From mboxrd@z Thu Jan  1 00:00:00 1970
X-Spam-Checker-Version: SpamAssassin 3.4.4 (2020-01-24) on polar.synack.me
X-Spam-Level: 
X-Spam-Status: No, score=0.4 required=5.0 tests=BAYES_00,INVALID_DATE,
	UNRESOLVED_TEMPLATE autolearn=no autolearn_force=no version=3.4.4
Path: utzoo!utgpu!water!watmath!clyde!cbosgd!ucbvax!STAR.STANFORD.EDU!ALLEN_S%ASD.SPAN
From: ALLEN_S%ASD.SPAN@STAR.STANFORD.EDU (Stanley Roger Allen, AdaDude)
Newsgroups: comp.lang.ada
Subject: Optimizing compilers
Message-ID: <8801220047.AA12713@ajpo.sei.cmu.edu>
Date: 21 Jan 88 22:25:00 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The ARPA Internet
List-Id: <comp.lang.ada>


Naom Tene states:

>I am NOT saying that the optimized code is or should be legal I am
>saying that IF you want faster execution you must PAY for it. VAX
>ADA does NOT use the optimizer by default (unless the system manager
>interferes with it) if it had it would have been a bug.

Actually, the VAX Ada compiler has the OPTIMIZE switch ON by default
(read the documentation).  Since a compiler is (supposedly) validated
with the default settings, it is natural to expect that this is how
the DEC Ada compiler passed the suite (can anyone out there speak on
the issue of compiler switches used at validation time?). 

As touching the larger issue, however, it must be said that optimized 
code STILL SHOULD BE LEGAL (of course, with a view toward the text
in LRM 11.6).  The Ada language has many optimizable features which 
are exploitable within the boundries of the standard.  Certainly the 
designers of Ada were aware that optimization was to be performed on 
Ada code, and intended that the standard would define the EFFECTS that 
a program would have, whatever optimizations were applied to it.  
To "have to pay" for faster execution at the expense of legality is
dangerous and unnecessary.

Therefore, the legality of a program is to be considered independent of 
optimization.  Just because an implementation of the language has bugs 
does not mean that Ada allows the kind of optimization that produces 
the bugs.  If the implementation does so, there is a bug in the 
implementation.

One more thing concerning bugs: I read recently a ComputerWorld 
article on cross-compilers in which someone (from the defense 
community) said "There is an Ada validation suite which guarantees 
that an Ada compiler has no bugs".  This is a gross misundertanding of 
the nature of the validation process.  The validation suite has its 
limits, and it DOES NOT indicate that a compiler is:

	 * good
	 * bug free
	 * useful for a particular application domain
	 * anything but a conformance to ANSI/MIL-STD 1815A-1983

Many Ada implementations are excellent, solid products with many 
man-years of effort behind them.  The fact that some bugs could exist 
in such large and complex products is not at all suprising.

As for those products which pass validation but are of poor quality,
the only thing that can be said is "to the victor go the spoils" -- 
meaning of course that the bad compilers will be driven out of the 
market.

			Stanley Allen
			allen_s%asd.span@jpl-vlsi.arpa