The Mythical Man-Month: Conceptual Integrity
November 20, 2011
Aside from being a fascinating inside-look at some of the challenges faced by the mainframe programmers of the sixties, The Mythical Man-Month presents many lessons-learned that are no less applicable today. This is the second article in a series exploring some of these lessons, in particular: conceptual integrity.
"Conceptual integrity" is one of the most important concepts in the book. Many of Dr. Brooks's recommendations throughout the book are predicated on the importance of ensuring a system's coherence, from the project management to user interface, from documentation to memory management optimizations. Brooks
"contend[s] that conceptual integrity is the most important consideration in system design. It is better to have a system omit certain anomalous features and improvements, but to reflect one set of design ideas, than to have one that contains many good but independent and uncoordinated ideas." (p42)
Brooks talks about simplicity being more valuable than diversity of function – because it is easier to construct and utilize a coherent, slightly simpler system than to create and use a complex one. How often have any of us opted for the simpler device rather than the gadget with all the bells and whistles, simply because we were concerned that the extras would break, or would be hard to learn? The same goes with software systems: keep the design simple. Only add as much diversity and complexity as are needed to get the job done. I suspect that this approach was influential in the formation of Agile principle number 10: "Simplicity – the art of maximizing the amount of work not done – is essential."
I think the author would agree that this principle is to be applied to the system generally, and must be applied very carefully when looking at individual components. As I read this section, I reflected back on mistakes made a few years before, when I first came onto a project in a file-processing environment. I couldn't articulate a significant problem until reading this: none of the .Net developers involved understood the system architecture, and therefore our solutions were incoherent. We had diverse ways of reporting errors (if any), of accepting input parameters, of creating output data.
Even as some of the individual lessons were learned, the bigger picture was not realized. Only after reading this book did I realize why some people have been less-than-thrilled about our use of Reporting Services for some new reports, instead of plain text, and I just installed a new application that does the same. We did so in order to have an easier time designing the report, and to have better looking output (simpler for the programmers). That's nice, but not at all what our users are used to, expected, or wanted.
What they want is something that prints the same way as all of the existing file-based reports. They want to be able to find the file on the system and reprint it when the existing paper copy gets chewed up. They don't want to learn how to go to Reporting Services to print the report; plus, we made the mistake of using the .Net version (RDLC). If reprints are needed, we'll have to either output a permanent PDF, or support two versions – the other loaded into SQL Server's Report Manager. Further, I've realized that testing is simplified when the report is in text: then it is easy to regression test the application, using a file diff tool to compare the output before and after an update.
We were the new kids on the architecture block, and we didn't get it. Sometimes we still don't. Brooks makes an analogy about ancient cathedrals. That RDLC report – it's like repairing a section of a Gothic cathedral by building a cantilevered steel-and glass structure. It might technically fit the need, but it probably won't be what anybody wanted.
TrackBack URL: http://www.safnet.com/fcgi-bin/mt/mt-tb.cgi/51