I don't like it either but seems like we just have to live with this 2 semantics 1 operator thing.
Does anyone have a better solution? Goals:- - Consistent semantic - Ease of use On Fri, May 16, 2008 at 2:30 PM, Utkarsh Srivastava <[EMAIL PROTECTED]> wrote: > > > > The first output column will have to be wrapped in Tuple whereas if we > > group > > by only one column we don't have to wrap. Is that the right logic? > > > > Yes, that is how it has worked so far. > > However, if I am not a big fan of this logic since it is confusing at > times and leads to several special cases in the code. I think it will be > cleaner to always wrap in a tuple. But that has 2 disadvantages: > > i) Will break backward compatibility > ii) Will lead to non-flat tuples which users won't be able to store > using default storage functions. > > Utkarsh > > > > > Pi > > > > On 5/16/08, Alan Gates <[EMAIL PROTECTED]> wrote: > > > > > > There really isn't any meaning to cogrouping with one field on one > > relation > > > and two fields on another. Given our definition of tuple, there > will > > never > > > be any tuples that match. I believe Santhosh has changed this to be > a > > > syntax error. > > > > > > Alan. > > > > > > pi song wrote: > > > > > >> Normally we do COGroup like this:- > > >> > > >> X = COGroup A By $0, B By $0 ; > > >> > > >> This first column of the output will be data atom. > > >> > > >> But if we do:- > > >> > > >> X = COGroup A By $0, B By $0, $1 ; > > >> > > >> What is the the first column then? I assume the B grouping will be > > wrapped > > >> to tuple and treated as atom. Am I right? > > >> > > >> Pi > > >> > > >> > > >> > > > >
