jcornado
07-23-2003, 09:09 AM
2 cnts,
Doing things manually now... and forever with the subsequent human error due
to tediousness and increasing test cycles. Not a good idea for the long run.
But who said that the software companies ever looked beyond the current
project? :-}
Have you thought about looking at it from the user perspective? A lot of
these variables will not likely be experienced by the average-joe-user.
An approach could be narrowing down the universe to those variables most
lilkely to be experienced by the normal user (whoever they might be: hard
core coders or internet surfers)
This would be an starting point that will hit most of the situations that
will affect support.
Testing and automating that would keep support and your project manager
happy while giving you the slack to increase the test base to address those
more-advanced but fewer in number users.
Matt Irwin wrote in message ...Hello,I've been trying to develop some automated testing for a very complexpiece of software. I've done some research into various methods and Iwould like to complete some model-based testing to verify thebehaviour of the program. Much of the material that I've come acrossrecommends constructing a Finite State Machine (FSM) and having aprogram generate random paths through it by selecting inputs andtriggering state changes. In this case, because the states are definedat each point, the output always known and can be easily verified.My problem is this: The function I am testing actually has a trivialstate machine. Inputs are set, the function executed, and the outputI'm interested in is immediately provided. The complexity in this casearises from the infinite ways to set up the test:a) Global configuration optionsb) Function switchesc) Test data based on boundary conditionsd) The environment with which this data will interact.So in order to develop a system to look for errors in this process, Ihave to determine the outcome for a very large number of conditions.If I sit down and calculate these outcomes line-by-line in a table, Imight as well be completing the testing manually.I've thought about testing each option acting independently first andthen moving on by forming intelligent groupings. Anyone have anythoughts on how I could incorporate model-based testing and have thesystem generate and verify cases?Thanks,Matt Irwin
Doing things manually now... and forever with the subsequent human error due
to tediousness and increasing test cycles. Not a good idea for the long run.
But who said that the software companies ever looked beyond the current
project? :-}
Have you thought about looking at it from the user perspective? A lot of
these variables will not likely be experienced by the average-joe-user.
An approach could be narrowing down the universe to those variables most
lilkely to be experienced by the normal user (whoever they might be: hard
core coders or internet surfers)
This would be an starting point that will hit most of the situations that
will affect support.
Testing and automating that would keep support and your project manager
happy while giving you the slack to increase the test base to address those
more-advanced but fewer in number users.
Matt Irwin wrote in message ...Hello,I've been trying to develop some automated testing for a very complexpiece of software. I've done some research into various methods and Iwould like to complete some model-based testing to verify thebehaviour of the program. Much of the material that I've come acrossrecommends constructing a Finite State Machine (FSM) and having aprogram generate random paths through it by selecting inputs andtriggering state changes. In this case, because the states are definedat each point, the output always known and can be easily verified.My problem is this: The function I am testing actually has a trivialstate machine. Inputs are set, the function executed, and the outputI'm interested in is immediately provided. The complexity in this casearises from the infinite ways to set up the test:a) Global configuration optionsb) Function switchesc) Test data based on boundary conditionsd) The environment with which this data will interact.So in order to develop a system to look for errors in this process, Ihave to determine the outcome for a very large number of conditions.If I sit down and calculate these outcomes line-by-line in a table, Imight as well be completing the testing manually.I've thought about testing each option acting independently first andthen moving on by forming intelligent groupings. Anyone have anythoughts on how I could incorporate model-based testing and have thesystem generate and verify cases?Thanks,Matt Irwin