StreamingHDU issue

Hi,

I’m running Debian unstable on 64 bit machines (and hence this is version 1.3 of astropy) and have come across an issue with StreamingHDU on 2 out of the 3 machines!

This is the test program:

#!/usr/bin/env python

import numpy as np
import astropy.io.fits as fits

n1 = 2048
n2 = 2048
n3 = 50
hdu = fits.PrimaryHDU()
hdr = hdu.header

hdr['NAXIS'] = (3, 'its a movie')
hdr['NAXIS1'] = (n1, 'number of n1 samples')
hdr['NAXIS2'] = (n2, 'number of n2 samples')
hdr['NAXIS3'] = (n3, 'number of frames in movie')
hdr['BITPIX'] = (-32, 'np.float32')

print hdr
shdu = fits.StreamingHDU('test_shdu.fits', hdr)

data = np.zeros((n1, n2), np.float32)

for f in xrange(n3):
    data[:,:] = f
    print 'Doing frame', f, data.nbytes
    shdu.write(data)

shdu.close()

On two of the machines, it fails with:

Doing frame 25 16777216
Traceback (most recent call last):
  File "./test_stream_fits.py", line 26, in <module>
    shdu.write(data)
  File "/usr/lib/python2.7/dist-packages/astropy/io/fits/hdu/streaming.py", line 174, in write
    raise IOError('Attempt to write more data to the stream than the '
IOError: Attempt to write more data to the stream than the header specified.

If I look at the output (using ds9) when the program fails, then every odd numbered frame is filled with zero, and the even frames are 1,2,3,…24. So it seems to be writing an additional empty image out before it writes the actual data. I suspect its incidental, but the two machines it fails on are both Intel, and the one it does work on is AMD.

Any suggestions, or is it a real bug?

Thanks,

Matt

[saimn]

It seems this was the last bug labeled “Python 2”, and since it works with Python 3 I guess we can close!

[embray]

This has never really been a well supported or tested feature in the first place.