Path: utzoo!utgpu!water!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!mailrus!ames!amdahl!uunet!pdn!alan
From: alan@pdn.UUCP (Alan Lovejoy)
Newsgroups: comp.lang.modula2
Subject: Re: more about ORD and VAL
Message-ID: <3638@pdn.UUCP>
Date: 3 Jul 88 17:43:42 GMT
References: <8806300811.AA00246@klaus.olsen.uucp>
Reply-To: alan@pdn.UUCP (0000-Alan Lovejoy)
Organization: Paradyne Corporation, Largo, Florida
Lines: 48

In article <8806300811.AA00246@klaus.olsen.uucp> Info-Modula2 Distribution List  writes:
>Pardon me, but what does the term "meaning-preserving coercion" denote?
>As far as I understand type coercions in other languages, I believe the
>current semantics are "meaning-preserving", that is, they preserve the
>meaning of the memory and that is all.  You cannot get rid of ORD, because
>it returns the "ordinal" value in terms of a type CARDINAL.  The type
>conversion CARDINAL merely subverts the type checking mechanism.  I know

I am sorry if the terminology I used confused you (or anyone).  And it
did confuse you.  The term coercion was used imprecisely.

In the current language, CARDINAL(-1) would have the value 65535 on most
systems.  This is a "bit-preserving type coercion" in the terminology
I usually use.  What the BSI committee proposed (a previous posting 
indicates they now have a new proposal), was that CARDINAL(3.0) become
a "meaning-preserving type conversion", which in the case of the example
would have the value 3.  The new proposal is that the syntax for this
type of conversion be CONVERT(CARDINAL, 3.0).  They want to avoid using
VAL for this since may compilers use it as a synonym for the
bit-preserving type transfers (even though Wirth's books define it as
a meaning-preserving conversion).  Perhaps they could use VALUE, instead
of CONVERT or VAL.  Although I liked the original proposal better (just
using the type name).

The proposed (and ISO accepted) syntax for bit-preserving transfers is
CAST(CARDINAL, -1).  CAST must be imported from SYSTEM.

Although bit-preserving type transfers are necessary for systems
programming, it is unfortunate that in the current language they are so
easily accomplished while the far safer and even more important meaning
preserving conversions are only haphazardly supported.  For example, there
is no standard way to convert between long and short types, INTEGERs and
CARDINALs, or REALS to INTEGERS (TRUNC returns type CARDINAL, for some
strange reason).  Even worse, the syntax for each such conversion that
is supported is different for each case!  This makes it impossible to
write portable code that converts one type into an equivalent value of
another type when the types involved may change from system to system,
because the types being converted are improperly and unnecessarily 
bound to the syntax of the conversion operator.  This should be fixed.




-- 
Alan Lovejoy; alan@pdn; 813-530-8241; Paradyne Corporation: Largo, Florida.
Disclaimer: Do not confuse my views with the official views of Paradyne
            Corporation (regardless of how confusing those views may be).
Motto: Never put off to run-time what you can do at compile-time!