|
Boost : |
From: John Maddock (john_at_[hidden])
Date: 2004-07-28 05:37:00
> > Yes you remember correctly. The reasons are rather simple: if one
> > directly refers to another library in the test, with <dll>* or <lib>*,
> > then the library is linked to directly and hence there is no need for
> > auto-linking. And auto-linking in that case would be redundant, and may
> > even cause errors and warnings from the linker. But the reason one
> > *must* refer to other libraries directly for tests is that auto-linking
> > is not available on all platforms. Hence since they are mutually
> > exclusive and tests need to be cross-platform auto-linking needs to be
> > disabled for regression testing.
They're not mutually exclusive as long as the toolsets are consistently
named so that the auto-link code can always generate the same name as the
toolset for the library. Unfortunately the toolsets (mainly the Intel ones)
aren't consistently named - we have the main toolset with no version number,
for which the autolink code does the right thing, and a bunch of other
versioned toolsets for which it does not :-(
John.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk