planner/core: get back range columns pruning after last refactor

[tiancaiamao]

What problem does this PR solve?

create table t (a varchar(32)) partition by range columns (a) (
    partition p0 values less than ('c'),
    partition p1 values less than ('f'),
    partition p2 values less than ('g'),
    partition p3 values less than ('n'),
    partition p4 values less than ('z')
);
select * from t;

Error Code: 1105. strconv.ParseInt: parsing "\"c\"": invalid syntax

What is changed and how it works?

Fix regression after #14834

Partition pruning for 'partition by range columns' table is broken after that refactor, as we only handle integer in the new partition pruning algorithm, 'by range columns' may contain non-integer data.

Now I've added expression comparing in addition to integer in this commit.

Check List

Tests

• Unit test

[tiancaiamao]

PTAL @imtbkcat @lysu

[codecov]

Codecov Report

Merging #15169 into master will decrease coverage by 0.0902%.
The diff coverage is 85.1239%.

@@               Coverage Diff                @@
##             master     #15169        +/-   ##
================================================
- Coverage   80.3669%   80.2766%   -0.0903%     
================================================
  Files           502        503         +1     
  Lines        133805     132280      -1525     
================================================
- Hits         107535     106190      -1345     
+ Misses        17823      17700       -123     
+ Partials       8447       8390        -57

[lysu]

should we let "range column by int" like this can use makeLessThanData?

create table t (a int) partition by range columns (a) (
    partition p0 values less than (0),
    partition p1 values less than (500),
    partition p2 values less than (1000),
    partition p3 values less than (1500),
    partition p4 values less than (2000)
);
select * from t;

(that way should be faster?)

[tiancaiamao]

Yes, it's faster, but the case is not general, we have to write more code for this optimizing

[lysu]

both RANGE COLUMNS partitioning and LIST COLUMNS partitioning support the use of non-integer columns for defining value ranges or list members. The permitted data types are shown in the following list:

• All integer types: TINYINT, SMALLINT, MEDIUMINT, INT (INTEGER), and BIGINT. (This is the same as with partitioning by RANGE and LIST.) Other numeric data types (such as DECIMAL or FLOAT) are not supported as partitioning columns.

• DATE and DATETIME. Columns using other data types relating to dates or times are not supported as partitioning columns.

• The following string types: CHAR, VARCHAR, BINARY, and VARBINARY. TEXT and BLOB columns are not supported as partitioning columns.

it seems partition by range columns only support three type int/date/string, so ideal way is impl two addition lessThanData type ---------- lessThanDataInt, lessThanDataTime and lessThanDataString to void new & eval expression for each compare?

but LGTM to use current impl to quick fix

[tiancaiamao]

If some type is not supported, we should prevent the case in DDL check, rather than in partition pruning.
lessThanDataTime and lessThanDataString seems good, we can optimize it later.

[lysu]

@winoros please help to take a look if free

[tiancaiamao]

/rebuild

[lysu]

LGTM, @winoros @eurekaka PTAL~

[tiancaiamao]

@imtbkcat PTAL

[imtbkcat]

LGTM

[tiancaiamao]

/merge

[sre-bot]

/run-all-tests