From: Vladimir Prus (ghost_at_[hidden])
Date: 2002-05-03 08:05:56
David Abrahams wrote:
> > I have implemented it by simply creating new feature attribute,
> > 'link-incompatible'. It says targets with different value of the
> > feature cannot be linked together. Any objections?
> Well, it doesn't capture all of the relationships we need. For example,
> <shared-linkable>true is link-compatible with everything, but
> <shared-linkable>false is not. In other words, it's not always a
> reflexive relationship.
> Hmm, maybe that is just an artifact of some v1 mis-design.
> Let me try to put it more clearly:
> The "normal" value of <shared-linkable> should be false, since we don't
> want to impost -fPIC where it's unneeded. A target built with
> <shared-linkable>true cannot be linked with any dependencies having
> <shared-linkable>false. However, a target built with
> <shared-linkable>false can be linked with a dependency which is
> <shared-linkable>true, no problem.
Yes, I understand that. I mean: should be devise a general scheme right now,
or just go ahead with simple 'link-incompatible' features, to refine
link-incompatibility later? I think that just about everything depends on
'apply-requirements' and that rule is only half usefull without
link-incompatibility in some form.
Boost-Build list run by bdawes at acm.org, david.abrahams at rcn.com, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk