Обсуждение: 'configure --disable-shared' and 'make check'

I notice that when I run 'make check' on a
statically linked HEAD, it fails during
'createlang' with

============== installing plpgsql                     ==============
ERROR:  could not access file "$libdir/plpgsql": No such file or
directory
command failed:
"/postgres/cvs/pgsql/src/test/regress/./tmp_check/install//magwien/postg
res-8.2.a/bin/psql" -X -c "CREATE LANGUAGE \"plpgsql\"" "regression"

This is not a problem of the regression test
since I see the same behaviour when I do it manually.

Yours,
Laurenz Albe

Albe Laurenz wrote:
> I notice that when I run 'make check' on a
> statically linked HEAD, it fails during
> 'createlang' with

Because createlang relies on *dynamic* loading.

-- 
Peter Eisentraut
http://developer.postgresql.org/~petere/

Peter Eisentraut wrote:
>> I notice that when I run 'make check' on a
>> statically linked HEAD, it fails during
>> 'createlang' with
>
> Because createlang relies on *dynamic* loading.

So that is working as designed.
I interpret that as 'static builds for the database
server are not supported'. Ok by me.

Yours,
Laurenz Albe

"Albe Laurenz" <all@adv.magwien.gv.at> writes:
> Peter Eisentraut wrote:
>> Because createlang relies on *dynamic* loading.

> So that is working as designed.
> I interpret that as 'static builds for the database
> server are not supported'. Ok by me.

Well, we're not supporting dynamically linked objects in a static build ;-)

It's at least theoretically possible that you could link selected PL
objects into a static backend build, but no one is particularly
interested in expending effort on it.  There don't seem to be any
platforms anymore on which --disable-shared is actually important.
(Should we just get rid of it?)
        regards, tom lane

Tom Lane wrote:
> It's at least theoretically possible that you could link selected PL
> objects into a static backend build, but no one is particularly
> interested in expending effort on it.  There don't seem to be any
> platforms anymore on which --disable-shared is actually important.
> (Should we just get rid of it?)
>
>     
>   

There is just one case where I think static linking might make some 
sense: pg_dump for use during an upgrade. Unfortunately, at least on 
Linux I found it to be close to impossible, as it uses some things that 
are apparently only available dynamically - I forget the details.

So basically, I would say yes, let's just drop --disable-shared.

cheers

andrew

Am Donnerstag, 21. September 2006 16:02 schrieb Tom Lane:
> There don't seem to be any
> platforms anymore on which --disable-shared is actually important.
> (Should we just get rid of it?)

IIRC, I added it in the past to test for static-only platforms like QNX.  I 
don't know if we plan to support such platforms in the future.

-- 
Peter Eisentraut
http://developer.postgresql.org/~petere/

Tom Lane wrote:

> 
> Well, we're not supporting dynamically linked objects in a static build ;-)
> 
> It's at least theoretically possible that you could link selected PL
> objects into a static backend build, but no one is particularly
> interested in expending effort on it.  There don't seem to be any
> platforms anymore on which --disable-shared is actually important.
> (Should we just get rid of it?)
> 
>

ISTR having to use --disable-shared to get a working profile-able build 
on some platforms (might have been Solaris 8 with sun complier, but I'm 
not really sure as it was a while ago).

Cheers

Mark