[csw-maintainers] Shared library placement proposal

[Maciej Bliziński]

2011/2/9 Philip Brown <phil at bolthole.com>:
> 2011/2/8 Maciej Bliziński <maciej at opencsw.org>:
>>
>> The text doesn't use the "a library or a symlink to it" expression
>> each time, because if we started to do this in every case where
>> symlinks could be used, the policy would be rather unreadable.
>>
>
>
>
>
> Perhaps we need to take a very different wording style, rather than
> attempting to just make small changes.
> Something like,
> "The library needs to have a presence in /opt/csw/lib"
> or "the library needs to be linkable with -L/opt/csw/lib."

This kind of wording sounds good to me.  In the specific case of -L, I
think it should be rather "Findable at runtime when RPATH is set to
/opt/csw/lib or /opt/csw/lib/$ISALIST".  The reason is that there's a
counterexample: the way we link against libnet, is
-L/opt/csw/lib/libnet-new.  The libnet.so.1 file is located in
/opt/csw/lib (it's a symlink to libnet.so.1.6.0, but it's an
implementation detail).  The same approach can be used with
berkeleydb.

The wording "The library needs to have a presence in /opt/csw/lib"
sounds the best to me.

> Although personally, i think if we just settle the issue of symlinks
> now vs later, then the writing will be simpler.

I have suggested writing a separate proposal for the issue of
symlinks.  Symlinks are a powerful tool, and practically any file can
be replaced with a symlink to another file located somewhere else.  If
we want to regulate how symlinks are used, we should have at least a
rough policy when the use of symlinks is appropriate.