Re: 2.4.5-ac4 es1371.o unresolved symbols

[Vojtech Pavlik]

On Thu, May 31, 2001 at 05:52:39PM +1000, Keith Owens wrote:

> On Thu, 31 May 2001 08:08:45 +0200, 
> Vojtech Pavlik <[e-mail]> wrote:
> >On Thu, May 31, 2001 at 11:29:06AM +1000, Keith Owens wrote:
> >> With your patch, if a user selects CONFIG_INPUT_GAMEPORT=m and
> >> CONFIG_SOUND_ES1370=y then the built in es1370 driver has unresolved
> >> references to gameport_register_port() which is in a module, vmlinux
> >> will not link.  That is why I derived CONFIG_INPUT_GAMEPORT based on
> >> the config options in two separate directories.
> >
> >Have you tried the patch? Because the gameport.h define has:
> >
> >#if defined(CONFIG_INPUT_GAMEPORT) || (defined(CONFIG_INPUT_GAMEPORT_MODULE) && defined(MODULE))
> >void gameport_register_port(struct gameport *gameport);
> >void gameport_unregister_port(struct gameport *gameport);
> >#else
> >void __inline__ gameport_register_port(struct gameport *gameport) { return; }
> >void __inline__ gameport_unregister_port(struct gameport *gameport) { return; }
> >#endif
> 
> When the user has gameport hardware compiled it as a module and they
> have es1371 bult into the kernel then es1371 silently ignores the
> gameport, even if the gameport modules has been loaded.  This violates
> the principle of least surprise, a user configuring both gameport and
> es1371 expects to use the gameport, kbuild should support that instead
> of silently ignoring the combination.

True. Is this worse than the ugliness in your patch?

-- 
Vojtech Pavlik
SuSE Labs