Re: Life in Verdix Hell

[elroy.jpl.nasa.gov!swrinde!zaphod.mps.ohio-state.edu!wupost!darwin.sura.n et!uvaarpa!software.org!blakemor@ames.arc.nasa.gov (Alex Blakemore)]

In article <EACHUS.91Dec9005333@Dr_No.mitre.org> eachus@Dr_No.mitre.org (Robert
 I. Eachus) writes:
>      I think that this posting is very inappropriate.  If it stated
> which versions of each compiler (and on which hardware and which OS
> versions).  Then it would be possible to determine if this had any
> relevance to current experience.

You are correct that he should have posted the version numbers, but
is it inappropriate to post negative experiences with a widely used
and expensive commercial product ?  Especially if backed up by detail ?
Or should only rosy scenarios be posted to this forum ?

There were some aspects of his posting that were unflattering,
but the problems that he described were real and are not that different
than the problems that were pervasive five years ago.

On the Sun (SPARC 4), the compiler version was :
    Verdix Ada Compiler, Copyright 1984-9, 1990
    Sun-4 SunOS Release 4.0 and 4.1, Version 6.0
    Thu Jun 28 11:45:27 PDT 1990, 6.0.3(c)
    SunOS 4.1.1

On the Apollo, the compiler version was :
    (c) Copyright 1987, 1988, 1989, 1990 Hewlett-Packard Company.
    Incorporates copyrighted program material from Verdix Corporation
    (c) Copyright 1984, 1985, 1986, 1987, 1988.
    Domain/Ada Version 3.0.2 - March 23, 1990 
    (Apollo changed Verdix's number for some reason - I believe Apollo 3.0 = Ve
rdix 6.0)
    Domain/OS 10.3.5 (bsd4.3)

They tried a later Apollo compiler version which had different errors that were
 
*more* difficult to work around and had to move back to the older version.  

>      The Verdix compiler is not perfect, neither is the DEC Ada
> compiler. But both are much better now than the old versions
> apparently being described.  From the description, I can guess that
> the Verdix versions discussed are probably 5.41 and an early 5.5
> release.  Things have changed a lot since then.  

You are jumping to the conclusion that this project was
using obsolete compilers, and that things are better now.
The compiler versions on both Sun and Apollo were both post 6.0.
The DEC compiler was a recent version - 2.2, but version 1.0
would have been sufficient in most respects.

The people at Apollo support were very helpful,
and even sent a beta version to attempt to fix some of the problems.

The crux of the matter is

  a.  For the past several years, the quality of Unix Ada compilers
    has been spotty. There were bright spots, but for large projects
    large amounts of time would certainly be spent fighting the tools.
    Compiler crashes and code generation bugs were common.
    Contrast this to DEC and Rational which somehow have had high quality
    tools for almost five years.

  b.  Now over 6 years after DEC released their Ada compiler, no Unix
     compiler that I know of can come close.  Using recent versions of
     a common Unix compiler, these people had to abandon a basic
     make utility, work around crashes and disassemble compiled code
     frequently.  Libraries were easily corrupted forcing long recompilations.
     It is hard to justify the long term benefits of Ada in this envioronment.

  c. Things dont seem alot better than several years ago in this case.
     Is this experience unique ?  Are other users experiencing similar
     difficulties ? Are other vendors significantly better - able
     to compare with DEC for example ?  Or should we all just learn C++?

> today's compilers are a lot better than those of a few years ago.

Judging from the mail John has received, this experience was not
unique to this project.  I would love to see counter examples, but
I can understand why people who first try Ada in a Unix environment
often turn into advocates AGAINST the language.

P.S.  Reproducing compiler bugs to ship back to the vendor is expensive
and time consuming. Yet how can the compilers get better if their 
customers dont report problems with examples ?
I think compiler vendors should compensate their customers for this effort, 
perhaps discount the maintenance cost 10-20% for every reproducible error.
It still wouldnt cover the time lost due to the problem,
but it might make it easier to tell your boss that you are spending 
a day paring down an example to report the bug.
(You are in effect working for the compiler vendor at that point.)
Does anybody do anything like this ?  Everyone would benefit.

Dislaimer: My experience with DEC and Verdix is recent, but with
Telesoft and Alsys is over 2 years old.
-- 
---------------------------------------------------------------------
Alex Blakemore           blakemore@software.org        (703) 742-7125
Software Productivity Consortium  2214 Rock Hill Rd, Herndon VA 22070