[Pacemaker] Proposing patch for ld --as-needed

[Simon Horman]

On Wed, Jul 21, 2010 at 09:40:10PM +0200, Ultrabug wrote:
> On Wednesday 21 July 2010 03:49:42 Simon Horman wrote:
> > On Tue, Jul 20, 2010 at 05:35:01PM +0200, Ultrabug wrote:
> > > ----- Original Message -----
> > > From: Simon Horman <horms at verge.net.au>
> > > To: The Pacemaker cluster resource manager
> > > <pacemaker at oss.clusterlabs.org> Sent: Tue, 20 Jul 2010 05:10:32 +0200
> > > (CEST)
> > > Subject: Re: [Pacemaker] Proposing patch for ld --as-needed
> > > 
> > > On Sat, Jul 17, 2010 at 01:12:20PM +0200, Ultrabug wrote:
> > > >Dear list,
> > > >
> > > >I would like to ask you about a possible upstream modification regarding
> > > >the -- as-needed ld flag for which we Gentoo users need to patch the
> > > >pacemaker sources to get it compile.
> > > >
> > > >I'm attaching the patch which, as you can see, is relatively small and
> > > >simple (looks to me at least). The question is whether or not you think
> > > >this could be done upstream ?
> > > >
> > > >Thank you for your interest in this and all you work,
> > > 
> > > Out of interest, could you explain why this is needed?
> > > Is it because gold is being used as the linker?
> > 
> > [ please don't top-post ]
> 
> [ noticed after sending, sorry ]
> 
> > 
> > Thanks for the link.
> > 
> > I guess what is happening without --as-needed is that the curses
> > library is being dragged in somewhere, somehow - without your proposed
> > change CURSESLIBS is used exactly nowhere.
> > 
> 
> Actually the two chunks of the patch have different purposes.
> 
> The first one is needed because the linking order has a meaning on a as-needed 
> system and libpengine uses functions that are defined in libpe_status.
> 
> Here is an example of failing build : http://paste.pocoo.org/show/239905/
> If you try to compile a program that uses libpengine that doesn't need 
> libpe_status, e.g. gcc -Wl,--as-needed ptest.c -o ptest -lpe_status -lpengine 
> (shortened version of actual linking from pacemaker build.log) linking will 
> fail. Linker evaluates that statement sequentially starting from the inner 
> most lib:
>   1) do I need libpe_status? No, forget about it.
>   2) do I need libpengine? Yes, please.
>   3) is everything all right? Ups, I don't know what `was_processing_warning' 
> is, die...
> 
> 
> The second one explicitly adds CURSESLIBS dependency to libpe_status.
> If pacemaker detects ncurses, you get HAVE_NCURSES_H and e.g.
> status_print (used in  lib/pengine/native.c etc.) becomes wrapper
> around "printw" (see configure.ac). You need to provide `printw' or any linking 
> with libpe_status will fail.
> Fail build example : http://paste.pocoo.org/show/239916/
> 
> > So I think that your change is a step in the right direction,
> > though for completeness I think that you also need to give the same
> > treatment to libpe_rule as common.c seems to make curses calls.
> > 
> > Could you considering updating your patch to include my proposed
> > additions below? And could you please include a description that describes
> > what the patch does? Perhaps something like this:
> > 
> 
> Sure, if you agree with the explanations above, I'll summarize them and add 
> them in the patch which I'll resubmit to you for integration.

I think tat you have a better handle on this problem than me.
So yes, please summarise your explanation above and use
it as a preamble to the patch.

[snip]