Re: Testing relational databases

From: S Perryman <a_at_a.net>
Date: Tue, 11 Jul 2006 08:40:22 +0100
Message-ID: <e8vklb$7mj$1_at_emma.aioe.org>


"Phlip" <phlipcpp_at_yahoo.com> wrote in message news:_Susg.129129$dW3.26057_at_newssvr21.news.prodigy.com...

>>S Perryman wrote:

>>> For example, Praxis CS (safety-critical s/w developers) showed on one
>>> project
>>> quite significant ROI on defect prevention and detection by using
>>> proof-based
>>> techniques, compared to the usual test approaches (unit testing etc) .

> That compares proof-first to test-last.

1.It does nothing of the kind. Here is the paper :

http://www.praxis-his.com/pdfs/cost_effective_proof.pdf

In it they compare the costs/ROI of formal proof vs unit tests of any kind. The results are quite telling.

2. If you wish to persist with your muppetry, then please feel free to cite references (publications akin to the above etc) for the following hypothesis.

For *any* unit test U, writing U *before* any implementation is provided (TW1) , as opposed to writing U *after* the implementation (TW2) :

  • results in TW1 producing U much faster than TW2
  • any U produced via TW1 will find more defects than the equivalent U produced via TW2

I'm sure the testing community will love to be shown that their unit tests can be produced more quickly, and find more defects, if only they had written the tests first ...

Regards,
Steven Perryman Received on Tue Jul 11 2006 - 09:40:22 CEST

Original text of this message