[SPARK-14545][SQL] Improve LikeSimplification by adding a%b rule

[dongjoon-hyun]

What changes were proposed in this pull request?

Current LikeSimplification handles the following four rules.

• 'a%' => expr.StartsWith("a")
• '%b' => expr.EndsWith("b")
• '%a%' => expr.Contains("a")
• 'a' => EqualTo("a")

This PR adds the following rule.

• 'a%b' => expr.Length() >= 2 && expr.StartsWith("a") && expr.EndsWith("b")

Here, 2 is statically calculated from "a".size + "b".size.

Before

scala> sql("select a from (select explode(array('abc','adc')) a) T where a like 'a%c'").explain()
== Physical Plan ==
WholeStageCodegen
:  +- Filter a#5 LIKE a%c
:     +- INPUT
+- Generate explode([abc,adc]), false, false, [a#5]
   +- Scan OneRowRelation[]

After

scala> sql("select a from (select explode(array('abc','adc')) a) T where a like 'a%c'").explain()
== Physical Plan ==
WholeStageCodegen
:  +- Filter ((length(a#5) >= 2) && (StartsWith(a#5, a) && EndsWith(a#5, c)))
:     +- INPUT
+- Generate explode([abc,adc]), false, false, [a#5]
   +- Scan OneRowRelation[]

How was this patch tested?

Pass the Jenkins tests (including new testcase).

[SparkQA]

Test build #55555 has finished for PR 12312 at commit 8d496f6.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[rxin]

i think you should have some comments explaining this rewrite, in particular the greaterthanorequal part is not as straightforward as the other ones.

[dongjoon-hyun]

Thank you for review. I'll add comments here.

[dongjoon-hyun]

Hi, @rxin . Sorry for late response.
I added comments and changed the testcase name according to your comments.
Thank you for review!

[SparkQA]

Test build #55618 has finished for PR 12312 at commit 16ab6c8.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[dongjoon-hyun]

Hi, @rxin .
For LikeSimplification, is there something to do more?

[rxin]

while you are at it, can you rename "pattern" in the startsWith case to prefix, and endsWith to suffix?

[rxin]

also rename utf to "pattern"

and just compute the value of GreaterThanOrEqual(Length(l), Literal(prefix.size + postfix.size)) directly since it is a literal

[dongjoon-hyun]

Sure. I will renames those parameters.
By the way, l of Length(l) is not literal here. It could be column.

[dongjoon-hyun]

Thank you for spending time here. I know your are very busy.

[rxin]

ah ok - let's leave the l one there

[dongjoon-hyun]

I updated to use pattern, prefix, postfix, and infix (as a similar manner).

[rxin]

sorry to be pedantic. do you mind changing l to something like input? "l" is too short and unclear what it means, and it also just looks like one.

[dongjoon-hyun]

You're welcome. :) I'll fix right now!

[SparkQA]

Test build #55826 has finished for PR 12312 at commit 5753437.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[SparkQA]

Test build #55830 has finished for PR 12312 at commit 111a78c.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds no public classes.

[rxin]

Thanks - merging in master.

[dongjoon-hyun]

Thank you for merging, @rxin !