[bananas] Linux woes

[Eric Greisen]

Several matters have come up that I want to draw to people's
attention:


1. The Linux 2.2.14-5 kernel (which ships with Redhat 6.2) as well
   as kernels 2.2.13-0.13 and 2.2.14-6.1 suffer from a context switch
   bug.  The manufacturers state:
 
     "Red Hat, Intel and Dell have uncovered a problem with the Red Hat
      Linux 6.2 for the x86 (Intel) processor. This problem has been
      duplicated and confirmed in our lab, though we have had no reports
      from customers at large. This problem affects all OEM system
      manufacturers shipping Red Hat."

   We (AIPS) were the first to duplicate it in the wild:
      IMAGR and FRING on largish problems can end up producing
      strange answers or NaNs unpredictably and unrepeatably.

   The problems were fixed in kernel version 2.2.16-3, so any 2.2.17-x
   or later should be free of this bug.  

                     -------------------------------

2. The AIPS group has received a report from a user who attempted to
   install and run AIPS on a PC running Red Hat 7.0.  The GNU
   compilers (gcc, g77) included with this version of Red Hat were
   merely a snapshot ("2.96") of a development system and not a
   properly released version, something for which the Red Hat people
   subsequently got severely criticized.  The "current" release of
   gcc/g77 at the time was (and still is) 2.95.

   While AIPS will appear to compile and link on such a system, there
   are significant problems:

   -  TVFLG fails to run (infinite loop)
   -  GRBOXS did odd things when resetting > 1 box
   -  UVMAP seems to taper all the data
   -  IMAGR ignores histogram, tries to load all pixels in Clean boxes

   and maybe more.  Recompiling with NOOPT seems to fix things, but
   then the performance of IMAGR severely degrades (you do not want to
   do this!)  Such subtle errors are unusual in modern Fortran
   optimizing compilers and are very worrisome.  The hard stop in
   TVFLG is actually far less serious than the others.

   Our Linux systems are using g77 version 0.5.24 dated 19981002 with no
   known problems; the RedHat 7.0 system (suspect) g77 was listed as
   0.5.26 dated 20000731 both from g77 --version.  The version IDs can
   also be examined with "rpm -qf /usr/bin/g77" to find out which rpm
   "g77" belongs to.  On our systems (updated RedHat 6.2) the answer
   is egcs-g77-1.1.2-30, on RedHat 7.0 it is gcc-g77-2.96-54.

   We have no idea whether one could substitute the older g77 from
   RedHat 6.2, but still run on a RedHat 7.0 system.

   We do not recommend trying AIPS under RedHat 7.0.  A more stable
   release 7.1 is expected in April or so and we will test AIPS under
   that as soon as possible.  (We already have a beta version in
   house.)

                         -------------------------------

3. One Linux 6.2 site has had an optimization problem with
   $APLNOT/GRIDTB causing TVFLG to go into an infinite loop while
   trying to build the grid file.  To cure this, do:

         COMRPL $APLNOT/GRIDTB NOOPT 

   followed by 

         COMLNK $YPGNOT/TVFLG

   I do not know why it has not been seen on sites with apparently the
   same compiler.