I'm not sure. If the type of two things that I am comparing (typically same field of tuples inside a bag) I expect it to throw an error instead of ordering the results by the datatype.
Because if it doesn't, it will either error out later on in the pigscript or it will be serialized out and some program and that program will read in offending field and crash. I'd prefer it fail early than late. Which is why I'm just casting to Comparable and calling compareTo. The problem with that is that it depends on each of the Comparable's compareTo method to handle errors in similar ways..... and I see that it does by calling into DataType.compare(circa l166 in DataByteArray for BYTEARRAY's...) ahh I see, so by casting to comparable it does the same as DataType.compare when the types are different. Hmmmm, I guess I want to stick to casting to Comparables, since the two ways of calling them are identical. Unless people have other comments. On Wed, Apr 28, 2010 at 3:57 AM, Gianmarco <gianmarco....@gmail.com> wrote: > Basically, DataType.compare() just calls the compareTo() method of the two > objects after checking that the two types are the same. > However, DataType.compare() does 2 things more than a simple compareTo(). > > Firts, it is specialized for Maps, for which sizes are taken into account > and keys are sorted. > > Second, it imposes an (arbitrary) order on different data types. In this > way > the types are not dependent on each other and there is a single point of > control. > > So I think you should use DataType.compare() unless you are sure you do not > need these features. > > Anyway, there is something that I do not understand. > > What I do not understand is why the function needs to switch on the > datatype > byte and cast the objects before calling the compareTo on them. Just > casting > them to Comparable and letting Java run the proper polymorphic method > should > work as well, right? > > > > > On Wed, Apr 28, 2010 at 07:12, hc busy <hc.b...@gmail.com> wrote: > > > guys, I'm implementing that ExtremalTupleByNthField and I have a question > > about comparison... > > > > > > So, when I have parsed out the two objects that I want to compare how do > I > > perform that comparison? My current implementation assumes the data is > > Comparable (which they invariably are within pig) so I do > > > > > > int c = ((Comparable)o1).compareTo((Comparable)o2); > > > > > > now I also see that there's another compare that compares the two objects > > by: > > > > > > int c = DataType.compare(o1, o2, DataType.findType(o1), > > DataType.findType(o2)); > > > > > > > > The initial methods works for all types I've tried (int, string, etc.) > But > > the latter is used by another UDF already in SVN. > > > > What are your suggestions? > > > > (PIG-1386 is ticket where I've checked in the patch). > > >
