1.1 May You Solve Interesting Problems

The primary motivation for learning sed and awk is that they are useful for devising general solutions to text editing problems.[1] For some people, myself included, the satisfaction of solving a problem is the difference between work and drudgery. Given the choice of using vi or sed to make a series of repeated edits over a number of files, I will choose sed, simply because it makes the problem more interesting to me. I am refining a solution instead of repeating a series of keystrokes. Besides, once I accomplish my task, I congratulate myself on being clever. I feel like I have done a little bit of magic and spared myself some dull labor.

[1] I suppose this section title is a combination of the ancient Chinese curse "May you live in interesting times" and what Tim O'Reilly once said to me, that someone will solve a problem if he finds the problem interesting. [D.D.]

Initially, using sed and awk will seem like the long way to accomplish a task. After several attempts you may conclude that the task would have been easier to do manually. Be patient. You not only have to learn how to use sed and awk but you also need to learn to recognize situations where using them pays off. As you become more proficient, you will solve problems more quickly and solve a broader range of problems.

You will also begin to see opportunities to find general solutions to specific problems. There is a way of looking at a problem so you see it related to a class of problems. Then you can devise a solution that can be reused in other situations.

Let me give you an example (without showing any program code). One of our books used a cross-referencing naming scheme in which the reference was defined and processed by our formatting software (sqtroff). In the text file, a reference to a chapter on error handling might be coded as follows:

\*[CHerrorhand]

"CHerrorhand" is the name giving the reference and "\*[" and "]" are calling sequences that distinguish the reference from other text. In a central file, the names used for cross references in the document are defined as sqtroff strings. For instance, "CHerrorhand" is defined to be "Chapter 16, Error Handling." (The advantage of using a symbolic cross-referencing scheme like this, instead of explicit referencing, is that if chapters are added or deleted or reordered, only the central file needs to be edited to reflect the new organization.) When the formatting software processes the document, the references are properly resolved and expanded.

The problem we faced was that we had to use the same files to create an online version of the book. Because our sqtroff formatting software would not be used, we needed some way to expand the cross references in the files. In other words, we did not want files containing "\*[CHerrorhand]"; instead we wanted what "CHerrorhand" referred to.

There were three possible ways to solve this problem:

1. Use a text editor to search for all references and replace each of them with the appropriate literal string.

2. Use sed to make the edits. This is similar to making the edits manually, only faster.

3. Use awk to write a program that (a) reads the central file to make a list of reference names and their definitions, (b) reads the document searching for the reference calling sequence, and (c) looks up the name of the reference on the list and replaces it with its definition.

The first method is obviously time-consuming (and not very interesting!). The second method, using sed, has an advantage in that it creates a tool to do the job. It is pretty simple to write a sed script that looks for "\*[CHerrorhand]" and replaces it with "Chapter 16, Error Handling" for instance. The same script can be used to modify each of the files for the document. The disadvantage is that the substitutions are hard-coded; that is, for each cross reference, you need to write a command that makes the replacement. The third method, using awk, builds a tool that works for any cross reference that follows this syntax. This script could be used to expand cross references in other books as well. It spares you from having to compile a list of specific substitutions. It is the most general solution of the three and designed for the greatest possible reuse as a tool.

Part of solving a problem is knowing which tool to build. There are times when a sed script is a better choice because the problem does not lend itself to, or demand, a more complex awk script. You have to keep in mind what kinds of applications are best suited for sed and awk.