Joerg Sonnenberger <joerg%britannica.bec.de@localhost> writes: > On Mon, Jan 24, 2011 at 10:48:53PM +0100, Adam wrote: >> I would like to remove libtool as a building tool for PostgreSQL. >> >> Arguments: >> >> 1. Libtool is not supported by PostgreSQL. Patching PostgreSQL to >> use libtool is a hack and maintaining the package is a headache. > > So is maintaining different linker rules. Yes, but that is what upstream is doing. Adam: can you comment on platforms where pgsql would build/work differently with native linking vs forced-libtool? Sounds like we'd gain OS X, and it's not clear to me if we'd lose anything. >> 2. With libtool, there are problems with other packages, like PostGIS. >> 3. Without libtool, shared libraries are built just fine. We have >> other packages with shared libraries built without libtool >> (textproc/icu comes to mind). > > I consider the latter more a bug and we have enough packages where the > native build system is completely wrong for anything but Linux... Agreed, but PostgreSQL is more reasonable than Linux-only code. >> 4. On Mac OS X, PostgreSQL crashes when built with libtool (see >> PR/44365). I guess other platforms might suffer, too. > > So we get the real reason. We have a platform that is known to behave > differently depending on a number of factors. We don't even try to debug > the problem, we just completely change the environment. Very > convincing... It's worse than 'mac is different'. On NetBSD, I chased an apparent problem in postgis for a while that was due to our putting libtool into pgsql when upstream did not. upstream postgis does not want to hear about problems arising from building pgsql other than as instructed by pgsql. Perhaps Adam could mail the proposed diff.
Attachment:
pgpYqmILrQi0o.pgp
Description: PGP signature