Re: [Boost-testing] vc6/7 for 1.34?

From: AlisdairM (alisdair.meredith_at_[hidden])
Date: 2006-02-12 14:26:31

Jeff Garland wrote:

> So the bottom line is I don't see any problem keeping the current
> level of support for 5_6_4.  Presumably when the config problems are
> cleared up 5_8_1 support should be similar. 

I believe that will be correct, once configuration issues are cleared
up.  I don't think there are any significant fixes in the compiler for
date time, although the change to Dinkumware STL may have an effect?

My main concern is not to suffer regressions, rather than get all the
latest features.  Obviously I would like them <g> but I am quite ready
to accept a library author, working on a voluntary basis, can assume a
full working C++ compiler.

Dropping support for VC/GCC2.95 is interesting though.
As you say, Spirit has already done this, but maintains a branch of the
previous version that remains compatible with the older compilers.  If
you do embark on a rewrite, it might be nice if the old version
remained available for those that wanted to support it.

An alternative might be to add a 'legacy' flag to the date time config,
detect 'legacy' compilers in the current headers, and redirect to an
archive of the existing implementation - a bit like MPL redirecting to
pre-processed headers.  You could then hand maintenance of the legacy
system over to their respective communities, and move on in the
mainline without a concern.

I would even accept Borland going down this 'legacy' route if it was
maintained in the mainline ;?)  so that I only needed one boost
distribution.  The problem with Spirit is applying the 1.6 branch on
top of Boost(latest) is a bit of a pain.

-- 
AlisdairM