[CBRD-24684] divide operation result into numeric type #4150

beyondykk9 · 2023-02-28T08:42:47Z

http://jira.cubrid.org/browse/CBRD-24684

It can see the difference in decimal point output between CUBRID and other DBMS (Tibero, Oracle).
In CUBRID, only integer values are displayed during calculation, and decimal points are output in other DBMS (Tibero, Oracle).
For this reason, we are experiencing great difficulties in migrating from other DBMS to CUBRID.
We need to modify the logic so that the result data type is numeric in case of numeric style division operation.

ctshim · 2023-05-15T22:56:05Z

src/base/system_parameter.c

@@ -728,6 +728,8 @@ static const char sysprm_ha_conf_file_name[] = "cubrid_ha.conf";

 #define PRM_NAME_STATDUMP_FORCE_ADD_INT_MAX "statdump_force_add_int_max"

+#define PRM_NAME_ORACLE_STYLE_DIVIDE "oracle_style_divide"


Should we use the word "oracle"?
How about replacing it with words like "for math" or "for science", "for engineering"?

The purpose of this parameter is compatibility with oracle.

youngjinj · 2023-05-16T09:20:16Z

src/parser/type_checking.c

@@ -6089,6 +6089,8 @@ pt_apply_expressions_definition (PARSER_CONTEXT * parser, PT_NODE ** node)
  int matches = 0, best_match = -1, i = 0;
  EXPRESSION_SIGNATURE sig;

+  static bool oracle_style_divide = prm_get_bool_value (PRM_ID_ORACLE_STYLE_DIVIDE);


Instead of the changes in this PR, how about just changing below and setting 'oracle_style_number_return=y"?

diff --git a/src/parser/type_checking.c b/src/parser/type_checking.c index d27a1891b..791d76fc4 100644 --- a/src/parser/type_checking.c +++ b/src/parser/type_checking.c @@ -10580,7 +10580,11 @@ pt_common_type (PT_TYPE_ENUM arg1_type, PT_TYPE_ENUM arg2_type) /* The common type between two ENUMs is string */ common_type = PT_TYPE_VARCHAR; } - else + else if (PT_IS_DISCRETE_NUMBER_TYPE (arg1_type) && PT_IS_DISCRETE_NUMBER_TYPE (arg1_type)) + { + common_type = PT_TYPE_NUMERIC; + } + else { common_type = arg1_type; }

csql> ;get oracle_style_number_return /* oracle_style_number_return=y */ create table t1 (s_c1 short, s_c2 short, i_c1 int, i_c2 int, b_c1 bigint, b_c2 bigint); insert into t1 values (1, 2, 1, 2, 1, 2); select s_c1/s_c2, i_c1/i_c2, b_c1/b_c2, 1/2, 1.0/2 from t1; /* s_c1/s_c2 i_c1/i_c2 b_c1/b_c2 1/2 1.0/2 ============================================================================================================== 0.5 0.5 0.5 0.5 0.5 */ select s_c1+s_c2, i_c1+i_c2, b_c1+b_c2, 1+2, 1.0+2 from t1; /* s_c1+s_c2 i_c1+i_c2 b_c1+b_c2 1+2 1.0+2 ============================================================================================================== 3 3 3 3 3 */ select s_c1-s_c2, i_c1-i_c2, b_c1-b_c2, 1-2, 1.0-2 from t1; /* s_c1-s_c2 i_c1-i_c2 b_c1-b_c2 1-2 1.0-2 ============================================================================================================== -1 -1 -1 -1 -1 */ select s_c1*s_c2, i_c1*i_c2, b_c1*b_c2, 1*2, 1.0*2 from t1; /* s_c1*s_c2 i_c1*i_c2 b_c1*b_c2 1*2 1.0*2 ============================================================================================================== 2 2 2 2 2 */ select typeof(s_c1/s_c2), typeof(i_c1/i_c2), typeof(b_c1/b_c2), typeof(1/2), typeof(1.0/2), typeof(1) from t1; /* typeof(s_c1/s_c2) typeof(i_c1/i_c2) typeof(b_c1/b_c2) typeof(1/2) typeof(1.0/2) typeof(1) ==================================================================================================================================== 'numeric (24, 9)' 'numeric (24, 9)' 'numeric (24, 9)' 'numeric (24, 9)' 'numeric (10, 9)' 'integer' */

The NUMERIC type has precision and scale info which might not be defined at compile time.
SELECT count(*) / 3 FROM t1 query returns integer.

It should not be changed in the pt_common_type function, it should be changed in the pt_common_type_op function so that the operator can be checked and changed.

diff --git a/src/parser/type_checking.c b/src/parser/type_checking.c index d27a1891b..cf40e6816 100644 --- a/src/parser/type_checking.c +++ b/src/parser/type_checking.c @@ -11310,6 +11310,12 @@ pt_common_type_op (PT_TYPE_ENUM t1, PT_OP_TYPE op, PT_TYPE_ENUM t2) switch (op) { + case PT_DIVIDE: + if (PT_IS_DISCRETE_NUMBER_TYPE (t1) && PT_IS_DISCRETE_NUMBER_TYPE (t2)) + { + result_type = PT_TYPE_NUMERIC; + } + break; case PT_MINUS: case PT_TIMES: if (result_type == PT_TYPE_SEQUENCE)

shparkcubrid · 2023-05-17T05:23:45Z

In Oracle, unnecessary '.0' is not printed. Please check test cases of test_medium.

SQL> select 4/2 from dual;

       4/2
----------
         2

shparkcubrid · 2023-05-17T05:28:07Z

The test case below is not normal. Please check.
test_medium

evaluate 150000000000 / 6543;
** Difference between Expected(-) and Actual(+) results:
@@ -1,4 +1,4 @@
 ===================================================
     
-22925263     
+2.2925263640531868E7

beyondykk9 · 2023-05-17T22:58:18Z

In Oracle, unnecessary '.0' is not printed. Please check test cases of test_medium.
SQL> select 4/2 from dual;

       4/2
----------
         2

trailing zeros are suppressed in oracle style printing numeric.

hyunikn

In the last comment of RND-1815, technical support team requested that the result type should be numeric rather than double. Was there a further discussion and decision on this point?

beyondykk9 · 2023-05-21T23:04:42Z

In the last comment of RND-1815, technical support team requested that the result type should be numeric rather than double. Was there a further discussion and decision on this point?

Finally, it would be NUMERIC type.

hgryoo · 2023-05-22T07:55:07Z

As a result of this issue, is it okay if the result of DIVIDE is numeric? Or should the precision or the resulting value be exactly the same?

hgryoo · 2023-05-22T08:20:28Z

Floating error could be accumulated a lot like the following because of scale:

SELECT (5555 / 7) / (7.5 / 11) FROM dual;

Oracle

1163.904761904761904761904761904761904763

CUBRID

1163.904761904761594 -- SELECT (5555 / 7) / (7.5 / 11) FROM dual;
1163.9047619047619047619047619047619358 -- SELECT (cast(5555 as numeric(38,34)) / 7) / (cast(7.5 as numeric(38,34)) / 11) FROM dual;

hgryoo · 2023-05-22T08:21:00Z

Floating error could be accumulated a lot like the following because of scale:

SELECT (5555 / 7) / (7.5 / 11) FROM dual;

Oracle

1163.904761904761904761904761904761904763

CUBRID

1163.904761904761594 -- SELECT (5555 / 7) / (7.5 / 11) FROM dual;
1163.9047619047619047619047619047619358 -- SELECT (cast(5555 as numeric(38,34)) / 7) / (cast(7.5 as numeric(38,34)) / 11) FROM dual;

Maybe It could be fixed by #4375

hgryoo

The issue was confirmed to be related to the format of that particular type rather than the actual value resulting from the division operation. Approved.

shparkcubrid · 2023-05-23T04:29:38Z

In Oracle, unnecessary '.0' is not printed. Please check test cases of test_medium.
SQL> select 4/2 from dual;

       4/2
----------
         2
trailing zeros are suppressed in oracle style printing numeric.

it's fail case of test_medium.

select country, rownum/10 as number
from public.location
where mod(rownum, 10) = 0 order by 1;
** Difference between Expected(-) and Actual(+) results:
@@ -1,6 +1,6 @@
 ===================================================
 country    number    
-Aruba     1     
-Bahamas     3     
-West Indies     2     
+Aruba     1.000000000     
+Bahamas     3.000000000     
+West Indies     2.000000000

[CBRD-24684] divide operation result into numeric type

10a89e6

beyondykk9 self-assigned this Feb 28, 2023

fix: add numeric_db_value_to_num

bb71cd7

beyondykk9 marked this pull request as draft March 1, 2023 23:53

beyondykk9 added 16 commits March 2, 2023 09:18

fix: number type to double

aba26f7

fix: overflow handling

007626f

fix: return error

84781dc

Merge remote-tracking branch 'upstream/develop' into CBRD-24684

9fe487a

fix: numeric precision setting

c811dce

temp

ab03185

fix: int/bigint/short divide to double

426bc49

Merge remote-tracking branch 'upstream/develop' into CBRD-24684

8ecc4b3

refine code

c954c20

fix csql_result_format and simplifying

b2b88dd

remove useless codes

c39ec6d

revert

a3e1e58

fix: remove trailing zero

a2e78a9

Merge remote-tracking branch 'upstream/develop' into CBRD-24684

87362fb

add system parameter oracle_style_divide

ceae786

temp: default true

753eea3

beyondykk9 changed the title ~~[CBRD-24684] divide operation result into numeric type~~ [CBRD-24684] divide operation result into double type May 10, 2023

beyondykk9 added 2 commits May 11, 2023 10:55

chagne macro TP_IS_DISCRETE_TYPE

b757be6

double format for oracle_style_divide

fc850dc

beyondykk9 requested review from hgryoo, shparkcubrid, hyunikn, ctshim and youngjinj May 15, 2023 06:43

hgryoo marked this pull request as ready for review May 15, 2023 07:24

ctshim reviewed May 15, 2023

View reviewed changes

youngjinj reviewed May 16, 2023

View reviewed changes

change: pt_common_op to NUMERIC for DISCRETE NUMBER type

f1175f0

hyunikn reviewed May 19, 2023

View reviewed changes

beyondykk9 added 2 commits May 19, 2023 16:19

modified as comments of Youngjin

b3b039c

add type cast to numeric for server execution

dbb42f5

hgryoo approved these changes May 23, 2023

View reviewed changes

youngjinj approved these changes May 25, 2023

View reviewed changes

hyunikn approved these changes Jun 1, 2023

View reviewed changes

Merge remote-tracking branch 'upstream/develop' into CBRD-24684

075579f

ctshim approved these changes Jun 1, 2023

View reviewed changes

fix default parameter

f8c34f2

beyondykk9 changed the title ~~[CBRD-24684] divide operation result into double type~~ [CBRD-24684] divide operation result into numeric type Jun 1, 2023

beyondykk9 merged commit 63133ba into CUBRID:develop Jun 1, 2023

beyondykk9 deleted the CBRD-24684 branch June 1, 2023 23:41

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[CBRD-24684] divide operation result into numeric type #4150

[CBRD-24684] divide operation result into numeric type #4150

beyondykk9 commented Feb 28, 2023

ctshim May 15, 2023

beyondykk9 May 15, 2023

youngjinj May 16, 2023 •

edited

Loading

beyondykk9 May 17, 2023

youngjinj May 17, 2023 •

edited

Loading

shparkcubrid commented May 17, 2023 •

edited

Loading

shparkcubrid commented May 17, 2023

beyondykk9 commented May 17, 2023

hyunikn left a comment •

edited

Loading

beyondykk9 commented May 21, 2023

hgryoo commented May 22, 2023

hgryoo commented May 22, 2023

hgryoo commented May 22, 2023

hgryoo left a comment

shparkcubrid commented May 23, 2023 •

edited

Loading

		@@ -728,6 +728,8 @@ static const char sysprm_ha_conf_file_name[] = "cubrid_ha.conf";

		#define PRM_NAME_STATDUMP_FORCE_ADD_INT_MAX "statdump_force_add_int_max"

		#define PRM_NAME_ORACLE_STYLE_DIVIDE "oracle_style_divide"

[CBRD-24684] divide operation result into numeric type #4150

[CBRD-24684] divide operation result into numeric type #4150

Conversation

beyondykk9 commented Feb 28, 2023

ctshim May 15, 2023

Choose a reason for hiding this comment

beyondykk9 May 15, 2023

Choose a reason for hiding this comment

youngjinj May 16, 2023 • edited Loading

Choose a reason for hiding this comment

beyondykk9 May 17, 2023

Choose a reason for hiding this comment

youngjinj May 17, 2023 • edited Loading

Choose a reason for hiding this comment

shparkcubrid commented May 17, 2023 • edited Loading

shparkcubrid commented May 17, 2023

beyondykk9 commented May 17, 2023

hyunikn left a comment • edited Loading

Choose a reason for hiding this comment

beyondykk9 commented May 21, 2023

hgryoo commented May 22, 2023

hgryoo commented May 22, 2023

hgryoo commented May 22, 2023

hgryoo left a comment

Choose a reason for hiding this comment

shparkcubrid commented May 23, 2023 • edited Loading

youngjinj May 16, 2023 •

edited

Loading

youngjinj May 17, 2023 •

edited

Loading

shparkcubrid commented May 17, 2023 •

edited

Loading

hyunikn left a comment •

edited

Loading

shparkcubrid commented May 23, 2023 •

edited

Loading