Re: Turning warning "returning reference to temporary" into error?

[Fergus Henderson <fjh at cs dot mu dot OZ dot AU>]

On 11-May-2003, Zack Weinberg <zack@codesourcery.com> wrote:
> Gabriel Dos Reis <gdr@integrable-solutions.net> writes:
> 
> > A compiler writer cannot imagine every possible usage.  Which is why
> > somethings are warnings.  In this case, if the return value is ignored
> > then the program is valid.  GCC cannot reject it just because
> > -pedantic  is specified.
> 
> In other places where we can prove that execution of a certain code
> path will cause undefined behavior, but we can't prove that that path
> will be taken at runtime, we issue a warning and generate a trap
> instruction.  An analogous tactic for this scenario would be if we
> generated code to return a distinguished pointer value that's
> guaranteed to cause a segmentation fault when dereferenced.  (This
> should _not_ be 0, as that may mask the bug.)

I don't think that would be a good idea.

Although the behaviour is undefined, there are some important programs
that rely on the assumption that the address of a local is an address
on the stack.  In particular, the Boehm et al. conservative garbage
collector does this.  If I recall correctly, some versions of
LinuxThreads also did this (to implement pthread_self()),
although I don't know off-hand if they still use that technique.

If GCC was changed in a way that stopped this technique from working,
conservative garbage collectors would have to resort to techniques
which are *less* portable, such as inline assembler.  I don't think
that would be an improvement, and I definitely don't think it would
be worth the transition cost.

-- 
Fergus Henderson <fjh@cs.mu.oz.au>  |  "I have always known that the pursuit
The University of Melbourne         |  of excellence is a lethal habit"
WWW: <http://www.cs.mu.oz.au/~fjh>  |     -- the last words of T. S. Garp.