pig-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alan Gates (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (PIG-2224) Incorrect arity test in AstValidator.g with ALL and column-based grouping condition together in cogroup
Date Wed, 17 Aug 2011 19:49:28 GMT

    [ https://issues.apache.org/jira/browse/PIG-2224?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13086546#comment-13086546

Alan Gates commented on PIG-2224:

We should not allow "cogroup A by x, B all" in any situations.  The result is not well defined.
 "group B all" means collect all records of B into one bag.  "group A by x" means collect
records for each unique value in x into a bag.  Only in the case where x has the same value
for all of A would "cogroup A by x, B all" make any sense.  Since Pig cannot know a priori
whether this holds it needs to fail this query in all cases.

So I agree we should fix this logic in the parser.  But we should fix it so it fails whether
the ALL is first or second.

> Incorrect arity test in AstValidator.g with ALL and column-based grouping condition together
in cogroup
> -------------------------------------------------------------------------------------------------------
>                 Key: PIG-2224
>                 URL: https://issues.apache.org/jira/browse/PIG-2224
>             Project: Pig
>          Issue Type: Bug
>          Components: grunt
>    Affects Versions: 0.9.0
>         Environment: Suse Linux 9/MacOS(10.7)
>            Reporter: JArod Wen
>              Labels: ALL, arity, astvalidator, cogroup, grunt
>             Fix For: 0.9.1
>         Attachments: pig-2224.diff.patch, pig-2224.diff.patch
> When ALL and column-based grouping condition are used together in COGROUP, the arity
test in AstValidator.g (line 242) incorrectly sets the arity and causes exception. For example,
assume we have the follow two relations:
> a = load 'A' as (col_a_0, col_a_1);
> b = load 'B' as (col_b_0, col_b_1);
> The following statement will throw an invalidation error:
> c = cogroup a by col_a_0, b ALL;
> It is because when processing a:col_a_0, the arity is set to 1; then when processing
b:ALL, due to the null value in join_group_by_clause will emit arity 0 for the second relation,
and arity test fails. 
> Reversing the two relations will be a work-around for this error:
> c = cogroup b ALL, a by col_a_0;
> However it is a lucky shot: when processing b:ALL, since join_group_by_clause is null,
arity is still 0; then when processing a:col_a_0, arity will be initialized so no arity test
is done in this case (so it passes).
> The main reason is the omission of the consideration on ALL keyword during the arity
test. I attached a patch to fix this, by separating the arity test for both join_group_by_clause
and ALL. The patch is tested locally and it works.

This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira


View raw message