HOL.thy:
"@" is no longer introduced as a "binder" but has its own explicit
translation rule "@x.b" == "Eps(%x.b)". If x is a proper pattern, further
translation rules for abstractions with patterns take care of the rest. This
is very modular and avoids problems with "binders" such as "!" mentioned
below.
let now allows pttrn (let (x,y) = t in u) instead of just idt (let x = t in u)
Set.thy:
UN, INT, ALL, EX, etc all use "pttrn" instead of idt. Same change as for "@"
above, except that "@" was a "binder" originally.
Prod.thy:
Added new syntax for pttrn which allows arbitrarily nested tuples. Two
translation rules take care of %pttrn. Unfortunately they cannot be
reversed. Hence a little ML-code is used as well.
Note that now "! (x,y). ..." is syntactically valid but leads to a
translation error. This is because "!" is introduced as a "binder" which
means that its translation into lambda-terms is not done by a rewrite rule
(aka macro) but by some fixed ML-code which comes after the rewriting stage
and does not know how to handle patterns. This looks like a minor blemish
since patterns in unbounded quantifiers are not that useful (well, except
maybe in unique existence ...). Ideally, there should be two syntactic
categories:
idts, as we know and love it, which does not admit patterns.
patterns, which is what idts has become now.
There is one more point where patterns are now allowed but don't make sense:
{e | idts . P}
where idts is the list of local variables.
Univ.thy: converted the defs for <++> and <**> into pattern form. It worked
perfectly.
Hints
=====
22-Oct-1993 MMW
20-Nov-1993 MMW
Some things notable, but not (yet?) covered by the manual.
- constants of result type prop should always supply concrete syntax
(elaborate on this in last sect of 'Defining Logics' (?));
- 'Variable --> Constant' possible during rewriting;
- 'trivial definitions' via macros (e.g. "x ~= y" == "~ (x = y)");
- patterns matching any remaining arguments are not possible (i.e. what would
be (f x y . zs) in LISP); e.g. HOL's @ (supposing it implemented via macros
which it is *not*): "@x. P" == "Eps(%x. P)", now the print rule doesn't
match things like Eps(%x. P, a, b, c);
- alpha: document the precise manner in which bounds are renamed for
printing;
- parsing: applications like f(x)(y)(z) are not parse-ast-translated into
(f x y z); this may cause some problems, when the notation "f x y z" for
applications will be introduced;