From: Johan Nilsson (r.johan.nilsson_at_[hidden])
Date: 2006-11-14 03:46:26
David Abrahams wrote:
> "Johan Nilsson" <r.johan.nilsson_at_[hidden]> writes:
>> Roland Schwarz wrote:
>>> Johan Nilsson wrote:
>>>> +1 for generating an error.
>>> The author has the intent to
>>> <threading>single <runtime-link>shared
>>> She specifies this in a platform independent manner.
>>> Now on platform A this might indeed work.
>>> But on platform B (e.g. windows) it won't.
>> Isn't this more of a compiler-specific issue than O/S specific? I
>> get your point though.
>>> Would you expect, only platform B generating an
>>> error, or fall-back B to a safe emulation?
>> I'd prefer only platform B generating an error. If I explicitly
>> request a shared runtime library I'd like to either get it, or fail.
> I think we need to distinguish "requires a shared runtime" from
> "I prefer a shared runtime if one is available."
Good point. Would it be possible to make that distinction stand out better
in some way?
> I was under the impression that we made that distinguish by putting
> <runtime>shared either in requirements or in a default build request.
So, in the case of requirements, either via command-line options or through
the Jamfiles, the above example should fail.
In the case of a default build request, a graceful fallback to either a
"safe emulation" or simply not to build that variant would be ok.
I've read the earlier discussion in the "Building Boost and impossible
features" thread and I think this is similar to what the conclusions were
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