6 Some History

Long, long ago in a lab far, far away, I became frustrated with the time required to write command line parsing routines, even with getopt (), for all of the new programs that I developed. I felt that it was tedious work that offered too many opportunities to cut corners and produce error-prone code. In late 1997, the seeds of Genparse fell together in my head, and I released version 0.1, written in C, on New Year’s Day, 1998. The only output language supported was C, and it did not allow long options. Parameter types were limited to flags, integers, floats, and strings, and range checking was supported.

It soon became clear that although Genparse was fairly useful (I was already using it to create parsers for my own projects), it was severely limited and could use a number of additional features. With the help of a handful of people who provided feedback and constructive criticism, version 0.2 was soon released. New features included a more flexible Genparse file format (essentially, the current format) which was parsed by lex and bison rather than by my C code. Parameter descriptions and callback functions were now supported.

A few bug fixes later, version 0.2.2 was released. It was to remain current for over a year. In June 1999, I revisited Genparse to add the #include and #mandatory directives, as well as to clean up the code a bit and perform some minor bug stomping. The resulting version 0.3 was distributed with Debian/GNU Linux.

During my revision that produced 0.3, I became aware of the acute coding slop that had evolved. To call the output functions "spaghetti logic" would have been an understatement. Naturally, I needed to modularize Genparse, and the best way of doing so seemed to be a re-write in C++. In December 1999, I undertook this task. The code became separated into three logical sections:

• The parser, a bison grammar with supporting code.
• An internal representation, encapsulated in a C++ class.
• Output functions for each supported language, built into the C++ class.

This design greatly improved the extensibility of Genparse. In order to support a new output language, one only needs to add appropriate member functions to the C++ class. This implementation became version 0.4.

An important part of 0.4 was support for C++ as an output language. Rather than returning a struct containing the command line parameter values, a C++ command line parser would be encapsulated by a C++ class. The user would call the member functions of this class to access parameter values, making for a cleaner interface than C can support.

Version 0.4 also included a more comprehensive set of test suites along with the documentation that you currently are reading.

In Fall 2000, I revisited Genparse. Version 0.4 did not support options with no short form, and thus was limited to 52 options at most. Version 0.5 lifted this restriction, included <stdlib.h> rather than the depreciated <malloc.h> in output files, and fixed up user-defined include files so that they worked with quotes as well as "<>". Also, I finally figured out how to get Genparse to reliably compile on systems both with and without the getopt_long () function.

Version 0.5.1 fixed a few problems with the lexical analyzer that resulted in default values for floats not being used properly. This version also eliminated the mandatory use of the -q option.