Minimal definitions proposal

[Marcin 'Qrczak' Kowalczyk]

Fri, 17 Aug 2001 23:45:55 -0400, Dylan Thurston <dpt@math.harvard.edu> pisze:

> Here's one cheap proposal:
> 
> Add a "MINDEFINITION" pragma which takes a boolean expression whose
> variables are the class functions.  Since it's a pragma, this should
> only produce warnings if violated.

You mean an expression using && and || and method names? They are
indeed enough almost always.

In my experience they are not enough only if the default definitions
are correct for some uses and incorrect for others, e.g. "foo and bar,
and also baz if this is a set rather than a bag" or "succ and pred,
and also enumFrom* functions if the type doesn't fit in Int". Since
these side conditions are hard to express at all, in these cases the
programmer would simply not be warned.

I like it, except the name MINDEFINITION, although I tend to call it
"minimal definition". REQUIRED?

Although "Minimal definition: nothing, all default definitions will
make a correct instance" could be spelled by omitting the pragma,
an explicit "True" should be allowed too, so one reading the class
definition can distinguish between this case and "I didn't bother
with specifying what is required". I don't feel a need of having a
special shortcut for "all methods must be specified, despite some
having default definitions" - just name the methods.

-- 
 __("<  Marcin Kowalczyk * qrczak@knm.org.pl http://qrczak.ids.net.pl/
 \__/
  ^^                      SYGNATURA ZASTĘPCZA
QRCZAK