[PATCH][libstdc++-v3 parallel mode] Avoid taking address of dereferenced random access iterator

[Johannes Singler]

On 03/10/2011 11:37 AM, Jonathan Wakely wrote:
> On 10 March 2011 09:47, Johannes Singler wrote:
>> The attached patch patch solves a conformance problem of the parallel mode
>> helper routine multiseq_partition.  I have added a test case for that.
>>   multiseq_selection has similar problems, but is unused, so I plan to remove
>> that completely (which might ask for renaming of the file and the test).
>
> Please update the copyright date in the changed file as well.

Done.

>> Should I use unique_ptr (or alloca, or something similar) here for a better
>> exception safety (this routine is not parallel itself)?
>
> unique_ptr is C++0x only, auto_ptr would work.  But I see other heap
> allocation in that function are already unguarded so there doesn't
> seem to be much point guarding one and not the others.  How about
> defining a local RAII type (and combining the three allocations into
> one) e.g.
>
>        struct _Guard
>        {
>            _DifferenceType* _M_ns;
>
>            ~_Guard() { delete[] _M_ns; }
>        } __guard = { };
>
>        __guard._M_ns = new _DifferenceType[__m*3];
>
>        _DifferenceType* __ns = __guard._M_ns;
>        _DifferenceType* __a = __guard._M_ns + __m;
>        _DifferenceType* __b = __guard._M_ns + 2*__m;
>        _DifferenceType __l;
>
> That ensures the _Guard destructor will clean up on exiting the
> function, so you can remove the delete statements.

Well, isn't it a bit ugly to define such a guard newly every time?
In other places, parallel mode uses std::vector, but I guess this is 
actually also discouraged for internal use, since it adds the <vector> 
dependency.

Johannes