Approach to long and complex SAS coding

[don_mcmahon]

I too use the approach of organizing my programs by import -> coding -> reports -> stats -> graphs.

To organize, each section and subsection starts with a comment with two asterisks in outline format. Example:

** 1.0 Import source data;

** 1.1 Import Questionnaires;

** 1.2 Import Labs;

...

** 2.0 Recode and transform data;

** 2.1 Collapse categories;

** 2.2 Normalize skewed distributions;

I then have a separate program that takes the name of a SAS program as it's input and then prints out the program outline as defined by the text following the double-asterisks. This is simple text parsing, but produces a document that provides a road map to the entire program.  It doesn't help with the problem of running specific sections of code. But helps a when you need to understand a program you wrote a long time ago.

...don

%macro pgmdoc(fn);
filename pgmfile "&fn";

data _pgmdoctemp_;
attrib Outline length=$255 format=$145.;
infile pgmfile linesize=255;
input Outline &;

if scan(upcase(Outline),1) ='DATA' then do;            Outline='        '||trim(Outline); output; end; else
if scan(upcase(Outline),1) ='PROC' then do;            Outline='        '||trim(Outline); output; end; else
if scan(upcase(Outline),1) ='SET' then do;            Outline='            '||trim(Outline); output; end; else
if scan(upcase(Outline),1) ='MERGE' then do;        Outline='            '||trim(Outline); output; end; else
if scan(upcase(Outline),1) ='LIBNAME' then do;         Outline='        '||trim(Outline); output; end; else
if substr(upcase(Outline),1,4)='%LET' then do;         Outline='        '||substr(Outline,1,length(outline)); output; end; else
if substr(upcase(Outline),1,8)='%INCLUDE' then do;  Outline='        '||substr(Outline,1,length(outline)); output; end; else
if substr(Outline,1,3)='** ' then do;                Outline=trim(substr(Outline,4,length(Outline)));output; end;
run;

title 'SAS Program Documenter';
title2 "&fn";
proc print data=_pgmdoctemp_ noobs;
var Outline;
run;
%mend;

[Reeza]

@don_mcmahon You may want to take a look at PROC SCAPROC if you aren't already familiar with it.

[Peter_L]

I use most of the methods mentioned by other contributors plus a few others.

.

Break code into cohesive sections.

Make the sections have as low a coupling as possible.

Use macro variables to parameterise code. There is a downside: too many global variables (= coupling).

.

Save code as "modules" and %include them as needed.

Write pre- and post- assertions for each module formally to define what it does.

Include the unit test code with every module commented out at the end.

.

I write the documentation in the code marked with special comment symbols of /** or **; A simple awk script then processes these into html in a similar way to javadoc.

.

Replace as much DATA step merge code with PROC SQL as possible. It makes the code shorter and easier to read.

Use data-driven methods as much as possible. This tends to make things simpler, shorter, more reliable and easier to test. SQL helps here.

Replace macro logic with DATA step logic and data-driven methods. DATA step logic is far easier to debug and quoting is not a problem.

.

To make your code easier to read change the editor to show a light grey background to comments. The colour extends to the screen width. Then use comments to break up sections of code. You will be surprised how big a difference it makes.