System and UAT testing

general discussion
1

In the past I have worked for clients where I have been involved in the whole spectrum of B.O. implementation from initial definition of user requirements, through Universe design and report development, to final roll out. An integral part of this process has been to test the definition of all objects and the data that measures returned (usually against existing data sources).

I am now working for a client that wants me to develop formal System and UAT test processes for other staff to utilise, and I’m having difficulty in deciding what degree of detail should be adopted.

For example, if a Universe / Object has been incorrectly defined it is possible that a measure might return incorrect results only when combined in a query with certain other measures. i.e. it’s okay when used in isolation. But if your Universe contains 100 measures, do you need to test each measure with every other combination of the other 99? Do you just restrict testing to measures within the same context etc. etc.?

In the past when testing against my own Universes, I knew instinctively the likely problem areas, and what to test, albeit in a rather ad hoc fashion. I.m now in a situation where my knowledge of the Universe design is far more limited, and I don’t have access to Designer.

I’d be very interested to hear of other peoples’ experiences and procedures that they have adopted. Has Business Objects ever produced white papers regarding best testing practices?

Thanks.

anorak :uk: (BOB member since 2002-09-13)

2

Hi Phil,

I guess we all have the same problem. As you mention Testing all combinations is just not practical. But it is the only one to guarantee that no bugs/errors exist in the universe.

We typically test for the following:

Measures and accurate aggregations (when combining with unexpected objects)
Contexts, Multiple contexts
Data provider synchronization
Filters
All objects with functions (or more that a select with other objects)

This is no guarantee but usually should uncover most of the issues

But you are never sure of anything.

We had one recently on Sybase where all the test where OK and it failed in production because people had more than 600 character of data in the group by function.
One test we now run it make sure that the unique key’s are unique (I know they should be but we had a few cases where this was not the case)

All this can only be best effort. It may still be that you leave some errors, and the may not be from BO but from the source.

ClaireB :de: (BOB member since 2002-08-09)