Re: [Boost-users] boost.container.vector::operator==() bug?

Subject: Re: [Boost-users] boost.container.vector::operator==() bug?
From: Ion Gaztañaga (igaztanaga_at_[hidden])
Date: 2014-11-10 02:52:13

On Mon, Nov 10, 2014 at 8:41 AM, niXman <i.nixman_at_[hidden]> wrote:

> niXman 2014-11-10 01:19:
>
>> Ion Gaztañaga 2014-11-10 00:44:
>>
>>> I don't think using std::equal will guarantees the use of memcmp. At
>>> least, the last time I checked some major implementations didn't use
>>> it and others only for char types. I've seen libstdc++ uses memcmp for
>>> non-char types, maybe that could be a good example to follow.
>>>
>>
>> Hmm..
>> std::equal()[1] uses __equal_aux()[2], which uses '__builtin_memcmp()'
>> for:
>> https://gcc.gnu.org/viewcvs/gcc/trunk/libstdc%2B%2B-v3/
>> include/bits/stl_algobase.h?view=markup#l832
>>
>>
>> [1] https://gcc.gnu.org/viewcvs/gcc/trunk/libstdc%2B%2B-v3/
>> include/bits/stl_algobase.h?view=markup#l1047
>> [2] https://gcc.gnu.org/viewcvs/gcc/trunk/libstdc%2B%2B-v3/
>> include/bits/stl_algobase.h?view=markup#l828
>>
>
> ping?
>

As said, listdc++ uses memcmp. MSVC uses it only for chars, libc++ does not
use memcmp. So Boost.Container will need to implement its own optimization
based on libcstd++ ideas: pointers and integers (asumming no padding or
trap representation) can be memcmp-ed on all platforms Boost runs. PODs,
due to float signed zero and Nan, can't be memcmp-ed.

A similar optimization could be done for lexicographical_compare, but only
for unsigned chars. Unsigned integers might work in Big Endian machines. In
any case, I consider this a very specific optimization that chould not be
crucial.

Ion

>
> --
> Regards, niXman
> ___________________________________________________
> Dual-target(32 & 64-bit) MinGW-W64 compilers for 32 and 64-bit Windows:
> http://sourceforge.net/projects/mingw-w64/
> ___________________________________________________
> Another online IDE: http://liveworkspace.org/
>