Bruce Momjian wrote:
> Tom Lane wrote:
> > Bruce Momjian <bruce@momjian.us> writes:
> > > It has been years since I ran src/tools/pginclude/pgrminclude to remove
> > > unnecessary include files. (I have already fixed things so include
> > > files can be compiled on their own.)
> >
> > > The attached patch removes unneeded include references, and marks some
> > > includes as needing to be skipped by pgrminclude.
> >
> > > I am sure applying this patch will break builds on some platforms and
> > > some option combinations so I will monitor the buildfarm when I apply it
> > > and make adjustments.
> >
> > The last time you did this was in July 2006. It took us two weeks to
> > mostly recover, but we were still dealing with some fallout in December,
> > cf
> > http://archives.postgresql.org/pgsql-hackers/2006-12/msg00491.php
> >
> > We had the buildfarm then, had had it for a couple years. The notion
> > that watching the buildfarm is enough is fully disproven by history.
> >
> > Unless you have a better test plan than last time (which this isn't),
> > I don't think this should be done at all. The benefits are microscopic
> > and the pain real.
>
> I don't have a better plan. There are #ifdef code blocks that often
> don't get processed and therefore this can't be done better. I will
> abandon the idea.
OK, try #2. I already had code that removed #if/#else/#endif code in
*.h files for better testing, so I extended that to all *.c files. This
reduces the size of the diff from 6.6k lines to 4.7k lines but it makes
it much less likely that there will be problems from running
pgrminclude.
The current patch is here:
http://momjian.us/expire/pgrminclude.diff
I tested the patch on BSD and Linux.
-- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB
http://enterprisedb.com
+ It's impossible for everything to be true. +
