nan [was: grftest failure]

[David Relson]

At 01:12 PM 1/5/03, Clint Adams wrote:

> > The next thing to try is building without optimization (remove "-O2" in
> > Makefile) and see what happens.  If nothing changes, I guess the next step
>
>Using -O0, t.grftest passes when not run from make, but fails in make
>check.  Now the only differences from CVS are configure.in and the sed
>expression in t.frame.

Sounds like you're having fun :-)  It begins to sound like two problems - 
(1) production of results and (2) recognition of correct/incorrect 
values.  Stated differently, they sound like (1) compiler (as in rounding 
or optimization or ???) and (2) shell (as in return codes, possibly the 
"set -e" command).

A couple of ideas:

modify t.frame so it doesn't delete the output directories and then compare 
the t.grftest runs.

modify t.systest.d/Makefile so it runs t.grftest in verbose mode and 
compare runs.

> > is for you to send me a tarball of your directory including executables
> > object files, etc.  Alternatively, I can send my copy to you.  Is is
> > possible for me to access the machine on which you're working?
>
>No.  spe170.testdrive.compaq.com is similar, but is 64-bit and has
>only gcc 3.0.4, which is buggy.

I just another look at the 'uname' output you sent and saw the 
'parisc'.  Given that, asking for executables wasn't such a good idea.