[bug-inetutils] Tracking of build failures.

[Mats Erik Andersson]

Hello all,

let me, inspired by the information from Ludovic Courtès, start a long time
thread for recording and discussing build failures of Inetutils, internal
and external.

Reading the build log journal at

    http://hydra.nixos.org/jobset/gnu/inetutils-master/

I can immediately record three causes for failures:

  1) FreeBSD fails due to an issue with IFF_NOTRAILERS.

     This will we be settled once 

        http://lists.gnu.org/archive/html/bug/inetutils/2010-11/msg00012.html

     is committed, for which I have gotten an ACK myself. The only additional
     showstopper for FreeBSD is contained in the same set.

  2) Darwin fails due to a missing <osockaddr.h> needed for the TALK client.
     On one hand, GNU/Linux and OpenBSD (and probably FreeBSD) define

        /* <sys/socket.h> */

        struct osockaddr;

     but do not define HAVE_OSOCKADDR_H. Therefore they succeed in compiling
     the client "talk". On the other hand, Darwin somehow believes it must
     define "HAVE_OSOCKADDR_H=1" on its own initiative. This leads straight
     into the build failure, since `talk/ctl.o' is then compelled to include
     `<osockaddr.h>', which in turn fails miserably.

     An analysis at the stage "checking for struct osockaddr" in "config.log"
     could possibly lead to a solution.

     Is the test for `struct osockaddr' correctly written in `configure.ac'?

     Could Darwin be using a location of `<osockaddr.h>' at a directory deeper
     than the top inclusion directory?

     Should we eliminate the build of `talk' and `talk' for any system where
     neither `struct osockaddr' nor `<osockaddr.h>' exist? Similar to the
     exclusion of `ping6' whenever "IPV6 != 1".

  3) GNU/Linux fails due to a suboptimal form of the script `tests/tftp.sh',
     an error unfortunately caused by my own efforts for portability! The 
present
     issue concerns paths to utilities on a running system, presently `netsed'
     and `egrep'. As a side note, once 1) is resolved, FreeBSD will be trapped
     also here, at least with the Hydra-NixOS build daemon.

     The good thing is that the same test script succeeds on every system with
     either GNU/Linux, GNU/kFreeBSD, OpenBSD, or FreeBSD I personally so far
     have considered. It is the more exotic systems that seem to intervene here.


Is there any information available for OpenSolaris, AIX, HP-UX, NetBSD, MirBSD,
Dragonfly, etcetera?


Regards,
Mats