expression: fix a bug caused by cast function signature wrong in aggregation push down

[tiancaiamao]

What problem does this PR solve?

Issue Number: close #28804

Problem Summary:

What is changed and how it works?

Remove the special handle of cast function in ColumnSubstituteImpl

The bug is caused by this:

Before ColumnSubstitute, the function signature of cast(floor(hour(ts) / 4) as char) is cast(Int -> Varstring) ...
Then, the ColumnSubstitute will substitute the column ts of expression floor(hour(ts) / 4)
After that, the expression floor(hour(ts) / 4) signature changed, it's return type is not Int any more.
Expression floor(hour(ts) / 4) returns Decimal instead.

But the signature of cast() function is still cast(Int -> Varstring), so the cast expect floor(hour(ts) / 4) to return Int and the later returns Decimal, the error is throw finally.

What's Changed:

Remove the special handle of the cast function.

How it Works:

The bug is caused by the special handle of the Cast function.
Its function args are re-generated (signature re-calculated), but the Cast function it self does not refresh.
So the signature mismatch...

OK, "why the signature change from Int to Decimal" is another story, it's caused by the wrong Flen and that's another big issue.
I'm not meant to solve it here.

Check List

Tests

• [] Unit test
• Integration test
• Manual test (add detailed scripts or steps below)
• No code

Side effects

• Performance regression: Consumes more CPU
• Performance regression: Consumes more Memory
• Breaking backward compatibility

Documentation

• Affects user behaviors
• Contains syntax changes
• Contains variable changes
• Contains experimental features
• Changes MySQL compatibility

Release note

Fix `ColumnSubstitute` cause wrong signature of the Cast() function, this bug could led TiDB return "baseBuiltinFunc.vecEvalInt() should never be called" error in some aggregation push down scenario.

[ti-chi-bot]

[REVIEW NOTIFICATION]

This pull request has been approved by:

• guo-shaoge
• wjhuang2016

To complete the pull request process, please ask the reviewers in the list to review by filling /cc @reviewer in the comment.
After your PR has acquired the required number of LGTMs, you can assign this pull request to the committer in the list by filling /assign @committer in the comment to help you merge this pull request.

The full list of commands accepted by this bot can be found here.

Reviewer can indicate their review by submitting an approval review.
Reviewer can cancel approval by submitting a request changes review.

[tiancaiamao]

Unstable CI #24679

[2021-10-26T08:26:43.833Z] ----------------------------------------------------------------------
[2021-10-26T08:26:43.833Z] FAIL: handle_test.go:957: testSerialStatsSuite.TestAnalyzeGlobalStatsWithOpts2
[2021-10-26T08:26:43.833Z] 
[2021-10-26T08:26:43.833Z] handle_test.go:968:
[2021-10-26T08:26:43.833Z]     s.checkForGlobalStatsWithOpts(c, tk, "test_gstats_opt2", "p1", 1, 50)
[2021-10-26T08:26:43.833Z] handle_test.go:913:
[2021-10-26T08:26:43.833Z]     // since the hist-building algorithm doesn't stipulate the final bucket number to be equal to the expected number exactly,
[2021-10-26T08:26:43.833Z]     // we have to check the results by a range here.
[2021-10-26T08:26:43.833Z]     c.Assert(numBuckets, GreaterEqual, buckets-delta)
[2021-10-26T08:26:43.833Z] ... compare_one int = 0
[2021-10-26T08:26:43.833Z] ... compare_two int = 15

/run-check_dev_2

[tiancaiamao]

CI failed again, I can't reproduce it on my local environment:

[2021-10-26T08:37:21.334Z] FAIL: integration_test.go:1911: testIntegrationSerialSuite.TestIndexMerge
[2021-10-26T08:37:21.334Z] 
[2021-10-26T08:37:21.334Z] integration_test.go:1933:
[2021-10-26T08:37:21.334Z]     tk.MustQuery(tt).Check(testkit.Rows(output[i].Plan...))
[2021-10-26T08:37:21.334Z] /home/jenkins/agent/workspace/tidb_ghpr_check_2/go/src/github.com/pingcap/tidb/util/testkit/testkit.go:66:
[2021-10-26T08:37:21.334Z]     res.c.Assert(resBuff.String(), check.Equals, needBuff.String(), res.comment)
[2021-10-26T08:37:21.334Z] ... obtained string = "" +
[2021-10-26T08:37:21.334Z] ...     "[IndexMerge 8.00 root  ]\n" +
[2021-10-26T08:37:21.334Z] ...     "[├─IndexRangeScan(Build) 1.00 cop[tikv] table:t, index:a(a) range:[1,1], keep order:false]\n" +
[2021-10-26T08:37:21.334Z] ...     "[├─Selection(Build) 0.80 cop[tikv]  1]\n" +
[2021-10-26T08:37:21.334Z] ...     "[│ └─IndexRangeScan 1.00 cop[tikv] table:t, index:b(b) range:[1,1], keep order:false]\n" +
[2021-10-26T08:37:21.334Z] ...     "[└─TableRowIDScan(Probe) 8.00 cop[tikv] table:t keep order:false]\n"
[2021-10-26T08:37:21.334Z] ... expected string = "" +
[2021-10-26T08:37:21.334Z] ...     "[IndexMerge 1.72 root  ]\n" +
[2021-10-26T08:37:21.334Z] ...     "[├─IndexRangeScan(Build) 1.00 cop[tikv] table:t, index:a(a) range:[1,1], keep order:false]\n" +
[2021-10-26T08:37:21.334Z] ...     "[├─Selection(Build) 0.80 cop[tikv]  eq(length(cast(1, var_string(20))), 1)]\n" +
[2021-10-26T08:37:21.334Z] ...     "[│ └─IndexRangeScan 1.00 cop[tikv] table:t, index:b(b) range:[1,1], keep order:false]\n" +
[2021-10-26T08:37:21.334Z] ...     "[└─TableRowIDScan(Probe) 1.72 cop[tikv] table:t keep order:false]\n"
[2021-10-26T08:37:21.334Z] ... sql:desc format='brief' select /*+ use_index_merge(t) */ * from t where a =1 or (b=1 and length(b)=1), args:[]

[tiancaiamao]

/run-check_dev_2

[tiancaiamao]

/merge

[ti-chi-bot]

This pull request has been accepted and is ready to merge.

Commit hash: b0072d9

[tiancaiamao]

This change is expected, because after fixing the bug, the constant propagate can infer that
b = 1 and length(b) = 1 is always true.
So the this filter condition changes and the cost of the plan changes...

[tiancaiamao]

/run-check_dev_2

[ti-srebot]

cherry pick to release-5.1 in PR #29316

[ti-srebot]

cherry pick to release-5.2 in PR #29345