AipsppNotes

Since this page was previously blank, I will provide some notes on issues I encountered when `porting' aips++ from Linux RedHat 9 to Scientific Linux 4. I had previously compiled aips++ in April 2005 with gcc 3.4.3 on RH9 and the easiest migration path turned out be to tar up the previously compiled system (libraries, binaries and all) and install it directly on the new SL 4 system. (gcc 3.4.3 is the defacto compiler with Sl 4.)

On RH9 glish was linked against Tcl/Tk 8.3 libraries. SL 4 comes with Tcl/Tk 8.4. So I downloaded Tcl/Tk 8.3 source code, compiled and installed it, and made soft links from /aips++/linux_gnu/lib to the Tcl/TK8.3 libraries et voila! (An attempt to directly compile glish against Tcl/Tk 8.4 produced a lot of compile errors which I didn't feel like tracking down.)

I then updated the whole system to the weekly build of July 9 2005 (V1.9, build 1112). Before you start the build make sure that you install cfitsio 3.0.2 beta. This will allow you to avoid aips++ FITS output errors (due to large file size interface problems) that seem to occur on occasion with the current cfitsio 2.51 package. This build successfully runs the aips++ scripts that are being used to create CLAR simulations. One thing that no longer works properly is the aips++ image display (the aips++ group are currently migrating to the Qt widget system). However I don't care as I always disliked the aips++ display and prefer to look at output FITS files with kview.

Note that aips++ no longer seems to build glish by default. As I mentioned above the starting point for this particular exercise was a direct copy of a working April 22 build so I'm using the previously built glish.

July 20 - Ger van Diepen suggests doing the following to get glish to build (do this before the rest of the installation):

  . aipsinit.sh
  cd code/aips/glish
  make
  cd ../..
  make

Beware of the following: the structure of /proc/meminfo has changed from 2.4 to 2.6 Linux kernels. This means aips++ will (as of July 22, 2005) not extract the right value for the actual amounts of real and available memory on a system with a 2.6 kernel. Consequently big imaging applications will swap everything out to disk and processing takes FOREVER. A fix for HostInfoLinux.h is available (contact Tony Willis) and has been sent to the aips++ group, but I'm not sure when it will be put into the aips++ distribution.