lclint-interest message 158

From evans@cs.virginia.edu Mon Sep 22 15:37:27 1997
Date: Mon, 22 Sep 97 15:36:00 -0400
From: evans@cs.virginia.edu (David Evans)
To: kir@iitb.fhg.de
Cc: lclint-interest@larch.lcs.mit.edu
In-Reply-To: Harald Kirsch's message of Mon, 22 Sep 1997 16:43:09 +0200 <199709221443.QAA06683@s424.iitb.fhg.de>
Subject: How to annotate


Harald writes: 

> I understand that the case is for lclint not as trivial as it may appear
> to us. It has to prove that the value of `i' does not change between the
> two if-statements.
> 
> However, I thought lclint does some data-flow analysis and I cannot
> decide, if it should know that `f' is defined at the place where it
> gives the error.

LCLint doesn't do deep enough analysis to determine that the second
predicate is true only if the first predicate was also true and f is
therefore defined.  

Richard Brooksby's point that the code should be rewritten not annotated
is a good one.  (Of course, I understand that this is just an example to
make things as simple as possible.)  In general, I would claim that many
of the times lclint reports spurious errors because it cannot determine
that the code is correct are the same types of code that it is hard for
a human to determine it is correct (and know what that correctness
depends on so it will break when maintainence changes are made).

If you want to suppress the warning, you could use the /*@-usedef@*/
syntactic comment:

	/*@-usedef@*/ float g = f; /*@=usedef@*/

--- Dave