Re: [PATCH][RFC] Add a subset of -Warray-bounds warnings to C/C++ front ends

[Simon Baldwin <simonb at google dot com>]

Richard Guenther wrote:
> On Fri, Apr 4, 2008 at 1:07 AM, Simon Baldwin <simonb@google.com> wrote:
>   
> > Attached below is a modest patch to provide a subset of the -Warray-bounds
> >  warnings from tree-vrp.c in the C and C++ front ends.  This permits the
> >  compiler to warn about egregious array bounds violations in unoptimized
> >  compilations or compilations that may use -fno-tree-vrp.  At present, array
> >  bounds checking is only done on optimized compilations.
> >
> >  A side effect of copying these warnings up into the language frontends is
> >  that warnings are now printed even if the array access is in dead or
> >  inaccessible code.
> >
> >  The current array bounds tests are modified to account for this new checking,
> >  and additionally there are two new tests for warnings from -O0 compilations,
> >  one for C and one for C++.
> >
> >  Bootstrapped, and regression tested on i686 Linux for gcc and g++.
> >
> >  Thoughts?  Okay for trunk?
> >     
>
> I think it is a good thing to move these to the frontends.  Did you test
> the warning on some real-world C and C++ code?  I expect some
> "false" positives from deliberately partial dead code coming from
> template instantiation or constant propagation from function arguments.

I tried it on a codebase consisting of more than 3,000 source modules, 
almost all C++.  Encouragingly, it produced no false positives, and only 
one true positive, the latter from a "plant" I'd deliberately placed 
into the code ahead of time just to be sure I'd set up the compilation 
correctly.