Return-Path: postman Delivery-Date: Wed, 15 Sep 93 12:57:22 PDT Return-Path: postman Received: by postgres.Berkeley.EDU (5.61/1.29) id AA18828; Wed, 15 Sep 93 12:50:29 -0700 Resent-From: postman (POSTGRES mailing list) Resent-Message-Id: <9309151950.AA18828@postgres.Berkeley.EDU> Sender: owner-postman@postgres.Berkeley.EDU X-Return-Path: aoki@postgres.Berkeley.EDU Received: from faerie.CS.Berkeley.EDU by postgres.Berkeley.EDU (5.61/1.29) id AA18820; Wed, 15 Sep 93 12:50:24 -0700 Received: from localhost (localhost [127.0.0.1]) by faerie.CS.Berkeley.EDU (8.1C/8.1B) with SMTP id MAA07371; Wed, 15 Sep 1993 12:50:53 -0700 Message-Id: <199309151950.MAA07371@faerie.CS.Berkeley.EDU> From: aoki@postgres.berkeley.edu (Paul M. Aoki) To: erik vangilder Cc: postgres@postgres.berkeley.edu Subject: Re: is postgres without dynamic load ok? In-Reply-To: Your message of Wed, 15 Sep 93 14:36:02 EDT <9309151834.AA17806@postgres.Berkeley.EDU> Date: Wed, 15 Sep 93 12:50:53 -0700 X-Sender: aoki@postgres.Berkeley.EDU X-Mts: smtp Resent-To: postgres-dist Resent-Date: Wed, 15 Sep 93 12:50:27 PDT erik vangilder writes: > Since postgres does not support my platform yet and I don't yet > have the time to implement the dynamic loader, I would like > to use postgress without the dynamic loader. My question is, > therefore, how usable is postgres without the dynamic loader? the answer, i'm afraid, is that it depends on what you want to do. if you don't intend to use user-defined types, or are willing to compile your type functions into the server, then there isn't any problem .. > On a related note, I think I remember reading of a HP/UX > implementation of postgres using shared libraries instead > of the dynamic loader. If shared libraries are as functional > as the dynamic loader and more portable, could the postgres > group support this method too? Would a port be easier using > shared libraries instead of mucking around with COFF internals? > (Speaking for only myself, but the COFF internals and dynamic > loaders seem rather daunting.) andrew yu recently converted the dynamic loader interface to assume a dlopen()-style interface. the up sides are: - it's pretty fast on most systems -- faster than ld -A. - since most major versions of unix support some kind of shared library interface this is a major portability win. (he wrote an emulator for ultrix.) the down sides are: - some shared library implementations are moderately broken. sunos4 has a few known problems in this regard (though i haven't yet bumped into them myself). - this pushes some link-editing complexity from inside of postgres into the user's lap, since the user gets to create loadable shared libraries. (there is sufficient wackiness in some interfaces that it's hard to automate reliably.) on most systems this is pretty easy. on aix (and svr4, i think) the appropriate link-editing can be a real pain. :-P and nobody, but NOBODY, has the same ld flags to create shared libraries. - almost everyone has a slightly different load mechanism ("load" on aix, "shl_open" on hpux, "dlopen" on sun/alpha, ...). still, it's better than trying to understand COFF/ECOFF/a.out/... -- Paul M. Aoki | CS Div., Dept. of EECS, UCB | aoki@postgres.Berkeley.EDU | Berkeley, CA 94720 | ...!uunet!ucbvax!aoki