Modeling With Data

Using a program as a library

PDF version

I often have this scenario: I have an analysis to do using some quirky data set. The first step in every case is to write a function to read in and clean the data. Along the way to doing that, I'll write functions producing summary statistics sanity-checking the data and my progress.

At this point I can get to the actual process of producing a descriptive model, and then testing that model's claims. This will all be in modelone.c

Next week, I have an idea for a new descriptive model, which will naturally make heavy use of the existing functions to clean data and display basic statistics. So how can I most quickly get to those functions while doing minimal damage to the original program?

In the context of C and many, many other systems, the only difference between a function library and a program is that a program includes a main function that indicates where execution should start. So the problem is basically in making sure that at compilation, there is exactly one version of main visible to the compiler at a time. Here are a few options, all of which are appropriate in some circumstances, though I'll focus on the last here.

Option one: Simply add more functions embodying the new model to modelone.c, and add a command-line switch to select the model.

Pros: immediate (especially if you don't use getopt to parse the command line). Cons: gets messy very fast. Something about a single several-page code file discourages reading.

Option two: Move all of the more useful functions in modelone.c to a second file, model_lib.c, write a header, and then #include the header in both modelone.c and the new modeltwo.c. Pros: very organized. Cons: can take time to do it right, which may not pay off for an isolated project.

Option three: Conditionally comment out the main function. Here is a skeleton for modelone.c:

void read_data(){ 
    ... 
}

#ifndef MODELONE_LIB
int main(){ 
    ... 
}

#endif

If MODELONE_LIB is not defined (note the use of #ifndef rather than #ifdef), then main will appear as normal, so you can compile modelone.c as normal.

If that variable is defined, then main will be passed over--and suddenly you have a library instead of a program. So modeltwo.c will look like this:

#define MODELONE_LIB
#include "modelone.c"

void run_second_model(){
    ...
}

int main(){ 
    read_data();
    run_second_model();
}

We've successfully used modelone.c, which had been a program file, as a library file.

Pros: you don't have to rewrite modelone.c, save for adding the if/endif. Thus, this is fast and doesn't require any re-testing of your original work. Cons: if you left a lot of globals floating around in modelone.c, you now have all of those globals floating around in your second model. This will be a good thing for some globals, but side-effects may creep in if you aren't aware of what else you're bringing in.

Adding main to a library

In the other direction, there's good reason to have a self-executing library: testing. Rather than writing the actual library and then a separate file for testing, just put all the tests at the bottom of the library file, along with a main routine to run them all. Here, the default usage is to not run main, so surround it with #ifdef RUN_LIB_TESTS ... #endif, and then define RUN_LIB_TESTS only during testing. You may be able to surround all of the testing functions in the #ifdef, so non-testing library users can't see any of the test functions at all.

You can define RUN_LIB_TESTS either via a #define line at the top of the file that you keep normally commented out, or during compilation, by specifying -DRUN_LIB_TESTS among the C flags to GCC (or via comparable means for other compilers).

[Previous entry: "Your genetic data--ethics"]
[Next entry: "Making integers two-dimensional"]