Re: Context in data decl

[fplangc-request@cs.ucl.ac.uk]

From: fplangc-request@cs.ucl.ac.uk
Date: 15 Feb 91  5:57
Sender: Joe Fasel <jhf@chaco.c3.lanl.gov>
To: fplangc@cs.ucl.ac.uk, peterson-john@cs.yale.edu
Subject: Re: Context in data decl

I agree with Lennart's observations about why we should want to
keep contexts for data declarations: that the abstract data type
solution doesn't allow pattern matching, and the "Hello" :+ "World"
problem.

I was mystified by John Peterson's statement that niladic constructors
of data types with contexts are a typing problem, until he
explained that the problem is with the single-overloading rule.
As he said,

	data (Eq a) => Foo a = C1 a | C2

is problematical, because

	x = C2

runs afoul of the single-overloading rule.  Actually, so does

	f = C1

although this can be fixed:

	f x = C1 x

Essentially, the constructors of a type like Foo are overloaded constants.
So, here is another case in which the single-overloading rule is
an annoyance.  I suggest that we keep contexts for data declarations
and nuke the single-overloading rule.

I'll say more about this tomorrow.  Right now, I've gotta go crash.

--Joe