forums.silverfrost.com

[JohnCampbell]

You say you have some experience with unit tests in C# and C++, so you should have some understanding of what you are asking.

I am really puzzled as to what sort of answers to this question you were expecting, given you describe the program as a huge piece of code developed in Fortran 77, presumably quite old.

If you intend to break up the code into smaller units and define some performance requirement for each part, devise sufficient input tests (in isolation?) and then validate the output, the most plausible insight may not be appreciated !

[JohnCampbell]

I would find it difficult to believe the task is as simple as you propose.

I would expect that the development of the "huge piece of code developed in fortran 77" would have involved considerable testing and adaptation over the years to get to where it is now.
To approach any realistic form of sign-off would involve replicating the testing process that occurred during that development.

If you have a legal requirement for this, it sounds like a huge task, especially for defining the range of testing that would be required..

Without knowing the field of analysis, a more likely alternative, is to treat the whole package as a unit and the testing may be via a (previously) developed set of sample data sets and published answers.
Given the changes to numerical precision (eg "old hardware or PC 80 bit vs 64 bit" or "serial vs vector or parallel") there will be noticeable differences in results, which will need to be assessed as reasonable difference vs unacceptable.
Acceptance testing for this might be very different to your past experience, given the way you have framed the question.