Boost logo

Ublas :

Subject: Re: [ublas] g++ 4.4.6 and boost ublas 1_48 compile problem
From: Thomas Klimpel (Thomas.Klimpel_at_[hidden])
Date: 2012-02-22 11:21:09

Umut Tabak wrote:
> Thomas Klimpel wrote:
> > Well, probably yes. I guess "std::inplace_merge" from g++ 4.6.1
> > plays dirty tricks, which it is allowed to do according to the
> > standard. A possible solution would be to use a different
> > implementation of "inplace_merge".
> Or continue using boost 1_47, I guess from ublas perspective the
> results would be the same.

Oh, I don't think that the version of boost has anything to do with this. The problem should be the same in boost_1_49 as in boost_1_40. The problem is in the libstd++ that comes with gcc > 4.4

> I use MTL4 generally for my operations however to
> assemble some FE system matrices generalized_vector_of_vector was the
> fastest option and I build my interface classes with this container so
> I would like to stay with that for the moment. One more question, any
> further and faster alternatives to this path?

To be honest, I don't fully understand what you are asking here.

> You mean some different implementations such as sgi or so, well up to
> now I have always depend on standard so I will see what I can do.

Well, I meant that boost::ublas should use its own implementation of "inplace_merge" (copied for example from sgi, if license permits). As I wrote in that message, the main problem would be to sort out potential license issues. But I have no idea what the actual status is. The last thing I remember is that David replied that he wanted to take a look at it.