Re: [Boost-build] --layout=system for *nix vendors

Subject: Re: [Boost-build] --layout=system for *nix vendors
From: Vladimir Prus (ghost_at_[hidden])
Date: 2008-11-03 02:35:13

On Monday 03 November 2008 10:20:04 Steve M. Robbins wrote:
> On Mon, Nov 03, 2008 at 09:33:46AM +0300, Vladimir Prus wrote:
> > On Monday 03 November 2008 07:00:29 Steve M. Robbins wrote:
> 
> > > How about the following change, which retains <base> <threading>
> > > <runtime> for --layout=system but also appends $(BOOST_VERSION) when
> > > appropriate?
> > 
> > So, with --layout=system we will get names like:
> > 
> > 	libboost_filesystem-mt.so.1.37.0
> > 
> > ? I believe this is step in the right direction. Philipp, what do you think?
> > 
> > Personally, I think believe we should drop threading and always build with MT mode,
> > as that's what other libraries seem to be.
> 
> I agree with the idea that MT mode should be the default variant.  It
> is likely the only variant shipped by vendors.
> 
> My concern is with history: the MT variant has been decorated
> heretofore with -mt, while the ST variant has not been decorated.  I'm
> afraid that all existing projects contain that assumption in their
> build environment, whether it is based on bjam, autoconf, or CMake.
> Changing the rules now would be very confusing.
> 
> Any ideas for dealing with this?

Well, no. We either live with the pain of -mt, or introduce migration pain.
Qt dropped -mt with transition to 4.*, so they've hidden the pain in a bigger
one. I'm not sure C++ Boost will have an equally fortunate moment. I can ask
on boost-user to get an feeling what users will prefer. Should I?

- Volodya