DejaGnu 1.6.3 release candidate 5

Release candidate 5 for DejaGnu 1.6.3 has been uploaded to alpha.gnu.org
at <URL:ftp://alpha.gnu.org/gnu/dejagnu/dejagnu-1.6.3-rc5.tar.gz>.

Release candidate 5 should resolve the issues on Solaris systems.

In the interim since release candidate 4, I have applied for and been 
granted access to the GCC compile farm, which I have used to test on 
Solaris and AIX.  Many minor bugs have been corrected, including some 
that were uncovered by testing in different environments; details are in 
the Git history and ChangeLog.  I would like to thank the operators of 
the GCC compile farm and of the various machines thereby made available.

This release candidate ships with a patched copy of config.guess that is 
compatible with classic Bourne shells; these patches have also been sent 
to the config-patches list.


Known Bugs
----------

Two bugs were reported after the 1.6.3 code freeze and will be present
in the 1.6.3 release; both are expected to be fixed in 1.6.4:
    - PR44462
       - get_multilibs blindly assumes that the target compiler is GCC
    - PR44602
       - loading files fails if the full name contains spaces

One bug was reported in release candidate 1 and will be partially fixed
in 1.6.3, with more extensive work planned for 1.6.4:
    - PR47385
       - framework code can incorrectly accept directories as executable

While PR47382 is expected to be resolved before the 1.6.3 release, a
related issue will remain and is expected to be fixed in 1.6.4:  the
build/host/target triplets given to configure are not propagated to
site.exp.  This is ultimately a limitation in Automake, for which 1.6.4
is expected to carry a workaround, with efforts made to upstream a
solution to add this information to site.exp files by default.  Note
that GCC (and probably others) will not be affected by the future
Automake change, as they currently override Automake's site.exp rule.

Other known bugs include the incorrect handling of libdirs and
thread-safety issues with the C unit testing support in dejagnu.h.

The nawk(1) program on AIX 7.1 (on gcc111 at the compile farm) appears 
to have a strange bug that causes the totals reported by "dejagnu 
report-card" to be incorrect in some cases.  I am interested if anyone 
knows a workaround other than using GNU Awk.

Future Directions
-----------------

There are no changes to the future directions previously announced with
earlier 1.6.3 release candidates at this time.


-- Jacob