This is an automated cherry-pick of #48845

Signed-off-by: ti-chi-bot <ti-community-prow-bot@tidb.io>

pkg/planner/core/casetest/physicalplantest/testdata/plan_suite_out.json

@@ -2171,11 +2171,11 @@
  },
  {
  "SQL": "select a from t where c_str like ''",
- "Best": "IndexReader(Index(t.c_d_e_str)[[\"\",\"\"]])->Projection"
+ "Best": "IndexReader(Index(t.c_d_e_str)[[\"\",\"\"]]->Sel([like(test.t.c_str, , 92)]))->Projection"
  },
  {
  "SQL": "select a from t where c_str like 'abc'",
- "Best": "IndexReader(Index(t.c_d_e_str)[[\"abc\",\"abc\"]])->Projection"
+ "Best": "IndexReader(Index(t.c_d_e_str)[[\"abc\",\"abc\"]]->Sel([like(test.t.c_str, abc, 92)]))->Projection"
  },
  {
  "SQL": "select a from t where c_str not like 'abc'",
@@ -2191,7 +2191,7 @@
  },
  {
  "SQL": "select a from t where c_str like 'abc%'",
- "Best": "IndexReader(Index(t.c_d_e_str)[[\"abc\",\"abd\")])->Projection"
+ "Best": "IndexReader(Index(t.c_d_e_str)[[\"abc\",\"abd\")]->Sel([like(test.t.c_str, abc%, 92)]))->Projection"
  },
  {
  "SQL": "select a from t where c_str like 'abc_'",
@@ -2203,31 +2203,31 @@
  },
  {
  "SQL": "select a from t where c_str like 'abc\\_' escape ''",
- "Best": "IndexReader(Index(t.c_d_e_str)[[\"abc_\",\"abc_\"]])->Projection"
+ "Best": "IndexReader(Index(t.c_d_e_str)[[\"abc_\",\"abc_\"]]->Sel([like(test.t.c_str, abc\\_, 92)]))->Projection"
  },
  {
  "SQL": "select a from t where c_str like 'abc\\_'",
- "Best": "IndexReader(Index(t.c_d_e_str)[[\"abc_\",\"abc_\"]])->Projection"
+ "Best": "IndexReader(Index(t.c_d_e_str)[[\"abc_\",\"abc_\"]]->Sel([like(test.t.c_str, abc\\_, 92)]))->Projection"
  },
  {
  "SQL": "select a from t where c_str like 'abc\\\\_'",
- "Best": "IndexReader(Index(t.c_d_e_str)[[\"abc_\",\"abc_\"]])->Projection"
+ "Best": "IndexReader(Index(t.c_d_e_str)[[\"abc_\",\"abc_\"]]->Sel([like(test.t.c_str, abc\\_, 92)]))->Projection"
  },
  {
  "SQL": "select a from t where c_str like 'abc\\_%'",
- "Best": "IndexReader(Index(t.c_d_e_str)[[\"abc_\",\"abc`\")])->Projection"
+ "Best": "IndexReader(Index(t.c_d_e_str)[[\"abc_\",\"abc`\")]->Sel([like(test.t.c_str, abc\\_%, 92)]))->Projection"
  },
  {
  "SQL": "select a from t where c_str like 'abc=_%' escape '='",
- "Best": "IndexReader(Index(t.c_d_e_str)[[\"abc_\",\"abc`\")])->Projection"
+ "Best": "IndexReader(Index(t.c_d_e_str)[[\"abc_\",\"abc`\")]->Sel([like(test.t.c_str, abc=_%, 61)]))->Projection"
  },
  {
  "SQL": "select a from t where c_str like 'abc\\__'",
  "Best": "IndexReader(Index(t.c_d_e_str)[(\"abc_\",\"abc`\")]->Sel([like(test.t.c_str, abc\\__, 92)]))->Projection"
  },
  {
  "SQL": "select a from t where c_str like 123",
- "Best": "IndexReader(Index(t.c_d_e_str)[[\"123\",\"123\"]])->Projection"
+ "Best": "IndexReader(Index(t.c_d_e_str)[[\"123\",\"123\"]]->Sel([like(test.t.c_str, 123, 92)]))->Projection"
  },
  {
  "SQL": "select a from t where c = 1.9 and d > 3",

pkg/util/ranger/checker.go

@@ -17,6 +17,7 @@ package ranger
 import (
  "github.com/pingcap/tidb/pkg/expression"
  "github.com/pingcap/tidb/pkg/parser/ast"
+ "github.com/pingcap/tidb/pkg/parser/charset"
  "github.com/pingcap/tidb/pkg/parser/mysql"
  "github.com/pingcap/tidb/pkg/types"
  "github.com/pingcap/tidb/pkg/util/collate"
@@ -166,11 +167,22 @@ func (c *conditionChecker) checkLikeFunc(scalar *expression.ScalarFunction) (isA
  if err != nil {
  return false, true
  }
+ likeFuncReserve := !c.isFullLengthColumn()
+
+ // Different from `=`, trailing spaces are always significant, and can't be ignored in `like`.
+ // In tidb's implementation, for PAD SPACE collations, the trailing spaces are removed in the index key. So we are
+ // unable to distinguish 'xxx' from 'xxx ' by a single index range scan, and we may read more data than needed by
+ // the `like` function. Therefore, a Selection is needed to filter the data.
+ // Since all collations, except for binary, implemented in tidb are PAD SPACE collations for now, we use a simple
+ // collation != binary check here.
+ if collation != charset.CollationBin {
+ likeFuncReserve = true
+ }
+
  if len(patternStr) == 0 {
- return true, !c.isFullLengthColumn()
+ return true, likeFuncReserve
  }
  escape := byte(scalar.GetArgs()[2].(*expression.Constant).Value.GetInt64())
- likeFuncReserve := !c.isFullLengthColumn()
  for i := 0; i < len(patternStr); i++ {
  if patternStr[i] == escape {
  i++

pkg/util/ranger/ranger_test.go

@@ -1098,7 +1098,7 @@ create table t(
  indexPos: 0,
  exprStr: `a LIKE 'abc%'`,
  accessConds: `[like(test.t.a, abc%, 92)]`,
- filterConds: "[]",
+ filterConds: "[like(test.t.a, abc%, 92)]",
  resultStr: "[[\"abc\",\"abd\")]",
  },
  {
@@ -1112,14 +1112,14 @@ create table t(
  indexPos: 0,
  exprStr: "a LIKE 'abc'",
  accessConds: "[like(test.t.a, abc, 92)]",
- filterConds: "[]",
+ filterConds: "[like(test.t.a, abc, 92)]",
  resultStr: "[[\"abc\",\"abc\"]]",
  },
  {
  indexPos: 0,
  exprStr: `a LIKE "ab\_c"`,
  accessConds: "[like(test.t.a, ab\\_c, 92)]",
- filterConds: "[]",
+ filterConds: "[like(test.t.a, ab\\_c, 92)]",
  resultStr: "[[\"ab_c\",\"ab_c\"]]",
  },
  {
@@ -1133,21 +1133,21 @@ create table t(
  indexPos: 0,
  exprStr: `a LIKE '\%a'`,
  accessConds: "[like(test.t.a, \\%a, 92)]",
- filterConds: "[]",
+ filterConds: "[like(test.t.a, \\%a, 92)]",
  resultStr: `[["%a","%a"]]`,
  },
  {
  indexPos: 0,
  exprStr: `a LIKE "\\"`,
  accessConds: "[like(test.t.a, \\, 92)]",
- filterConds: "[]",
+ filterConds: "[like(test.t.a, \\, 92)]",
  resultStr: "[[\"\\\\\",\"\\\\\"]]",
  },
  {
  indexPos: 0,
  exprStr: `a LIKE "\\\\a%"`,
  accessConds: `[like(test.t.a, \\a%, 92)]`,
- filterConds: "[]",
+ filterConds: "[like(test.t.a, \\\\a%, 92)]",
  resultStr: "[[\"\\\\a\",\"\\\\b\")]",
  },
  {

tests/integrationtest/r/explain_generate_column_substitute.result

@@ -415,7 +415,8 @@ id estRows task access object operator info
 StreamAgg 1.00 root funcs:count(Column#6)->Column#4
 └─IndexReader 1.00 root index:StreamAgg
  └─StreamAgg 1.00 cop[tikv] funcs:count(1)->Column#6
- └─IndexRangeScan 250.00 cop[tikv] table:tbl1, index:expression_index(md5(`s`)) range:["02e74f10e0327ad868d138f2b4fdd6f","02e74f10e0327ad868d138f2b4fdd6g"), keep order:false, stats:pseudo
+ └─Selection 250.00 cop[tikv] like(md5(cast(explain_generate_column_substitute.tbl1.s, var_string(20))), "02e74f10e0327ad868d138f2b4fdd6f%", 92)
+ └─IndexRangeScan 250.00 cop[tikv] table:tbl1, index:expression_index(md5(`s`)) range:["02e74f10e0327ad868d138f2b4fdd6f","02e74f10e0327ad868d138f2b4fdd6g"), keep order:false, stats:pseudo
 select count(*) from tbl1 use index() where md5(s) like '02e74f10e0327ad868d138f2b4fdd6f%';
 count(*)
 64