Re: Re: Re: High level thoughts

Greg Ewing wrote:

>...
> Experiences I've had in getting C libraries from different sources to
> work together has led me to the view that the best way to ensure
> interoperability between components is to design them to use nothing
> but bog-standard, language-defined data types and structures when
> interfacing with the outside world.

That is the best you can do when components have no common context other 
than the language. e.g. a random GUI libarary integrated with a random 
networking package. But every framework in the universe DOES define 
standardized interfaces. You're a Mac user so I'm sure you know about 
the various standardized interfaces like "KVO" and "Drag data source".

Every big enough program also grows many standardized interfaces.

Plus there are standardized protoocols in the Python world like SAX and 
DBAPI.

Given how common protocols and interfaces are in real programs (to the 
extent that people fake them and hack them) I do think that the language 
or library requires first-class support for them. In particular I think 
that their implications for unit testing are important because a 
contract enforced (as much as possible) with tests is much more reliable 
than one that is just expressed in prose.

> In Python terms, that means sticking to the basic, well-known
> protocols -- number, sequence, mapping, iterator, file, etc.  

If that were possible there would be no DBSIG API, DOM, SAX, EJB, ADO 
Datasource, etc.

  Paul Prescod