expression: make in more powerful. #4969
Conversation
| for i := stkLen - lLen; i < stkLen; i++ { | ||
| if er.ctxStack[i].GetType().Tp != mysql.TypeNull && er.ctxStack[i].GetType().EvalType() != leftEt { | ||
| for _, arg := range args[1:] { | ||
| if arg.GetType().Tp != mysql.TypeNull && expression.GetAccurateCmpType(args[0], arg) != leftEt { |
There was a problem hiding this comment.
Can we collect all the argument types and merge them to one type, then just cast all the expressions to this type and always use in?
There was a problem hiding this comment.
We can group them to the form x in(type_x_expr) or cast(x to type_a) in(type_a_expr)or....But cannot to convert to one type. e.g. x in ('a', 1.1). x's type is string type. We should compare x and a in string type, x and 1.1 in double type. But we can't compare x and a in double type since this will make result absolutely wrong. And we can't compare x and 1.1 in string type since this don't obey the compare system in MySQL.
There was a problem hiding this comment.
How about build DNF with different types?
(intValLeft in (intA, intB, intC)) or (floatValLeft in (floatA, floatB))?
There was a problem hiding this comment.
But currently this is not very necessary to do this. This case can't push down to tikv, and can't be used to calculate range.
|
LGTM |
util/ranger/range_test.go
Outdated
| @@ -161,10 +161,12 @@ func (s *testRangerSuite) TestTableRange(c *C) { | |||
| resultStr: "[(-inf,-1] [1,+inf)]", | |||
| }, | |||
| // TODO: cast will change the extraction behavior of accessCondition. | |||
|
/run-all-tests |
|
/run-unit-test /run-integration-ddl-test |
|
mysql-test needs to do some change. Don't merge for a while. |
|
/run-common-test tidb-test=pr/396 |
Let
incan handle more case, so we can make more expression able to calculate range.PTAL @XuHuaiyu @zz-jason @hanfei1991