Re: [patch] proposed fix for libstdc++/59829

[Jonathan Wakely <jwakely at redhat dot com>]

On 27/01/14 23:37 +0000, Jonathan Wakely wrote:
> On 27 January 2014 20:35, Jonathan Wakely wrote:
> > On 27 January 2014 20:12, Marc Glisse wrote:
> > > On Mon, 27 Jan 2014, Jonathan Wakely wrote:
> > >
> > > > This is the best I've come up with to avoid dereferencing an invalid
> > > > pointer when calling vector::data() on an empty vector.
> > > >
> > > > For C++03 we reurn the vector's pointer type, so can just return the
> > > > internal pointer, but for C++11 we need to convert that to a raw
> > > > pointer, which we do by dereferencing, so we must check if it's valid
> > > > first.
> > >
> > >
> > > For comparison, libc++ has 2 paths. If pointer really is a pointer, it just
> > > returns it, no need to pay for a comparison in that case. And otherwise, it
> > > calls _M_start.operator-> and crosses its fingers. There is a helper
> > > function doing that used throughout the library.
> >
> > Ah yes, I remember Howard posting a get_raw_pointer() function to the
> > reflector that used operator->() on user-defined types ... I don't
> > really like calling that on a potentially invalid pointer though. The
> > user-defined pointer type in my new testcase could just as easily
> > throw if operator-> is called on an invalid pointer.  As Paolo also
> > mentioned avoiding the branch for built-in pointers I'll do that.

Here's what I'm committing, the testcase is simplified by reusing the
new PointerBase type I added to testsuite_allocator.h

Tested x86_64-linux, committed to trunk.

Attachment: patch-2.txt
Description: Text document