Path: utzoo!attcan!uunet!husc6!bloom-beacon!tut.cis.ohio-state.edu!mailrus!ames!pasteur!ucbvax!phast.phys.washington.EDU!SEYMOUR
From: SEYMOUR@phast.phys.washington.EDU
Newsgroups: comp.os.vms
Subject: VMS Fort v4.8 bug PATCH
Message-ID: <8805232008.AA23674@ucbvax.Berkeley.EDU>
Date: 19 May 88 17:58:00 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 38

Two weeks ago I reported a bug concerning corruption of Comment lines
beginning with lowercase c if the /D_LINE compiler switch was used.

Fortran v4.8 has about 7 major bugs as distributed.
They've had so many SPRs that they're bypassing (read: overloaded)
the normal response center on the east coast and having Colo Springs
generate patch kits (read: complete replacement distribution) too.
I SPR'd it (by mail), received a phone call one week later and the
tape the next day.

The original (buggy) Fortran is v4.8-276
the repaired Fortran which i received is v4.8-283

(if you get a NEWER version, why not tell INFO-VAX, along with
a description of the bug that got it.  Then we can SPR for a new copy
(of ANY software product, not just Fortran) without having
to be bitten by the bug first...)

-- dick seymour
seymour@uwaphast.bitnet

p.s. the bugs were:
 1) bad code generated (BCG) for BLOCK DATA subroutines without
     any COMMON declarations
 2) compiler goes into an endless loop when a DO WHILE is
     determined to be endless at compile time is inside a DO loop
        which can be eliminated at compile time (due to no results
         being needed later in the routine)
 3) BCG involving array references in complicated DO-loops. a register
     was being mis-loaded.
 4) invalid diagnostics when a lowercase c with /d_lines (mine!)
 5) internal consistency failure if a routine had more than 64k declaration
        statements [2 comments: WOW! and why does DEC still use 16 bits?]
 6) BCG with some block moves and block initializations if a subroutine
        had two dummy array arguments which "happen to be" the same actual
        argument.
 7) BCG sometimes involving a character constant as the run-time FORMAT
     of an i/o statement contained in a DO-loop