design by contract versus doctest

[aku]

On 2004-04-05, Peter Hansen <peter at engcorp.com> wrote:
>  aku wrote:
> 
> > On 2004-04-05, Peter Hansen <peter at engcorp.com> wrote:
> > 
> >> aku wrote:
> >>
> >>
> >>>Isn't it so, that by applying DBC, a lot of unittests can be
> >>>made redundant?
> >>
> >> How would you propose verifying that your code will work if you
> >> don't run tests before you ship?
> > 
> > 
> > in debugging mode you obviously have to ensure that the caller
> > makes sure that all preconditions are met...then - by contract - 
> > the postconditions are met on returning from the function.
> 
>  What is "debugging mode"?
> 
>  You aren't actually talking about _manual_ testing of some
>  kind, are you?  

no I'm not

>  If not, I don't understand how you can be
>  talking about automated testing, yet claiming that unit
>  tests are redundant.  Or are you simply saying that proper
>  use of DBC can somewhat _reduce_ the number of unit tests
>  you have to write?  

yes - the latter. Indeed reduce, not omit. A nr of unittests will
still be necessary to ensure preconditions, but in
the (oversimplified) example I gave (n = 1), the test
"if n < 0: ..." isn't necessary anymore and certainly not *in*
the called function itself.

The "downside" with DBC I see is: when the caller *cannot*
ensure a precondition, then the behavior in the called
function becomes unpredictable. 

(In which case I ask how do you propose
>  to verify that your precondition contracts are written properly?)
> 
>  -Peter