Thread: more massaging on pgxs postresql extension infrastructure
Dear patchers, after Peter's massaging on pgxs, I think the infrastructure deserves some more massaging because: (a) some files are missing (namely libpgport.? needed by pgbench for instance, and I guess possibly by others). (b) I think it is a key feature that one should be able to compile contrib with the already installed postgresql, without having to reconfigure. so I submit this new patch. I've tried to preserve Peter modifications to my initial submissions, plus: (1) add libpgport installation under "install-all-headers" target (2) all contrib Makefiles can ALSO use of the dynamic pgxs stuff => no direct inclusion of Makefile.global (which is generated by configure) => USE_PGXS macro enables that ("make USE_PGXS=1 install"), otherwise it is just as before the patch, the static "configured" infrastructured is used. If there are still issues, which is perfectly possible, I'm really willing to fix them while preserving the ability to compile postgresql contribs with an already installed postgresql. Basically it works for me. Thanks in advance, -- Fabien Coelho - coelho@cri.ensmp.fr
Attachment
Am Mittwoch, 11. August 2004 16:25 schrieb Fabien COELHO: > (b) I think it is a key feature that one should be able to compile > contrib with the already installed postgresql, without having > to reconfigure. Why? I think it would be an entirely useless feature. -- Peter Eisentraut http://developer.postgresql.org/~petere/
Dear Peter, > Am Mittwoch, 11. August 2004 16:25 schrieb Fabien COELHO: > > (b) I think it is a key feature that one should be able to compile > > contrib with the already installed postgresql, without having > > to reconfigure. > > Why? I think it would be an entirely useless feature. (1) this "entirely useless feature" would have simplified *my* little life once for some tests I needed. (2) I used contrib to test that pgxs works! I think they are needed to test that it continues to work. That is how I figured out that libgport was missing, because I could not compile pbbench with PGXS. (3) if there is an infrastructure for extensions, there is no reason why we could not demonstrate it on contribs, if it is available for tests anyway, and as it may be useful to other people. -- Fabien Coelho - coelho@cri.ensmp.fr