Debugging MDEV-10046, I have discovered that range optimizer will call records_in_range() when we have a full equality range for an extended index.

Example:

create table t1 (

  pk int primary key,

  col1 int,

  ...

  index IDX1(col1)

);

explain select * from t1 where col1 IN (c11, c12, ...) and pk IN (c21, c22, ..)

Here, when doing range analysis for IDX1, we get ranges in form:

(col1, pk)= (c11, c21)

When PK is a part of the index, there is no need to call records_in_range: we don't call it for unique indexes, InnoDB internally is set to return 1 when it sees a zero estimate.