[Python-ideas] Simplicity of C (was why is design-by-contracts not widely)

Stephen J. Turnbull turnbull.stephen.fw at u.tsukuba.ac.jp
Sun Sep 30 08:11:27 EDT 2018


Steven D'Aprano writes:

 > (1) Distance 

 > (2) Self-documenting code

 > (3) The "Have you performed the *right* tests?" problem

 > (4) Inheritance
 > 
 > Contracts are inherited, unit tests are not.

What does "inherited" mean?  Just that methods that are not overridden
retain their contracts?

 > (6) Separation of concerns: function algorithm versus error checking
 > 
 > Functions ought to validate their input, but doing so obfuscates the 
 > function implementation. Making that input validation a pre-condition 
 > separates the error checking and input validation from the algorithm 
 > proper, which helps make the code self-documenting.

There's nothing in the Eiffel syntax that distinguishes *which*
contract(s) was (were) violated.  Is there some magic?  Or does the
Eiffel process just die?  (ISTR that is a typical error "recovery"
approach in systems implemented in Eiffel, maybe from the Beautiful
Code book?)

 > (7) You can't unit test loop invariants

I don't see how a loop invariant can be elegantly specified without
mixing it in to the implementation.  Can you show an example of code
written in a language with support for loop invariants *not* mixed
into the implementation?

 > (8) Executable documentation

I don't see how any of your points fail to be satisfied by use of
asserts with a convention that (except for loop invariants) they're
placed either at the beginning or the end of the function, depending
on whether they're pre- or post-conditions.  This requires a single-
exit style which may sometimes be unnatural, of course.

AFAICS you can program in contract style already in Python.  Contracts
involving both beginning and end state would require annoying local
variables, definitely.  But other than that, all I see is a desire for
unnecessary syntax, or stdlib, support.



More information about the Python-ideas mailing list