Re: [Boost-users] bug in boost::multi_index::sequenced splice (patch)

From: JOAQUIN LOPEZ MU?Z (joaquin_at_[hidden])
Date: 2006-09-30 06:06:58

----- Mensaje original -----
De: Hubert Schmid <kf1gmtvro69bnet5_at_[hidden]>
Fecha: Viernes, Septiembre 29, 2006 4:48 pm
Asunto: [Boost-users] bug in boost::multi_index::sequenced splice 
(patch)
Para: boost-users_at_[hidden]
CC: Hubert Schmid <kf1gmtvro69bnet5_at_[hidden]>

> Hi,
> 
> I guess there is a bug in boost/multi_index/sequenced_index.hpp. The
> following program fails to compile, because there is no matching
> operator== for class C. But this operator shouldn't be required for
> splice.
[...]
> The following program compiles, but both assertions fail at runtime:
[...]
> I have made both tests with the boost library in Debian/unstable
> (version 1.33.1-7). I have also compared the source file
> sequenced_index.hpp with the CVS-MAIN version (1.12). I guess the
> following patch should fix this bug:
> 
> --- sequenced_index.hpp.orig    2006-09-29 11:38:06.000000000 +0200
> +++ sequenced_index.hpp 2006-09-29 11:39:43.000000000 +0200
> @@ -355,7 +355,7 @@
>     BOOST_MULTI_INDEX_CHECK_DEREFERENCEABLE_ITERATOR(i);
>     BOOST_MULTI_INDEX_CHECK_IS_OWNER(i,x);
>     BOOST_MULTI_INDEX_SEQ_INDEX_CHECK_INVARIANT;
> -    if(x==*this){
> +    if(&x==this){
>       if(position!=i)relink(position.get_node(),i.get_node());
>     }
>     else{
> @@ -388,7 +388,7 @@
>     BOOST_MULTI_INDEX_CHECK_IS_OWNER(last,x);
>     BOOST_MULTI_INDEX_CHECK_VALID_RANGE(first,last);
>     BOOST_MULTI_INDEX_SEQ_INDEX_CHECK_INVARIANT;
> -    if(x==*this){
> +    if(&x==this){
>       BOOST_MULTI_INDEX_CHECK_OUTSIDE_RANGE(position,first,last);
>       if(position!=last)relink(
>         position.get_node(),first.get_node(),last.get_node());

Hello Hubert,

This is indeed a bug, and a gross one I must embarrassingly admit :)
Also, the proposed patch is the way to go. Thank you very much for
spotting this, I'll be committing the fixes next Monday (during
weekends I'm CVS-disabled.) FYI, the same probem also shows at
other places like random-access indices and some other ops like
merge, yet the reference docs use the correct expression &x==this
everywhere, it's so disturbing, I wonder how this could go unnoticed
for so long.

Best regards,

Joaquín M López Muñoz
Telefónica, Investigación y Desarrollo