r199452 - Issue a warning if a throwing operator new or operator new[] returns a null

[Richard Smith]

On Mon, Feb 10, 2014 at 1:02 PM, Joerg Sonnenberger <joerg at britannica.bec.de
> wrote:

> On Fri, Jan 17, 2014 at 02:09:33AM -0000, Richard Smith wrote:
> > Author: rsmith
> > Date: Thu Jan 16 20:09:33 2014
> > New Revision: 199452
> >
> > URL: http://llvm.org/viewvc/llvm-project?rev=199452&view=rev
> > Log:
> > Issue a warning if a throwing operator new or operator new[] returns a
> null
> > pointer, since this invokes undefined behavior. Based on a patch by
> Artyom
> > Skrobov! Handling of dependent exception specifications and some
> additional
> > testcases by me.
>
> OpenJDK contains code like
>
>         void *foo::operator new(size_t, bar &, int, TRAPS) {
>           if (error_condition)
>             return 0;
>           return some_value;
>         }
>
> Should this warning really apply here?
>

Yes, this is exactly the code that we wanted this warning to catch. If
'error_condition' is true, this code has undefined behavior.

Another issue is that it doesn't check whether the return statement is
> actually reachable, e.g. a __builtin_unreachable() just before it does
> not silence the warning. I think it should.


Sure, if you want to go ahead and change the Diag to DiagRuntimeBehavior in
CheckReturnValExpr, that seems fine to me.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20140212/4711dfbd/attachment.html>