strange unit testing

Recently I had to extend a load balancing algorithm to take into account more variable (which they were isn’t pertinent here). Unfortunately the actual code to determine a candidate is in the middle of a long function, and that function can hardly be exercised without running the whole program and its communication relations.

I wasn’t inclined to extract that part into a separate function, because it would also needed to be in a separate file, and also I didn’t want to risk breaking anything in that step (and also I don’t think that testability should greatly affect program design).

So instead I went placing comments like

/*# BEGIN test1 */
/*# END test1 */

around the pieces of code that I needed to exercise and later modify, and wrote a ruby script to extract exactly those pieces into a file as named in the comment. The test code then includes those files (one each) after setting up the necessary scaffolding (including the variables used from other parts of that big function); the result then is compiled and run to perform the tests. (Part of the testing was to simulate the disk usage distribution caused by the load balancing by simulating everything for some time. Doing load balancing by a number of weighed factors can have counterintuitive results.)