This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [Patch] Adding fedfs to /etc/rpc


On Thu, 8 Mar 2012, Ulrich Drepper wrote:

> On Wed, Mar 7, 2012 at 07:42, Joseph S. Myers <joseph@codesourcery.com> wrote:
> > I think making libtirpc a build dependency is bad - in general circular
> > dependencies should be avoided where possible - as it makes bootstrapping
> > unduly hard.
> 
> Some people goal to do everything themselves by bootstrapping an
> entire system cannot possibly be allowed to have any influence on the
> design.  All that counts is the flexibility and robustness of the
> system.

In my view, flexibility and robustness includes build system flexibility 
and robustness for a wide range of uses to which people wish to put glibc.  
And it improves flexibility and robustness to have clean interfaces with 
good support for building pieces separately rather than having circular 
dependencies.  That's not just about NSS modules and libtirpc; there are 
several other ways in which I think it would be good to split up the 
source tree and build.

Given the disagreements, we should clearly seek a wider range of opinions 
to get a rough community consensus on the way to handle this.  (Testing to 
establish just what needs doing to make libtirpc ready for this purpose 
would also be good; it's unfortunate we didn't do that before the sunrpc 
code in glibc was deprecated, given subsequent findings that libtirpc is 
still not really ready to replace all the deprecated functionality from 
glibc.)  Once there is a community consensus, that can then be followed in 
subsequent development.  Saying "cannot possibly be allowed to have any 
influence" is clearly premature at this point and prejudges what the 
community might decide; we need to see what conclusion is reached by the 
public discussion among people thinking carefully about the issues.  What 
matters in the end is the community consensus at the point when libtirpc 
is ready, and it seems known not to be ready yet.

> Not having an nss module like that in glibc causes problems as anyone
> who tried to use the ldap modules has seen.  After every interface
> improvement it takes ages to (if ever) for those modules to adapt
> them.  Take nss_mdns. Lennart a long time ago brought up that the
> gethost* interfaces should be different to better handle mdns.  I made
> the change years ago and still the nss_mdns module doesn't use the new
> interface despite the authors being told about it.

You can have NSS modules that are maintained by the glibc developers, but 
in a separate repository and released at the same time in separate 
glibc-nss tarballs, built separately rather than as part of the glibc 
build.  That way glibc developers can update them as desired for new 
interfaces, without having circular build dependencies.  I think that 
would address both our concerns.

-- 
Joseph S. Myers
joseph@codesourcery.com


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]