View on GitHub

IRAF Community Distribution

IRAF maintained by the community

Home | Installation | Packages | X11IRAF | PyRAF | Forum

2.13beta2 on Mac OS X/ppc 10.3.9:problems with float images?

Emilliano Gregori wrote on Jun 22, 2006

Hello Mike,
I installed 2.13beta2 ppc binaries on my powerbook G4 with OS X 10.3.9.

The first problem is that ecl crashes, because it needs libncurses 5.4. The version installed by 10.3.9 is 5, but the problem is easily solved installing version 5.4 with fink, and then making a symbolic link in /usr/lib.
In this way ecl starts, and everything seems ok.

But then I get the following strange problem: when I run various image manipulation tasks (imcombine, imhist, pcol, prow, disp...) with float images (BITPIX=-32), I get a "bus error" message. Maybe could be the same problem of this post ?


authored by: caspin on Sunday, June 18 2006 @ 01:52 AM MDT
Mike, I tried the 2.13-beta2 port to Mac Intel and got a "Bus Error" when trying
to run scombine. I went back to 2.12a (PPC) and the same command worked
ok. Went back to 2.13-beta2 and got the bus error again - Colin


With integer images everything seems ok... Any hint? Maybe, if you compiled ppc binaries on 10.4, could it be some kind of compiler ABI or system libraries mismatch?

Many thanks,
Emiliano.

Mike Fitzpatrick wrote on Jun 22, 2006

Emilliano,

I'm booting and installing on a 10.3.9 system now and will have a look. Note the earlier post was a 10.4 MacIntel system using a particular task. I'd be surprised if everything using float images failed and nobody's reported it yet. I'll follow up when I have it tracked down.

Cheers,
-Mike

Emilliano Gregori wrote on Jun 22, 2006

This is the problem:


PACKAGE = immatch
   TASK = imcombine

input   =           bias*.fits  List of images to combine
output  =               b.fits  List of output images
(headers=                     ) List of header files (optional)
(bpmasks=                     ) List of bad pixel masks (optional)
(rejmask=                     ) List of rejection masks (optional)
(nrejmas=                     ) List of number rejected masks (optional)
(expmask=                     ) List of exposure masks (optional)
(sigmas =                     ) List of sigma images (optional)
(logfile=               STDOUT) Log file
(combine=              average) Type of combine operation
(reject =                 none) Type of rejection
(project=                   no) Project highest dimension of input images?
(outtype=                 real) Output image pixel datatype
(outlimi=                     ) Output limits (x1 x2 y1 y2 ...)
(offsets=                 none) Input image offsets
(masktyp=                 none) Mask type
(maskval=                    0) Mask value

Jun 23  0:09: IMCOMBINE
  combine = average, scale = none, zero = none, weight = none
  blank = 0.
                Images 
  bias.00000001.BIAS.fits
  bias.00000002.BIAS.fits
  bias.00000003.BIAS.fits
  bias.00000004.BIAS.fits
  bias.00000005.BIAS.fits
  bias.00000006.BIAS.fits
  bias.00000007.BIAS.fits
  bias.00000008.BIAS.fits
  bias.00000009.BIAS.fits

  Output image = b.fits, ncombine = 9
dataio> imhist b.fits
ERROR: bus error


The bias*.fits are 16bit "unsigned" integer:

SIMPLE  =                    T/CCDSOFT-SOFTWARE BISQUE 2           
BITPIX  =                   16                                       
NAXIS   =                    2               
NAXIS1  =                 1024                                       
NAXIS2  =                 1024                            
BSCALE  = +1.000000000000e+000                
BZERO   = +3.276800000000e+004                                   
....


and this is the b.fits image:

SIMPLE  =                    T / Fits standard    
BITPIX  =                  -32 / Bits per pixel                  
NAXIS   =                    2 / Number of axes             
NAXIS1  =                 1024 / Axis length                 
NAXIS2  =                 1024 / Axis length                
EXTEND  =                    F / File may contain extensions   
ORIGIN  = 'NOAO-IRAF FITS Image Kernel July 2003' / FITS file originator
DATE    = '2006-06-22T21:11:41' / Date FITS file was generated  
.....


The funny thing is that image b.fits created by imcombine is a good image... I can display it with ds9... but I can't use it within iraf tasks!

I also downloaded some fits floating point image from here:

http://www.cv.nrao.edu/fits/data/samples/image/

I have the same problem with primary HDU of file vogtstar_awt.fits... Again: I can display the image in ds9.

Cheers,
Emiliano.

Mike Fitzpatrick wrote on Jun 22, 2006

There are apparently two issues here, although one is clearly a bug: In the IRAF kernel code the FPE exception flag was being set using a local flag rather than a pointer to the flag. The second issue is that between 10.3.9 and 10.4 this flag went from being an "unsigned int" to a "long" and may explain why Colin saw the error on a MacIntel system but not generally elsewhere.

In any case, I've uploaded patch MacPPC binaries to

http://iraf.net/pub/fitz/ib.macx.ppc.gz
http://iraf.net/pub/fitz/nb.macx.ppc.gz

for you to try on your 10.3.9 system. I did only some quick tests but these appear to fix the problem. The same change will need to be made for the Mac/Intel binaries -- I'll wait to decide whether things are close enough for a Beta-3 release to put it off a few days or simply update the Beta-2 binaries with this (fairly critical) bug fix.

Thanks for the report, let me know if things are fixed.

Cheers,
-Mike

Emilliano Gregori wrote on Jun 22, 2006

oops! I can't download the 2 updated packages:

http://iraf.net/pub/fitz/ib.macx.ppc.gz
http://iraf.net/pub/fitz/nb.macx.ppc.gz

I get a "404 Not Found" error...

Cheers,
Emiliano

Mike Fitzpatrick wrote on Jun 22, 2006

My bad, the directory is http://iraf.net/ftp/pub/fitz/

Emilliano Gregori wrote on Jun 22, 2006

Mike,

replaced the binaries, and run install script again.
Now it works great :D :D :D

Many many many thanks!
Cheers,
Emiliano.

mat wrote on Jun 22, 2006

Hi Mike,

I don't know if you are still following this thread.

I'm using Mac Version 10.4.8 and I'm fairly sure I'm having the same issues.

My graphics display, thanks to splot, is happy to go with any ushort images, but gives me a bus error when I use any real images.

Are the binaries that you've included here just for 10.3? I've tried them for my system and don't seem to solve the problem.

Thanks

Mat

Mike Fitzpatrick wrote on Jun 22, 2006

Mat,

I'm also using 10.4.8 (macintel) and just tried this with no error, the V2.13 binaries were in fact built on a 10.4 system. The bus error may be data related. Was the message in response to some particular command or right away at startup? Is it only this data or does it happen with test data as well? Is this a ppc or intel system?

-Mike

mat wrote on Jun 22, 2006

Hi Mike,

I'm on a ppc system. The error came when I used splot. It works fine when I'm splotting ushort files (raw data in this case), but when I try some files that are real integers (reduced data here) it gives me the error. They work without any issues on ds9. I know the files are fine as I can splot them on other systems.

I'm going to try wiping my version of iraf completely and starting again. I originally installed it with the binaries on the download page, got this error, then found this thread and tried it with the binaries you posted. I'm guessing there's still some remnant of my previous install that's causing the problem.

I'll let you know how it goes

Mat

Mike Fitzpatrick wrote on Jun 22, 2006

The patched binaries should fix the FPE issue. To rule out problems with the real pixels you might try using CHPIXTYPE on the ushort images to convert them to real to see if SPLOT still works. If it's really only the reduced data please post a sample image and relevent details.

-Mike

Tom Greene wrote on Jun 22, 2006

I also had problems today with IRAF tasks bombing with bus errors
on floating point files when I installed the macosx nb and ib files
from http://iraf.net/ftp/iraf/V2.13-BETA/ (dated 19 Apr 2006).
V2.12.2 had been working fine on my PPC Mac running OS X 10.4.9.

I followed the tip in this thread and installed the v2.13 ib and nb
files from http://iraf.net/ftp/pub/fitz/ and the IRAF tasks worked
fine on floating point images and spectra again - thanks for the
support!

However I do have another problem that may be related. I
installed the external IRAF package rvsao 2.5.7 (see
http://tdc-www.harvard.edu/iraf/rvsao/),
and it seemed to build fine, but now rvsao tasks die with bus errors
when I run them on floating point spectra. For example, xcsao exits
with a bus error on my PPC Mac, but it works fine on my Mac Intel
laptop (also running IRAF V2.13-BETA and rvsao-2.5.7.

Is this a problem with IRAF or the external rvsao package? If it
is in RVSAO, could you please pass any tips on to Doug Mink?

Thanks & Cheers,

-Tom Greene

Mike Fitzpatrick wrote on Jun 22, 2006

Hi Tom,

If it's the same problem then most likely explanation is that the bug fix is in the libos.a library in the iraf kernel, but you linked against a version with the OLD libos.a. It appears the AS distribution on the site doesn't include the new version, so I just built the RVSAO v2.5.7 on a system with the fix and put the binaries at http://iraf.net/ftp/iraf/extern for you. Try using these binaries and post back if you still have problems.

Cheers,
-Mike

Tom Greene wrote on Jun 22, 2006


If it's the same problem then most likely explanation is that the bug fix is in the libos.a library in the iraf kernel, but you linked against a version with the OLD libos.a. It appears the AS distribution on the site doesn't include the new version, so I just built the RVSAO v2.5.7 on a system with the fix and put the binaries at http://iraf.net/ftp/iraf/extern for you. Try using these binaries and post back if you still have problems.


Mike,

Thanks very much - the new rvsao binaries did the trick.
What excellent support - on a weekend even!

-Tom

rfinger wrote on Jun 22, 2006

Hi Mike,

My colleagues and I are running IRAF from an Apple XServe with OS X and we are having the same bus error problems mentioned in this thread. Our sys admin attempted to install your patch, but it didn't help, and he says that he had trouble figuring out where the patch files go (he is new at this :wink: )

Would it be possible for you to provide us with specific installation instructions for installing the patch?

Thank you hugely!
-r

Mike Fitzpatrick wrote on Jun 22, 2006

This thread got a bit blended so I'm not quite sure which patch you mean: The files in the http://iraf.net/ftp/pub/fitz directory are PPC binaries for 10.3 systems that patch a bug in the FPE handler. These new binaries you would put in the /iraf/irafbin/bin.macosx (IB) and /iraf/irafbin/noao.bin.macosx (NB) directories respectively. There's also a patch to the ECL to workaround the libncurses thing
and I could give you a patched libos.a if you're compiling something locally.

Is this a 10.3 or 10.4 system? Intel or PPC? Which specific patch are you after?

Cheers,
-Mike

sirmarcos wrote on Jun 22, 2006

This also fixes some sort of bug is displaying certain fits images in Tiger. I had a real fits file that I could use display on in the Intel build of OS X iraf, but not in the PPC. Using the replacement binaries removed the "bus error."

I'm incorporating the fixed binaries into my .pkg installer of IRAF 2.13 BETA 2 for PPC Macs.

Last post on Jun 22, 2006