Re: pgsql: Move pg_dump memory routines into pg_dumpmem.c/h and restore com - Mailing list pgsql-committers

From Andrew Dunstan
Subject Re: pgsql: Move pg_dump memory routines into pg_dumpmem.c/h and restore com
Date
Msg-id 4ED2C1CA.6060506@dunslane.net
Whole thread Raw
In response to Re: pgsql: Move pg_dump memory routines into pg_dumpmem.c/h and restore com  (Tom Lane <tgl@sss.pgh.pa.us>)
List pgsql-committers

On 11/27/2011 12:42 PM, Tom Lane wrote:
> Andrew Dunstan<andrew@dunslane.net>  writes:
>> MSVC is still broken with this change, but now I think we've exposed a
>> long-standing error in the MSVC build system. ...
>> AddSimpleFrontend() calls AddDir() which harvests the contents of
>> $(OBJS) from the Makefile for the target. But pg_dumpall doesn't want
>> $(OBJS). We've been benignly but mistakenly building it with them for a
>> quite a few years, but now we can't do that any more, given Bruce's
>> changes. It looks like the fix is to call AddProject() for pg_dumpall
>> instead of AddSimpleFrontend() and then do a little more work ourselves
>> to select exactly what we need.
> Maybe it would be better to refactor the makefile so that $(OBJS) is
> still sensible to use this way?
>
>

I don't think so. That would make things worse for the pg_dump and
pg_restore targets which do use $(OBJS) as it is now.

Attached is a slightly hackish but very small patch that I have tested
that resolves the problem. Unless there's an objection I'll apply it
shortly (with an explanatory comment). Essentially it still uses
AddSimpleFrontend, so we still get the useful things it does, like
linking in libpq and setting up Windows resources, but the removes the
unwanted sources for $(OBJS) files from the object. Then we add in just
the files we do want.

cheers

andrew

Attachment

pgsql-committers by date:

Previous
From: Tom Lane
Date:
Subject: pgsql: Use IEEE infinity, not 1e10, for null-and-not-null case in gistp
Next
From: Andrew Dunstan
Date:
Subject: pgsql: Make pg_dumpall build with the right object files under MSVC.