >= 128
+					&& templ->charset <= 144)) {
+				/* There are two bytes per char, so the length
+				has to be an even number. */
+				ut_a(!(templ->mysql_col_len & 1));
+				data = mysql_rec + templ->mysql_col_offset;
+				len = templ->mysql_col_len;
+				/* Pad with 0x0020. */
+				while (len >= 2) {
+					*data++ = 0x00;
+					*data++ = 0x20;
+					len -= 2;
+				}
+			} else {
+				memset(mysql_rec + templ->mysql_col_offset,
+					pad_char, templ->mysql_col_len);
+			}
 		}
 	}