Re: state of libstdcxx_so_7-branch

[Paolo Carlini <pcarlini at suse dot de>]

Chris Jefferson wrote:

> The rvalue changes should probably (in my opinion at least) go
> straight into mainline at the same time as the compiler support is
> added. I'm convinced that they won't break ABI at all, although I
> think the most optimal string change might require an ABI break (of
> course long term rvalue references encourage the idea of a non-COW
> string).

Yes, at some point we even wanted to add to mainline the *simulated* 
rvalue references stuff, appeared safe from the binary compatibility 
point of view. Then we decided to wait a bit for the standardization 
process to proceed a bit more.

I think it would be safe now to proceed as you are suggesting, the only 
minor issue being the 'forward' un-uglyfied name being injected in 
namespace std also in C++03 mode. I think we can live with that...

Probably you could start experimenting with a current, mainline compiler 
together with the libstdcxx_so_7-branch library, I don't expect major 
issues, besides the __is_pod name conflict and a bit of hacking 
c++config for C++0x mode. That way you can work as soon as possible on 
replacing simulated rvalues with the real ones. Note, for mainline, 
until we break the ABI for real, I say better not fiddling at all with 
basic_string, ext/vstring.h is ok.

Paolo.