This is a classic example of how the decision to use eval to access variables just makes code more complex and introduces more problems than it solves. The name MATLAB comes from "MATrix LABoratory": when you put your data into matrices/arrays then this problem is easy to solve. Thus the first step is to simply put the input data into one cell array (which they should be in anyway, rather then in separate variables):
>> nameOrder = {'sur','given'};
>> sname = 'Smith';
>> gname = 'Jane';
>> tmp = {gname,sname};
>> [~,idx] = ismember(nameOrder,{'given','sur'});
>> sprintf('%s_%s',tmp{idx})
ans = Smith_Jane
Simple, reasonably efficient, and it avoids all of the pointless problems of eval. I changed the variable names and char vectors to make it clear that the nameOrder contents are not the same as the variable names. You could simplify it even more by storing the names in a standard format, e.g. perhaps a structure:
S.surname = 'Smith';
S.givenname = 'Jane';
tmp = {S.givenname,S.surname};
Note that using a cell array has the advantage that you could write code which takes into account cultures which have multiple or only one name, all of which could be handled trivially with indexing (and not an eval in sight!).
Summary: keep data together as much as possible, rather than splitting it apart. Keeping data together makes it easier to work with.
