Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!uwm.edu!wuarchive!cs.utexas.edu!rice!uw-beaver!fluke!mce
From: mce@tc.fluke.COM (Brian McElhinney)
Newsgroups: comp.lang.c++
Subject: Re: C++ Not Ready for Commercial Use
Message-ID: <11859@fluke.COM>
Date: 20 Oct 89 23:49:10 GMT
References: <24.UUL1.3#913@acw.UUCP> <2169@dataio.Data-IO.COM>
Sender: news@tc.fluke.COM
Organization: Software of the Mist
Lines: 22

In article <2169@dataio.Data-IO.COM> bright@dataio.Data-IO.COM (Walter Bright) writes:
>Prog Anal:	I always thought that most software metric analysis progs
>		were pretty worthless.

Have you ever looked at the "McCabe Cyclomatic Complexity Metric"?  I've
found, for C code at least, that it does a very good job at finding
error-prone portions of code.  There is also a very convincing article in the
Hewlett-Packard Journal (April 1989).  Basically, the McCabe metric turns code
into a graph, where each line of code is a node (the edges are the control
flow between nodes).  Applying some math from graph theory, you get a small
integer number that "measures" complexity.  From experience, a number greater
than 10 is usually worth investigating/testing/redesigning/whatever.

Unfortunately, I'm sure the McCabe metric is not going to be directly useful
for C++ code.  It is limited to the complexity of a single function.  C++ gets
complicated in the inter-relationships of methods, and of course, of classes.
Methods are typically not very complex.  Still, the same approach should be
useful for "measuring" the complexity of a class or set of classes.
 
 
Brian McElhinney         "If always without desire, one can observe marvels; if
mce@tc.fluke.com          always desireous, one sees merest traces."    Lao-Tzu