Re: Is the Writing on the Wall for Ada?

["Robert I. Eachus" <rieachus@attbi.com>]

Wes Groleau wrote:

 > Interesting how you were not able to persuade any of us ignoramuses
 > that temporaries are always needed without it nor that they are never
 > needed with it.

Let me tell you a story with names removed because I don't want to seem
to be dissing any company or compiler--because in this case the reality 
was just the opposite.

A woman, who was a fairly well respected compiler expert called me for 
advice.  She was implementing an Ada compiler for a vector machine, and 
was concerned that as she read the rules, Ada required copying the 
arguments in cases where Fortran 90 did not.

I gave her a few lines of Fortran over the phone.  She said, yes, that 
is exactly an example of what I am talking about.  Okay, I said, go 
compile it, run it, and look at the results...

After lunch she called me back.  "As you must have expected, our Fortran 
compiler produced garbage."

"Yep, and I can give you chapter and verse from the Fortran standard 
where it says that case is not erroneous."

"No need, I can do that myself.  Do I need to fix the Fortran compiler 
to make copies of vectors in all the cases that Ada requires them?"

"Nope, if you look at the wording carefully, all that both languages 
require is that you get the right answer.  It is just a little more 
obvious in Ada that sometimes that requires an 'extra' copy."

She fixed her Fortran compiler, and also validated her Ada compiler. 
But let me explain why she had problems.  On a vector machine, the 
hardware can step through an array with a "stride" set to more than one. 
  In other words extract a vector from a 2 dimensional, or even 
n-dimensional array.  But in both Ada and Fortran you can store the 
result of that computation in the same array.  Now imagine a case where 
the vector machine grabs more than one value--this one took 8 byte 
chuncks so a single precison array would do.  But for the code I 
suggested the vector engine wrote back the modified eight-byte value 
with the second half modified, just after writing the first four bytes. 
  The hardware couldn't guarantee that the two writes would occur in any 
particular order, and in practice they didn't.

Both Ada and Fortran were willing to allow her not to use temporaries, 
as long as the compiler could verify that this particular gotcha didn't 
occur.

-- 
                                          Robert I. Eachus

"As far as I'm concerned, war always means failure." -- Jacques Chirac,
President of France
"As far as France is concerned, you're right." -- Rush Limbaugh