|
Boost-Build : |
From: Steve M. Robbins (steve_at_[hidden])
Date: 2008-04-15 00:24:51
On Mon, Apr 14, 2008 at 04:23:19PM +0200, Roland Schwarz wrote:
> However one question about debian policy:
>
> http://www.netfort.gr.jp/~dancer/column/libpkg-guide/libpkg-guide.html#sonameapiabi
Err, that's actually a packaging guide; a HOWTO manual, if you like.
Debian Policy may be found at http://www.debian.org/doc/debian-policy/
> "In most cases, if a package version matches the SONAME, it is a sign
> that there is a problem with the versioning scheme."
>
> So, if following your suggestion, would this indicate we introduce a
> "problem with the versioning scheme"?
This comment is directed at the mainstream libraries that maintain ABI
compatibility across patch versions. Since the SONAME is effectively
the ABI identifier, you only expect a SONAME change when the major
version changes. A typical scheme for such library, say foo version
x.y.z, is to use a SONAME of "libfoo.so.x".
A library whose SONAME matches its package version is saying "we
cannot promise ABI compatibility between any two versions".
So: no, following my suggestion does not introduce a problem; that
"problem" has been there all along. Namely, that Boost does
not maintain ABI compatibility.
> Shouldn't boost libs rather:
>
> libboost_program_options.1.34.1.so
>
> and have a unversioned link from
>
> libboost_program_options.so
>
> Could you please comment on this?
My only comment is: the established convention on linux (and I believe
some other unix systems) is to use "libfoo.so.x.y.z" rather than
"libfoo.x.y.z.so". I don't know why one form was chosen over the
other.
-Steve
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