On 2/26/10 8:37 AM, Daniel Russel wrote: > I think it probably makes sense to require a real boost install to > build IMP since > - Dina's application uses program-options > - I use file system a bit in kernel and it would make things much > simpler (and, perhaps, more reliable on windows), not to have to work > around its absence if it is not there > - it would simplify various scripts > - it is, in general, easier to get a full install of boost than a > headers only one (all normal platforms have some variant of rpm, mac > ports, the windows installer etc). We are telling mac people to install > mac ports as it is.
I agree with all points except the last. With Windows you have the problem of multiple C runtimes - last time I checked the installer was unusable because the DLLs installed were linked against a different runtime than the one we link against. We would have to rebuild the DLLs ourselves. This would be a hassle and I don't think it's wise to make such a change this close to release. My assumption is that you don't consider Solaris to be a "normal platform", but building Boost libraries on Solaris is non-trivial too. We can certainly look into it post-release though.
Ben