Re: Patch for various warnings in gcc

[Dale Johannesen <dalej at apple dot com>]

On Thursday, August 8, 2002, at 10:32 AM, Zack Weinberg wrote:

> On Thu, Aug 08, 2002 at 12:05:05PM +0200, Michael Matz wrote:
> > > > ra-build.c:1818: warning: `web' might be used uninitialized in this 
> > > > function
> > This warning is wrong, as web in fact is initialized in all possible
> > paths.  There is no need for a fix, except maybe to silence the warning.
> > Wasn't there a gcc extension ala "int a = a;" to silence it?
> This topic has come up in the past; the consensus was that when GCC
> couldn't prove a variable initialized, it was best to initialize it
> at declaration to a value that would cause an immediate failure if
> used.  (NULL, for pointers.)  The cost of doing a single
> load-constant-zero instruction is unlikely to be measurable, and using
> a deterministic initializer is simply good defensive programming.  We
> have unconditional 'if (impossible) abort()' checks all over the
> compiler - those are more expensive, surely, and no one suggests
> removing them.
Well, there's a difference between checks that affect only the compiler
itself and generating extra code in user programs.  Still, I'd expect
the optimizer to get rid of the dead store in most cases where it mattered,
so I don't have a problem with this.

> Personally I agree with Dale that 'int a = a;' is too obscure a way of
> silencing this warning.  If we must have a way to disable it without
> generating extra code, let us create __attribute__ ((initialized))
> which states explicitly what is going on.
I like this.