Path: utzoo!utgpu!water!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!bloom-beacon!apple!voder!pyramid!oliveb!sun!chiba!khb
From: khb%chiba@Sun.COM (Keith Bierman - Sun Tactical Engineering)
Newsgroups: comp.lang.fortran
Subject: Re: FORTRAN 8x ... An intermediate standard?
Summary: rebuttal
Keywords: FORTRAN 8x
Message-ID: <74002@sun.uucp>
Date: 21 Oct 88 17:08:59 GMT
References: <656@convex.UUCP> <6148@june.cs.washington.edu> <867@ncar.ucar.edu> <2874@sugar.uu.net>
Sender: news@sun.uucp
Distribution: comp.lang.fortran
Lines: 59


It appears that my last several postings have been eaten by the local
buffer monster ;->


In article <2874@sugar.uu.net>, ssd@sugar.uu.net (Scott Denham) writes:

> In hindsight, it would seem that exactly such an intermediate step might
> have avoided the mess that the FORTRAN standard is in now.  
>

No. The last couple of YEARS of commitee time dealt with this at GREAT
length. There has been ballot after ballot on deleting "unnecessary
features". EVERY FEATURE in f88 WAS DEMANDED BY ONE GROUP or ANOTHER.

f88 is not "too" ambitious. If anything it is too conservative (now
that we have abstract data types and overloading(*) what do we do
about overloading IO ?))

(*) Many are claiming that ADT and overloading is "not fortran...etc".
FORTRAN is supposed to be FORMula TRANslator. To illustrate why these
features are useful for "real" applications...

When I write

	U*D*U**T

It is understood that this means multiply an upper triangular matrix
by a diagonal matrix by a lower triangular matrix (basis of Kalman filtering)

The "determinstic time step" of a kalman filter requires

	U*D = PHI*U*D

where, by convention, UD contains the "factors" as a vector of length
n*(n+1)/2. PHI, of course, is a full square matrix.

With overloading, and ADT, I can define a datatype UD as a special
case of an upper triangular matrix. Overload * so that a square matrix
* a triangular matrix, returns a triangular matrix. Now the code looks
like the algorithm!

How do we do this in f77

      call PHIU(PHI,MAXPHI,IRPHI,JCPHI,U,N,W,IMAXW)
      call WGSG(W,IMAXW,IW,JW,DW,F,V,U)

Now which is closer to the formula ? Which will be easier to maintain ?

With generics I can even make everything work for double precision,
w/o changing the code. Is it really better to have

	call DPHIU()
	call DWGSG()

Which features of the proposal do you want to drop, why, and what does
it do to the consistency and expressive power of f88 ?
Keith H. Bierman
It's Not My Fault ---- I Voted for Bill & Opus