[
https://issues.apache.org/jira/browse/PIG-4184?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Daniel Dai updated PIG-4184:
----------------------------
Attachment: PIG-4184-piggybank-1.patch
Going through the piggybank and fix the null handling part. Note those UDFs do
not handle null before, even without PIG-4184-1.patch. However, before
PIG-4184-1.patch, some NPE does not manifest if we don't feed null to UDF
directly.
> UDF backward compatibility issue after POStatus.STATUS_NULL refactory
> ---------------------------------------------------------------------
>
> Key: PIG-4184
> URL: https://issues.apache.org/jira/browse/PIG-4184
> Project: Pig
> Issue Type: Bug
> Components: impl
> Reporter: Daniel Dai
> Assignee: Daniel Dai
> Fix For: 0.14.0
>
> Attachments: PIG-4184-1.patch, PIG-4184-piggybank-1.patch
>
>
> This is the same issue we discussed in PIG-3739 and PIG-3679. However, our
> previous fix does not solve the issue, in fact, it make things worse and it
> is totally my fault.
> Consider the following UDF and script:
> {code}
> public class IntToBool extends EvalFunc<Boolean> {
> @Override
> public Boolean exec(Tuple input) throws IOException {
> if (input == null || input.size() == 0)
> return null;
> Integer val = (Integer)input.get(0);
> return (val == null || val == 0) ? false : true;
> }
> }
> {code}
> {code}
> a = load '1.txt' as (i0:int, i1:int);
> b = foreach a generate IntToBool(i0);
> store b into 'output';
> {code}
> 1.txt
> {code}
> 1
> 2 3
> {code}
> With Pig 0.12, we get:
> {code}
> (false)
> (true)
> {code}
> With Pig 0.13/0.14, we get:
> {code}
> ()
> (true)
> {code}
> The reason is in 0.12, Pig pass first row as a tuple with a null item to
> IntToBool, with 0.13/0.14, Pig swallow the first row, which is not right. And
> this wrong behavior is brought by PIG-3739 and PIG-3679.
> Before that (but after POStatus.STATUS_NULL refactory PIG-3568), we do have a
> behavior change which makes e2e test StreamingPythonUDFs_10 fail with NPE.
> However, I think this is an inconsistent behavior of 0.12. Consider the
> following scripts:
> {code}
> a = load '1.txt' as (name:chararray, age:int, gpa:double);
> b = foreach a generate ROUND((gpa>3.0?gpa+1:gpa));
> store b into 'output';
> {code}
> {code}
> a = load '1.txt' as (name:chararray, age:int, gpa:double);
> b = foreach a generate ROUND(gpa);
> store b into 'output';
> {code}
> If gpa field is null, script 1 skip the row and script 2 fail with NPE, which
> does not make sense. So my thinking is:
> 1. Pig 0.12 is wrong and POStatus.STATUS_NULL refactory fix this behavior (we
> don't need related fix in PIG-3739/PIG-3679)
> 2. ROUND (and some other UDF) is wrong anyway, we shall fix it
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
