FC6: library global symbol abuse

[Tom Lane]

Joe Orton <jorton at redhat.com> writes:
> On Thu, Nov 16, 2006 at 11:58:50PM -0500, Tom Lane wrote:
>> Uh ... why are these being compared?

> Just because they're both there.  It is intentional that two different 
> versions of this library are being shipped (and in one package)?

[ squints... ]  Good question --- it's intentional on upstream's part,
but the .1 .so is for Postgres 7.1 which we haven't supported for a long
time.  Probably could lose it.  Actually I've been thinking of stripping
both the pgsql and mysql drivers from unixODBC, because there are more
up-to-date versions shipped as separate packages.

> Also, regarding:
> Clashes for /usr/lib/libodbcminiS.so.1.0.0:

> is it correct that all these libraries must only ever be used indirectly 
> via libodbc (which will dlopen them) - and no app may link directly 
> against them?

Yeah; in fact they all *have* to define conflicting symbols because
that's the ABI they're supposed to present to libodbc.

> If so they can go on the whitelist, but they should not 
> really be in /usr/lib to start with; dlopen-able modules should go in 
> /usr/lib/<somedirectory>.

This has come up before, but the problem is that it'd break existing
user-written odbcinst.ini files.  I'm not convinced it's worth that.

			regards, tom lane