[daip] RedHat 9 Compiler problem?

[Craig Walker]

I installed AIPS on my home machine on 17 June 2003.  I have RedHat 9 with 
g77 and gcc 3.2.2.  Here is the main output from g77 -dumpversion:
g77GNU Fortran (GCC 3.2.2 20030222 (Red Hat Linux 3.2.2-5)) 3.2.2 20030222 (Red Hat Linux 3.2.2-5)

I installed with OPT2="-O2" in FDEFAULTS.SH and OPT0="-O3 ..." in
CCOPTS.SH.  I had two serious problems.  When POSSM (and another task)
were asked to do a bandpass calibration, the task simply died of
unnatural causes with no hints as to why.  The same inputs, but with
DOBAND=-1 produced reasonable results.  When IMAGR was asked to use
the multiresolution clean option, with inputs that otherwise worked
fine, something went terribly wrong.  The sidelobes were roughly
circular on a 4 degree dec source, and the DECSHIFT did not offset the
source by the requested amount.  The output source position and
offsets in the header seemed fine.  A back transform of the beam
showed UV tracks that all had their maximum extent at V=0, but their
range in U was about right.  The range in V seemed a bit larger than
expected, but not to the full extent of a polar source.  A wild
speculation is that UVW were recalculated but with the station Z
coordinate zeroed.

Tonight I blew away the whole AIPS installation and reinstalled with
OPT2="-O0" in FDEFAULTS.SH and OPT0="-O0" in CCOPTS.SH.  Both the
bandpass calibration problem and the IMAGR multiresolution problem are
now gone.  All seems to be working fine, although presumably slower
than before.

Note that I had tested the bandpass calibration problem at the AOC
using Eric's version of the RedHat 9 compilers.  I used a data set the
showed the problem at home, even after a reload from the FITS transfer
tape.  The problem did not occur there.

The compiler saga continues, I fear.  Or is there something else I
could have done to mess up the installation that might cause such
effects?

Craig