On Sun, 5 Jul 1998, Peter J. Braam wrote:

> Steven,
> 
> Libc 5 is really gone -- H. J.  Lu has just announced that he will do no
> more work on it.  We certainly won't be using it anymore and therefore
> likely even if we put your patches in now, they will gradually disappear
> again.

No problem.  I've had far too many problems with portability of system
utilities under glibc to switch all my boxes, so I just did what was
expedient in my environment.  For a quick example, try building the
user-land support for knfsd under Alpha linux.  I agree 1000% with Ted
Tso's sentiments; many ill-considered choices were made by the glibc
developers. 


> > --- lib-src/mlwp/lwp.c.orig	Wed Jun 24 09:48:05 1998
> > +++ lib-src/mlwp/lwp.c	Mon Jun 29 19:44:12 1998
> > @@ -133,6 +133,8 @@
> >  typedef void *register_t;
> >  #endif
> >  
> > +typedef unsigned long register_t;
> > +
> 
> This will have to go into the #ifdef clause.  Most systems have register_t
> and in that case we definitely don't want to redefine it.

register_t is missing in action for libc-5 under Linux.

> This won't go in.  The coda.h file is the _only_ file shared between the
> kernel and and userland.  I absolutely cannot include time.h without
> running the risk of a time conflict.  Perhaps you can tell me what the
> problem is in the compile, using MAKE CODA not just MAKE.

Under libc-5 (at least), the definition for 'timestruct' is missing.

(snip)

> If you do make coda (not make) then all these symlinks are unnecessary.
> The hdrinstall target (part of make coda) installs the include files in
> the include directory. 

Peter, I assure you that the coda source (as distributed on your ftp site)
will NOT build correctly without pre-making all the symlinks.  It fails
immediately (and repeatedly) with missing header files and defines.

And, yes, I WAS using 'make coda'.  Do me a quick favor and unpack one of
the rpm's on a non-networked vanilla box and try it yourself?  After
forcing the symlinks, it does eventually install the requisite headers.
However, they must pre-exist before the build can even get to the point
where the script installs them.  Catch-22.

I have seen this problem in every source release.  Sorry for not
mentioning it sooner.

Steve