[Mondrian] RE: Cron <jhyde at marmalade> nice bash --login/home/jhyde/open2/mondrian/testAll --nightly

[Julian Hyde]

> Robin Tharappel wrote:
>
> I went ahead and changed the test to use a NonEmptyCrossJoin 
> instead of
> a CrossJoin. Using the NonEmptyCrossJoin still reproduces the bug
> reported with tracker 1630754 and no longer occurs with the Mondrian
> tip. The test case passes with the mondrian.native.nonempty.enable set
> to true and false.
> 
> However I was not sure if changing the mondrian.native.nonempty.enable
> to false should result in a different result set size if a 
> CrossJoin is
> used. The customer dimension has a total of 10281 members with 5581
> members being non empty. Using a CrossJoin you would expect the result
> set to have a row for all 10281 customers. However if the
> mondrian.native.nonempty.enable is set to true only 5581 rows are
> returned (using the same CrossJoin query). 

Mondrian should produce the same results regardless of the value of
mondrian.native.nonempty.enable, so what you are seeing is a bug, not merely
a problem with a unit test.

Can you log a new bug for this issue. Re-enable the test, but leave it
disabled for this specific case:

    public void testBug1630754() {
        if (!MondrianProperties.instance().EnableNativeNonEmpty.get()
            && !Bug.BugXxxxFixed) {
            return;
        }
        ...
    }

BugXxxxFixed will be a new boolean constant in Bug.java named after the bug
you log.

Julian