I noticed that in 0.8.x and earlier pig.jar , hadoop classes such as
FileInputFormat are also packaged,

then there is a potential problem that the behavior specified in the pig
jar could be different from those in
the hadoop installed on the running box.  for example if we assume that
some params should work since
they are specified by the system hadoop installation, but the classes in
pig jar do not provide them:

I see that in the latest github source code for pig, 2 jars are generated,
one without hadoop. I hope this is the direction
for the future.

thanks
Yang

Yes, we use pig-withouthadoop.jar starting from 0.9. pig.jar is only for
local mode.

On Sat, Jun 16, 2012 at 7:55 PM, Yang <[EMAIL PROTECTED]> wrote:

> I noticed that in 0.8.x and earlier pig.jar , hadoop classes such as
> FileInputFormat are also packaged,
>
> then there is a potential problem that the behavior specified in the pig
> jar could be different from those in
> the hadoop installed on the running box.  for example if we assume that
> some params should work since
> they are specified by the system hadoop installation, but the classes in
> pig jar do not provide them:
>
>
>
> I see that in the latest github source code for pig, 2 jars are generated,
> one without hadoop. I hope this is the direction
> for the future.
>
> thanks
> Yang
>