[postgis-users] 1.3.0 comments from a packaging perspective

[Greg Troxel]

Mark Cave-Ayland <mark.cave-ayland at ilande.co.uk> writes:

> On Tue, 2007-08-21 at 21:01 -0400, Greg Troxel wrote:
>> I am in the midst of updating the postgis package in pkgsrc to 1.3.1.
>> 
>> templategis has been removed from the build/install targets (r2652) I
>> found no NEWS entry about this, or anything in the README.  As a
>> packager, should I take this as a hint that templategis is deprecated,
>> and binary packages should not include it?  This seems worthy of NEWS as
>> a user-visible change in what is installed.
>
> Well, if you actually got template_gis to build in the first place then
> you are in the minority ;)

I haven't got it to work, but the build/install finished.  Looking at
it, I immediately see issues.

> The code for template_gis wasn't really reviewed before commit, and so
> contains several debian-only commands which caused it to fail on just
> about every other distribution (in particular it tries to write to /etc
> which broke every installation as a non-root user). There have been lots
> of emails about this on the list, so apologies if you weren't aware of
> this.

OK - in my view a NEWS entry is in order for the removal from the
default install, as well as changing the README in templategis to note
the status.

I was aware of the /etc issues, but that was patched around in pkgsrc,
and I noticed that the config file was in a new place within $prefix in
this release.  But I didn't follow that running it caused problems.

I'll just omit it from the package.

>> I build with -j4 on a dual-processor system (NetBSD/i386, Thinkpad T60).
>> The build failed, and restarting it got:
>
> Not sure about this - I can't say I've ever built PostGIS in a
> multi-threaded fashion...

Perhaps you could try simply 'gmake -j4' after 'gmake clean'?  My view
is that a package that doesn't build this way should be fixed.  I'll see
if I can figure out the issue and send a patch.

>> There are some warnings (full logs on request):
>
> The GEOS warnings are known about - unfortunately the code has to be
> written in this way to allow PostGIS to build with both GEOS 3 and GEOS
> 2. The dbfopen.c warning comes from our version of shapelib - I've never
> really looked into this in any detail as it comes from an external
> source, and the iconv() message I remember seeing here before but never
> really explored in detail.

Thanks - I guessed these had been seen before.


Also, here's a patch from pkgsrc for DragonFly:

$NetBSD: patch-aa,v 1.3 2006/12/06 17:43:27 joerg Exp $

--- lwgeom/Makefile.shlib.orig	2006-12-06 17:36:49.000000000 +0000
+++ lwgeom/Makefile.shlib
@@ -140,6 +140,10 @@ ifeq ($(PORTNAME), freebsd)
   endif
 endif
 
+ifeq ($(PORTNAME), dragonfly)
+  LINK.shared		= $(COMPILER) -shared -Wl,-x,-soname,$(soname)
+endif
+
 ifeq ($(PORTNAME), netbsd)
   ifdef ELF_SYSTEM
     LINK.shared		= $(COMPILER) -shared -Wl,-x,-soname,$(soname)