Re: 2.5 kbuild: use of '-z muldefs' for LD?

[Sam Ravnborg]

On Mon, Jun 09, 2003 at 01:04:38PM +0100, Christoph Hellwig wrote:
> On Mon, Jun 09, 2003 at 01:56:59PM +0200, Jaroslav Kysela wrote:
> > one object file for more targets. Example:
> > 
> > ------
> > snd-ice1712-objs := ice1712.o delta.o hoontech.o ews.o ak4xxx.o
> > snd-ice1724-objs := ice1724.o amp.o revo.o aureon.o ak4xxx.o
> > 
> > # Toplevel Module Dependency
> > obj-$(CONFIG_SND_ICE1712) += snd-ice1712.o
> > obj-$(CONFIG_SND_ICE1724) += snd-ice1724.o
> > ------
> > 
> > The ak4xxx.o module is shared and has defined a few public functions.
> > Unfortunately, the default build-in.o rule fails when targets are 
> > requested to be included into the solid kernel because the public 
> > functions are duplicated in snd-ice1712.o and snd-ice17124.o.
> > 
> > I can instruct the ld compiler to ignore the multiple definitions using 
> > '-z muldefs':
> > 
> > EXTRA_LDFLAGS = -z muldefs
> > 
> > But it seems like a hack for me.
> > Does anybody have another idea to solve my problem?
> 
> Move ak4xxx.o out of the multi-obj rules.  Just declare a new helper-
> config option CONFIG_SND_AK4XXX that gets defined by all drivers
> using it and add
> 
> obj-$(CONFIG_SND_AK4XXX)	+= ak4xxx.o

Would it be worthwhile to resolve common functions from a library instead?
On request from Linus I made the lib-y change, and it is getting
a lot easier to create libraries.
So ak4xxx.o would be used to create lib.a in that particular directory.

A limitation would be that libaries would only be valid for current
directory - but that is OK for this situation.

On the other hand there should be very good reasons to clutter up the
build-system with this, so more users than sound is required.

Comments?
[Will there be problems with modules exporting symbols?]

	Sam