I'm not 100% certain that I understand the problem. Let me describe a situation that I faced, and my solution -- if it is relevant to your problem, I can provide more details.
We are running behavioral tests to study how well people can track sounds (in the dark). We perform a series of trials where we move speakers around, play sounds, flash lights, ask the subject to point with a Laser Pointer (mounted on a gimbal so we know where it is being pointed), etc. We design the Experiment using an Excel Worksheet where each Row is a Trial and each Column is one of 70-100 parameters that can be independently varied (how many sound pulses, how much time between pulses, how long is the pulse, what is its frequency range, etc.).
We need to "capture" the data from the Excel row to give us the "parameters" for the Trial. Partly because some parameters are integers, some Floats, some strings, we designed a "Trial Cluster" that basically mimics the Excel row, each element being named from the name of the corresponding Excel Column. Our Experiment Header File is an XML file that has an entry for every Trial consisting of this Cluster, saved as XML.
And then one of the Students wants to do something "a little different -- I want to use Headphones instead of movable speakers, so I'll need an additional 10 parameters added to the WorkSheet to hold Headphone-specific parameters". Great! That means a redesign of the Cluster (grrr).
When we originally designed this system, we wanted to keep things as simple as possible, and avoid humongous Worksheets if certain Trial Cluster parameters were not being used in the Study. So we said "If you are not using a Parameter, simply delete that Column from the WorkSheet". When we start the program, we read the WorkBook, parse the first row of the Trials WorkSheet, and map the Excel Columns to the Trial Cluster using the Column "names" contained in the first Row. We also flag which Cluster Elements are not present, and assign them a suitable Default Value (typically <blank> for Strings and 0 for numerics, though we use the unsigned equivalent of -1 for Delays, or "Delay forever" (=Don't Do That). So we can easily accomodate missing Columns, but how do we accomodate new columns?
Simple (sort of). We redesign a new "Master Excel WorkBook" with all of the Columns present, then run a LabVIEW Scripting routine that parses the WorkBook and creates for us a brand new Trial Cluster (we also increment the Version Number of the software -- at present, we are on our fourth Excel-driven Version). Because our Cluster is defined by a LabVIEW TypeDef, adding elements to the Cluster has a minimal impact on the existing Code (though we do check to make sure "Trial Delay" doesn't get substituted for "Pulse Width" within the code) -- we just have to add new code to handle the new Cluster elements.
There are a few more "complications", particularly at the Analysis end, but as the Software has progressed through its four Versions, experiments designed for the first Version's WorkBook continue to run just fine (remember, if a parameter isn't specified in the WorkBook, it is given a default value that the software interprets as "Do not do this"). Since we've "only" gone through four Versions, I've only had to run the Create Trial Cluster From Excel routine four times (note -- I ran it quite a few more times -- debugging, demonstrations, etc.), and "propagate" the new TypeDef throughout the code a few times. Made updates "relatively painless" (I'd say a week or three to implement and test).
And we didn't need to make any changes to the XML for saving data. Of course, when reading back the data, we did need to know which Version of the code wrote the data so we could use an appropriate Trial Cluster to convert the XML back to something that "made sense". That's when I started using LabVIEW Libraries (I'm a relative new-comer to them ...).
Bob Schor
