On December 27, 2014 12:47:40 PM EST, Peter Dimov wrote:

Andrzej Krzemienski wrote:

...

2014-12-26 12:05 GMT+01:00 Antony Polukhin :

...

Are there any reasons not to change the existing implementation of Boost.Swap to Ion's version?

You may find it a silly argument, and it goes back to these dependency graph discussions. If you look at the dependency report, like this one http://www.pdimov.com/tmp/report-d5ca5be/module-levels.html It says Boost.Core is level 2 (so, fairly primitive) while Boost.Move is Level 6.

It's in fact not a silly argument at all. Boost.Core is explicitly not allowed to acquire new dependencies - this was the reason for creating it. We can in principle extract Boost.Swap into its own module though.

I'm sure Antony meant to change the implementation to match the ideas in Ion's code, not use the test implementation he showed. ___ Rob (Sent from my portable computation engine)

El 28/12/2014 a las 19:42, Pete Bartlett escribió:

...

I'm sure Antony meant to change the implementation to match the ideas in Ion's code, not use >the test implementation he showed.

But isn't the whole idea to depend on Boost.Move so that swap can benefit from move emulation? As Peter said, a new Boost.Swap could depend on Move, but Core should not.

The part used by adl_move_swap is about Level 2 (because it depends on BOOST_STATIC_ASSERT). Without BOOST_STATIC_ASSERT it would be Level 0. Boost.Move is Level 4 because some header depends on TypeTraits. Obviously the dependency report can't show this. Is the dependency report based on master or develop branch? I can't see any mention to this. It would be nice if the report could show this in the last "Generated on ...." line. Thanks Ion

On December 28, 2014 1:42:54 PM EST, Pete Bartlett wrote:

...

I'm sure Antony meant to change the implementation to match the ideas in Ion's code, not use the test implementation he showed.

But isn't the whole idea to depend on Boost.Move so that swap can benefit from move emulation? As Peter said, a new Boost.Swap could depend on Move, but Core should not.

It could be implemented based upon C++11 move semantics, when available. If there were a way to detect and use Boost.Move without requiring a reference to it, that would be ideal, of course. ___ Rob (Sent from my portable computation engine)

Andrzej Krzemienski wrote: ...

http://www.pdimov.com/tmp/report-d5ca5be/module-levels.html

I've generated a new report at http://www.pdimov.com/tmp/report-b1adfee/module-levels.html