Regarding to original user tables, my observation is column names are passed to each Rel from bottom to top (and saves in row type) unless there is an alias.
But please do correct me if my observation is wrong. -Rui On Sun, Sep 8, 2019 at 2:19 PM Rui Wang <[email protected]> wrote: > Hi Dustin, > > 1. My understanding is RexInputRef's index is an offset from its inputs > and indexed from the first column from the first input(offset starts from > 0). For a Rel with a single input, the offset is the column position from > index 0. In BeamSQL's implementation, if I recall correctly, for a Join > Rel, the left join input is indexed from 0 to len(left_schema) - 1, and > right join input is from len(left_schema) to len(left_schema) + > len(right_schema) - 1. > > 2. After you locate the column based on RexInput, you can find column name > from input.getRowType().getFieldList().get(i).getName(). > > RexInputRef saves the index and column type. I am not aware that it saves > any column name directly. Usually column name is not useful inside a > logical plan. > > > -Rui > > > On Sun, Sep 8, 2019 at 1:48 PM Dustin Jiang <[email protected]> > wrote: > >> Hi, >> I am new to calcite. if you can provide any info about the following >> questions, I would really appreciate it. >> >> I have a RexInpuRef which refers to a column from input schema. 1) how do >> I find the column name and column position in its input schema? The the >> input schema is formed by joining two tables, 2) how do I find original >> column name and original column position in original user-defined table? >> >> Do we store any such info in the RefInputRef node itself? >> >> Thanks in advance, >> >> Dustin >> >
