[sane-devel] Build system issues.

[Chris Bagwell]

Chris Bagwell wrote:
> Couldn't find any documentation related to what libtool behaviour 
> should be in tihs area.
>
> I just now installed sane 1.0.19 tarball into a private directory and 
> then installed current CVS on top of it.  The behaviour I got seems to 
> be the behaviour you want.  In the end, all symlinks were updated and 
> pointed to latest so.  I had the following in ${prefix}/lib and other 
> backends followed same behaviour.
>
> libsane.la
> libsane.so -> libsane.so.1.1.0
> libsane.so.1 -> libsane.so.1.1.0
> libsane.so.1.0.19
> libsane.so.1.1.0
>
> Let me know if you continue to see different behaviour and we can 
> continue to debug it.
>
> Also, this means I won't be porting over the Makefile symlink logic I 
> previously mentioned since it was broke and current libtool seems to 
> be OK on all platforms I tested.
>
Hmm, I did notice one interesting thing.  I went ahead and re-installed 
sane-1.0.19 back on top of the pre-existing private directory that had 
both 1.0.19 and 1.1.0-cvs installed.

sane-1.0.19's libtool did overwrite all symlinks (based on timestamps) 
but chose different values at different levels.

libsane.so -> libsane.so.1.0.19
libsane.so.1 -> libsane.so.1.1.0
libsane.so.1.0.19
libsane.so.1.1.0

Re-installing CVS points *everything* back to 1.1.0 so.  Its hard to 
tell if 1.0.19 behaviour was a bug in libtool or if there is some rule 
that *.so should point to last installed while *.so.x should always 
point to highest installed version.

Anyways, current CVS seems to be doing as expected since it has highest 
so version released yet.

Chris