On Sun, 12 Apr 2009, Andreas Schuldei wrote:

> trying to package the current libcurl i discover that the soname that should
> be specified in configure.ac in the AC_INIT macro is set to [-], which
> corresponds nicely with the resulting soname of 0.0.0 that i get when
> building the lib.
>
> i hear that the soname should be in AC_INIT for "make dist" to work
> correctly later on, even though it could be overwritten/inserted in make the
> Makefile.am, too.

That's not quite true. On several points:

AC_INIT() defines a _version_, not soname. So 7.19.4 and not 4.1.0 or similar.
But 'make dist' works fine without it set by force by AC_INIT() and instead we
have the configure script extract the number from the include file. Can you be
more explicit about in what particular situation this breaks? We've done it
like this for MANY years so it's a bit surprising something broke now.

> where is the soname specified in the current source? (i have been looking at
> the tarball and the cvs version to figure it out, but cant find it.)

The soname is set with the -version-info flag to libtool, and that number is
currently set in lib/Makefile.am only.

> could we have the soname in AC_INIT in the future or is there a reason
> against that?

soname is plain wrong to have there. We _could_ make the released configure.ac
and associated configure script get the "correct" version number in there by
generating a specific release version of the configure.ac with maketgz and
generate the release configure from that but I've refrained from that so far
since it'd require us to always rerun autoconf within 'maketgz'. (And the sole
reason I don't like to rerun it is because it makes the operation a lot
slower.)

-- 
  / daniel.haxx.se