Boost logo

Boost-Build :

From: Vladimir Prus (ghost_at_[hidden])
Date: 2006-09-24 03:38:41


On Wednesday 20 September 2006 13:53, John Maddock wrote:
> Vladimir Prus wrote:
> > Done in CVS. Can you try again? I've removed version both in windows
> > and linux. Is that right?
>
> Library names look right now (the same as bbv1 at least), but linking still
> fails for the same reason as Borland: see below.
>
> >> It seems that the lib's are being named correctly (but I haven't
> >> checked them all), but the link fails because the directory that
> >> contains the lib isn't in the lib search path. Adding a
> >> -Lpath-to-lib on the command line would fix that.
> >
> > What's in response file? I get absolute path to the library there,
> > but doing this on Linux I've no idea what you'd get on Windows.
>
> Yes the absolute name is correct and in the linker response file, however
> the library is also selected via auto-linking and that gets just the
> filename (no path). MS's linker appears to ignor any autolinked lib name
> if it's the same as a lib already specified on the command line. Intel's
> and Borland's don't: they insist on looking up the lib whatever. This is
> harmless, and it's a useful check that the autolink and bbv2 code are doing
> the same thing, but only works if *both* the full lib name is given *and*
> the directory containing the lib is specified with -Lpath.

The question is we we want to add -L options just when testing auto-linking,
or always? I don't like the idea of always adding it, since that would be a
bit pointless with V2 -- you have absolute path to the library already, so
the only thing autolinking can do is to link to the same library.

In other words, autolinking is very helpful when you link to installed Boost
using some existing build system of yours, and we need to test that it works,
but when using Boost.Build, autolinking is not necessary.

So, I think we need an extra feature that specifies if -L options derived from
full paths of all used libraries should be added, and use that feature when
testing auto-linking code.

Does this sound reasonable?

- Volodya

-- 
Vladimir Prus
http://vladimir_prus.blogspot.com
Boost.Build V2: http://boost.org/boost-build2

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