hadoop-pig-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "pi song" <pi.so...@gmail.com>
Subject Re: Handling of untyped execution in type branch?
Date Tue, 20 May 2008 23:59:17 GMT
Thanks!!!

On 5/21/08, Santhosh Srinivasan <sms@yahoo-inc.com> wrote:
>
> Pi,
>
> I pushed out a patch today to handle null schemas in LOUnion. I am
> reviewing your patch which has the changes for merging schemas.
>
> We should go with 2 where every operator supports null schemas.
>
> Thanks,
> Santhosh
>
>
> -----Original Message-----
> From: pi song [mailto:pi.songs@gmail.com]
> Sent: Tuesday, May 20, 2008 4:23 PM
> To: pig-dev@incubator.apache.org
> Subject: Re: Handling of untyped execution in type branch?
>
> Attention!!!: This is an important design decision. Within 24 hours if
> nobody replies, I will assume option (1) to keep the work going.
>
> Pi
>
>
> On 5/20/08, pi song <pi.songs@gmail.com> wrote:
> >
> > How do we handle untyped executions?
> >
> > There are two ways:-
> >
> > 1. Stick dummy schema of ByteArrays in LOLoad.  This way is a bit ugly
> but
> > very simple to do.  Bad things: 1) This enforces maximum tuple arity
> for
> > untyped execution.  2) We also have to carry a number of schema fields
> to
> > downstream operators.
> >
> > 2. Implement all LOs to support null schema. This is more clean but
> > requires a bit more work. I still see some LOs like LOUnion that will
> die
> > immediately in getSchema() if one of the inputs have a null schema.
> >
> > I am more tempted to do (2) just because I want it to be clean but we
> could
> > do (1) first if we want to get the whole thing done quickly.
> > What about other people's opinion?
> >
> > Pi
> >
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message