That makes sense. We should only check the requested columns, not every
column in parquet file, to decide which parquet reader to use.


On Wed, Sep 13, 2017 at 4:17 PM, Damien Profeta <[email protected]>
wrote:

> Hi,
>
> I was looking at the code that read the parquet file and noticed there is
> a switch 'isComplex' to choose if it is possible to use the new reader or
> if we have to use the old one.
> The switch is based on the columns of the files (complex type or
> repetition level) but it doesn't care about the columns that have to
> effectively read.
>
> If for the ongoing query, we only read simple column and not with
> repetition level, couldn't we use the new reader? That would be a minor
> optimization but it could be worth.
>
> Thanks
> Damien
>
>

Reply via email to