[Mondrian] Problem with aggregate tables..

[John V. Sichi]

We're looking forward to being able to reuse shared dimensions within 
virtual cubes, so we're certainly glad about the work Will is doing. 
It's just that taking on intermediate regressions makes planning very 
difficult.

For example, LucidEra also has a release planned for end of Jan, which 
needs to stabilize towards end of Dec.

We already use a stabilization branch for this purpose, but very soon we 
were hoping to

- have Rushan check in her COUNT DISTINCT optimizations (another 
significant Mondrian roadmap item) into Mondrian main

- pass our regression tests (functionality and performance) on Mondrian main

- integ latest from Mondrian main to our stabilization branch

- continue with any stabilization needed there while Mondrian main moves 
forward

Now, the second step may have just gotten a lot more complicated.

So, what I'm proposing is this.  If someone has followed the process and 
submitted a regression test for something they expect to have working 
forever in the future, then anyone else who wants to disable that test 
as part of some other change needs to raise the issue on the mailing 
list first, and get it resolved before going ahead with checkin.

Would that be acceptable?

JVS

Julian Hyde wrote:
>> JVS wrote:
>>
>> In eigenchange 10203, Will Gorman disabled one of the tests 
>> Rushan had 
>> added earlier as part of a very difficult fix for a serious bug 
>> (SharedDimensionTest.testVirtualCube).
>>
>> Will created a sf.net bug to track this regression from his change 
>> (1833528), but I thought the idea was that if we create regression 
>> tests, we're buying ourselves protection against regressions? 
>>  In other 
>> words, if progress needs to march on (maybe the test was relying on 
>> something it shouldn't have), the case can be discussed, but 
>> shouldn't 
>> that happen (with buy-in) before disabling the test?
>>
>> We make extensive use of virtual cubes, as well as NECJ, so 
>> when I see 
>> this, and the other one (1833526) logged at the same time, I get very 
>> uneasy.
> 
> We're all fighting for the same side here. Will's change was a significant
> rework of how we handle shared dimensions, with the explicit aim to improve
> how we handle virtual cubes. These bugs will be fixed before mondrian 3.0 is
> released (Dec 2007, may slip to Jan 2008).
> 
> Julian
> 
>