Hello, On Tue, Nov 09, 2010 at 09:37:43PM -0200, Fernando Lemos wrote:
> I don't plan to do anything about this for the time being. The reason > for the FTBFS (as reported by others[2]) is that boost::filesystem > uses boost::system in its internal headers (in inline methods invoked > by the boost::filesystem::path constructor, for instance). A > boost::filesystem user is not supposed to know in advance that > boost::filesystem requires linkage against boost::system as this is an > implementation detail that could change in the future. > > btag does not use boost::system at all, and therefore it would not > make any sense to think this is a problem btag must solve. I see a few > possibilities for this being solved the right way: > > 1. It might be possible to change booost::filesystem to make those > methods that rely on boost::system (I think it's just > boost::system::error_code) not inline anymore. This would work fine, > but it might not be viable as C++ templates must be inlined. I don't > see this happening, good luck convincing the Boost team that the way > gold works is the right way. Speaking only for myself: please do explain why gold is an improvement. It seems to be a regression to me and my suggestion would be to avoid it and use the current linker. I have no idea how one would properly handle dependencies in inlined code, such as C++ templates. Cheers, -Steve
signature.asc
Description: Digital signature
