At 04:44 PM 4/27/99 -0700, Randy Bush wrote:
I find it rather curious that QA procedures were not in place to prevent the bug from happening in the first place.
to paraphrase dijkstra because i am too lazy to look up the reference, testing can demonstrate the presence of bugs, it can not demonstrate their absence. i.e. if they could do what you suggest, there would be very few new bugs in the world. this would be a truly great advance.
So what's the problem in *testing* your software before putting in general release? Running the "upgraded" software on a testbed machine before putting it on-line? Dropping contact information is kinda noticable...it's not like this is some obscure, only-happens-on-wierd-configurations, blue-moon bug; this is pretty obvious. Hell, even *I* test a new part in a computer before closing the case and pronouncing it fit for use...you'd think NSI could do the same. It's basic common sense, after all.
I also find it curious that there are no rollback procedures in place to recover quickly from a bug in generating whois data.
good point. a possible explanation is that they changed the back end, and hence the front end. while one might roll back the front end, the back end could be much more difficult as o new updates had flowed in, i.e. can't just roll back, need to convert the data back, and o when people write database conversion code, they tend to think of it as one way, and do not double the cost by writing un-conversion code.
Or maybe testing their code? Or would that have been more expensive than writing a patch?
in private email, an acquaintance suggested that extensive alpha and beta testing might have caught it. with a product such as whois this might be hard. i.e. alpha/beta tests are usually done by shipping product to a select few. how would one do this with whois? with registration services? etc.? not saying one could not, just that this is far from trivial or obvious.
Hmmm...how about using a partial database to test the revision before deploying it? Say, just a random sampling to make sure everything works the way it should before you step on your privates in front of a few million people. That's certainly what I would have done! It's pretty obvious and yes, even trivial...one machine (or two, if you want to verify network performance).
randy, who spent 20+ years in software development
And never learned to test before deployment? I learned that in high school BASIC classes.... "Let there be light!"...and God invented Thomas Edison. Dean Robb Owner, PC-EASY (757) 495-EASY [3279] On-site computer repair, upgrades and consultations Lead SimOps columnist/reviewer on http://WWW.TheGamers.Net