executor: fix a bug for 'int column <cmp> non-int constant'

[wshwsh12]

What problem does this PR solve?

#10675
Wrong result for int column <cmp> non-int constant when the constant is inf or -inf

CREATE TABLE `t` (
  `a` int(11) DEFAULT NULL
);
insert into t values(1);
select * from t where a < -184467440737095516167.1;

Before this pr:

select * from t where a < -184467440737095516167.1;
+------+
| a    |
+------+
|    1 |
+------+

This pr:

select * from t where a < -184467440737095516167.1;
Empty set (0.00 sec)

What is changed and how it works?

When the compare expression is int column <cmp> non-int constant or non-int constant <cmp> int column, TiDB will refine the constant to the int column type.
E.g., a < 1.1 will be rewritten to a < 2.

Before this pr, the logic of refine function is wrong. It is only judged whether the constant is Overflow.
a < -184467440737095516167.1 will be rewritten to a < inf
a > 184467440737095516167.1 will be rewritten to a > -inf

This pr, I use the leading sing and Overflow flag to decide the constant is 'inf' or '-inf'
a < -184467440737095516167.1 will be rewritten to a < -inf
a > 184467440737095516167.1 will be rewritten to a > inf

Check List

Tests

• Unit test
• Manual test (add detailed scripts or steps below)

Code changes

Side effects

Related changes

[wshwsh12]

/run-all-tests

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (master@47f449a). Click here to learn what that means.
The diff coverage is 82%.

@@            Coverage Diff             @@
##             master    #11050   +/-   ##
==========================================
  Coverage          ?   81.259%           
==========================================
  Files             ?       423           
  Lines             ?     90006           
  Branches          ?         0           
==========================================
  Hits              ?     73138           
  Misses            ?     11569           
  Partials          ?      5299

[wshwsh12]

/run-all-tests

[XuHuaiyu]

We can define this intFieldType as an input parameter. It should be the FiledType of the int column.

tidb> create table t (a int);
Query OK, 0 rows affected (0.01 sec)

tidb> desc select * from t where a > 128; -- It should be TableDual for the sql as well.
+---------------------+----------+------+------------------------------------------------------------+
| id                  | count    | task | operator info                                              |
+---------------------+----------+------+------------------------------------------------------------+
| TableReader_7       | 3333.33  | root | data:Selection_6                                           |
| └─Selection_6       | 3333.33  | cop  | gt(test.t.a, 128)                                          |
|   └─TableScan_5     | 10000.00 | cop  | table:t, range:[-inf,+inf], keep order:false, stats:pseudo |
+---------------------+----------+------+------------------------------------------------------------+
3 rows in set (0.00 sec)

tidb> desc select * from t where a > 11111111111111111111111111111;
+-------------+-------+------+---------------+
| id          | count | task | operator info |
+-------------+-------+------+---------------+
| TableDual_6 | 0.00  | root | rows:0        |
+-------------+-------+------+---------------+
1 row in set (0.00 sec)

[qw4990]

Could you please describe more details about the root reason causing this bug.
Then it's more convenient for reviewers to review your changes. @wshwsh12

[wshwsh12]

/run-all-tests

[wshwsh12]

/run-all-tests

[wshwsh12]

/run-all-tests

[qw4990]

LGTM

[qw4990]

/run-all-tests

[XuHuaiyu]

/run-unit-test

[wshwsh12]

/rebuild