On Wed, 30 Jun 2004 21:08:27 CDT, Paul Schmehl <pauls@xxxxxxxxxxxx> said: > I attended a presentation yesterday for a security product in the > application firewall field. During the presentation, the CISSP stated that > "in every 1000 lines of code there will be 15 errors". I don't know if I'd > agree with that - I suspect most coders are a bit better than that - but I > had to chuckle, because, of course, I immediately thought, "So you admit > that your code is riddled with holes!" Actually, I suspect most coders are *worse* than that. Sendmail 8.13.0 weighs in at just about 90K lines of C code for the main program. By that metric, there should only have been 135 bugs in it. In fact, there are 441 occurrences of 'Problem noted by' in the release notes. BIND 9.2.3 has 1,525 entries in the CHANGELOG file, of which 774 are listed as '[bug]' entries. I'm fairly sure that BIND9 is well under 510,000 lines of code, so again we're running well above 15 bugs per KLOC. So either (a) Sendmail and BIND were written by people who were *incredibly* worse than "the average programmer", or 15 errors/KLOC is a vast understatement. Now although Sendmail may not be a paragon of excellent programming practice, it would be hard to argue that it's literally 4 times as buggy as code written by "the average programmer" - think back to your "intro to programming" class and ask what the *lower* half of the class would have done if they had done a rewrite of Sendmail... ;) I might be willing to accept 15 *security-critical* errors per 1,000 - the vast majority of bugs are *not* a security issue.
Attachment:
pgp00007.pgp
Description: PGP signature