Re: [patches] Patch for OPTION_EGLIBC_NSSWITCH option group.

[Jim Blandy <jimb@xxxxxxxxxxxxxxxx>]

Daniel Jacobowitz <dan@xxxxxxxxxxxxxxxx> writes:
> On Thu, Jul 12, 2007 at 07:21:49PM -0500, Mark Hatle wrote:
>> Would it work to include the default (compiled in?) name service
>> libraries in the libc.so linker script?
>> 
>> I'm just trying to think of an automated way to include the libnss
>> libraries w/o having to modify the applications.
>
> When I made static NSS work for a customer recently, I dumped those
> libraries directly into libc.a.  This is the analagous problem for
> dynamic linking.  Jim, do you think that the static analysis benefits
> are really worthwhile?

I don't know, myself.  But the idea for the nsswitch option group came
out of discussion with the EGLIBC partners, and I believe the
motivation was the desire to eliminate uses of dlopen, to help things
like libopt.

The way the patch is set up at present, the system has no way to tell
whether you've used some functions from a service library, but not
others; they're all referenced from some service_user's 'known'
function list.  If I stare at the getXXbyYY_r.c files a little more, I
might be able to see how to clean that up.

If I remember correctly, there's also a dlopen in the gconv character
set conversion stuff.  I need to refine the OPTION_EGLIBC_LOCALE
option group so that goes away.  Then we can look for others.