Mike Kupfer wrote:
> Roland> Are the test failures always the same ?
> 
> Yes and no.  The failure with path.sh was with BUILD_KSH93_AS_BINKSH=1.
> I only tried it once.

Can you try it again, preferably with B37 to avoid we're seeing any
"ghost" bugs caused by any libc/libsocket/libnsl/libm/etc.-flag day or
other kind of disturbance ?

> The failure with io.sh is with
> BUILD_KSH93_AS_BINKSH=0 and is repeatable.

See below...

> Roland> # Note that "builtins.sh" is allowed to fail until upstream fixes
> Roland> # -- snip --
> Roland> # "ld.so.1: ksh: fatal: relocation error: file
> Roland> xxx/usr/src/cmd/ksh/i386/ksh: symbol sh_waitnotify:
> Roland> # referenced symbol not found
> Roland> ../../../lib/libshell/common/tests/builtins.sh: line 313: 28473:
> Roland> # Killed        builtins.sh[314]: "name=value exec -c ..." not 
> working"
> Roland> # -- snip --
> 
> Okay, this will have to be addressed prior to putback

BTW: I reported this issue in
https://mailman.research.att.com/pipermail/ast-users/2006q3/001085.html
to upstream.

> either get the
> fix from upstream, or skip the test. 

Erm... the test is currently treated in a way which makes a failure
non-fatal. Isn't this enougth ?

> We might even want to change
> cmd/ksh/Makefile.com so that "all" doesn't depend on "testshell".

Umpf... only if we can't resolve the failures this week, Ok ?
I would be very unhappy with loosing the tests as part of the default
build because this could be a very usefull instrument to catch worse
things (like hiccups in libc&co.) very early...

> Roland> The failures in "path.sh" and "io.sh" are new for me... how does
> Roland> your ${PATH} look like ?
> 
> athyra--proto2$ echo $PATH
> /opt/onbld/bin:/opt/onbld/bin/sparc:/opt/SUNWspro/bin:/usr/ccs/bin:/usr/bin:/usr/sbin:/usr/sfw/bin:.

Mhhh... this should work...
In http://polaris.blastwave.org/changeset/325 I changed the way how
${PATH} is created - instead of adding some stuff I now use a "fixed"
string which only references content in $(ROOT)/ , however this is only
half-functional (at least in theory, to keep it working I added
"/usr/bin" as the last element in ${PATH}) since usr/src/cmd/ksh/ is
built in the mittle of the cmd/ commands and not at the end (of the list
used in usr/src/cmd/Makefile)). One possible solution may be to move
"ksh" to the end in "usr/src/cmd/Makefile", but I am not sure whether
this is allowed since the list seems to be in alphabetical order...

> Here's some more information:
> 
> I hacked cmd/ksh/Makefile.com to only run io.sh, and I reran the test
> with truss -f.  truss shows that close1 is getting made mode 775, and it
> looks like the echo command is getting written to it.  But the exec is
> indeed failing with ENOEXEC.
> 
> If I hack io.sh to add a "#! /bin/ksh" line to the start of close1, the
> test passes.

Ok... I can now reproduce the problem. It only happens when the
executable is not called "ksh" (like "ksh93" or "kshtentaclemonster")
... ;-(
Anyway, I reported this to upstream as
https://mailman.research.att.com/pipermail/ast-users/2006q3/001089.html
and commited http://polaris.blastwave.org/changeset/330 to allow that
"io.sh" fails until we figure out what's the best solution to fix this.

----

Bye,
Roland

-- 
  __ .  . __
 (o.\ \/ /.o) roland.mainz at nrubsig.org
  \__\/\/__/  MPEG specialist, C&&JAVA&&Sun&&Unix programmer
  /O /==\ O\  TEL +49 641 7950090
 (;O/ \/ \O;)

Reply via email to