[
https://issues.apache.org/jira/browse/PIG-2153?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13060298#comment-13060298
]
Pradeep Kamath commented on PIG-2153:
-------------------------------------
I don't have full context and given that I have not actively looked at Pig code
in quite a while, my comments should be taken with a grain of salt. I am
assuming POProject.getNext(Tuple) is being called because the schema (of load?)
says that a tuple field should be projected. If that is indeed the case, then
shouldn't the LoadFunc be returning a Tuple (with the bag in it)? The outer
tuple that the LoadFunc returns simply represents a record and does not count -
the types of the fields inside the outer tuple are the ones that matter in the
schema and if the schema says there is one field of type Tuple, then POProject
would except a type Tuple - so am wondering if the cast is correct as it is.
Again, I have been out of touch with Pig for a good 8 months now - so my
thinking above could be completely wrong :) - hopefully the more active Pig
committers can confirm/refute my hypothesis.
> POProject throws an error with tuples containing a single non-tuple field
> -------------------------------------------------------------------------
>
> Key: PIG-2153
> URL: https://issues.apache.org/jira/browse/PIG-2153
> Project: Pig
> Issue Type: Bug
> Affects Versions: 0.8.1
> Reporter: Ken Goodhope
>
> When POProject.getNext(tuple) processes a tuple with one field, the field is
> pulled out. If that field is not a tuple, a cast exception is thrown. This
> is happening in the folliwing block of code at line 401.
> if(columns.size() == 1) {
> try{
> ret = inpValue.get(columns.get(0));
> ...
> res.result = (Tuple)ret;
> I am seeing this error in a unit test that is loading an array of floats.
> The LoadFunc is converting the array to bag, and wrapping the bag in a tuple.
>
> ({(3.3),(1.2),(5.6)})
> This results on POProject attempting to cast the bag to a tuple. Looking at
> the code, it appears that if I wrapped the previous tuple in another tuple,
> then it would work.
> (({(3.3),(1.2),(5.6)}))
> In this case it would work because POProject would extract the first inner
> tuple and return it. But this would require the LoadFunc to check for tuples
> with a single non-tuple field and only wrap those.
> This could be fixed by first checking that the tuple does actually wrap
> another tuple.
> if(columns.size() == 1 && inpValue.getType(0) == DataType.TUPLE)
> {...
> I don't know the original intent of this code well enough to say this is the
> appropriate fix or not. Hoping someone with more Pig experience can help
> here. Right now this is preventing the unit tests in AvroStorage from
> working. I can change the unit test, but I think in this case the unit test
> is catching a real bug.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
