From: <mcalabre_at_email.protected>
Date: Thu, 28 Jul 2005 11:09:37 +1000

<pre id="body">
<a name="start" accesskey="j" id="start"></a>Greetings,
The FITS onslaught continues! However, the current proposals, relating
to handling 64-bit integers in FITS, should be light work compared to
WCS Paper III.
The proposals have been split into four separate votes by Bill Pence who
writes
    Some general background information about the issue
    of supporting 64-bit integers in FITS is available at
    http://fits.gsfc.nasa.gov/fits_64bit.html. In addition, a PDF
    file is available from that web page which shows in detail the 11
    specific modifications to the wording of the FITS Standard that
    are now being considered.
    Please ask your committee members to vote separately on the
    following 4 specific issues to provide as much guidance as possible
    to the IAU-FWG. In each case, members may vote either "Yes",
    "No", or "Abstain - no opinion".
    1) Do you favor adding support to FITS for primary arrays and image
    extensions containing 64-bit integer pixels, with BITPIX = 64?
    (Changes 1 - 3 in the PDF file).
    2) Do you favor adding support to FITS for 64-bit descriptor 'Q'
    columns in binary tables (analogous to the 32-bit 'P' descriptor
    columns). This will support FITS files with a variable length array
    heap much larger than the 2 GB limit of 'P' descriptor columns.
    (Changes 4 - 6 and 8 - 9).
    3) Do you favor adding support to FITS for columns of 64-bit integer
    values in binary tables, with TFORMn = 'K'? (Changes 4, 7, and 10).
    4) Do you support changing the definition of the 'P' descriptor
    pointers from 'signed integers' to 'unsigned integers'. This change
    will not invalidate any existing FITS files and will double the
    maximum allowed size of the variable length heap from ~2 GB to
    ~4 GB. (Change 11).
Note that the URL cited above does not mention the discussion on
fitsbits that occurred in April, May, and June 2005:
http://listmgr.cv.nrao.edu/pipermail/fitsbits/2005-April/date.html
http://listmgr.cv.nrao.edu/pipermail/fitsbits/2005-May/date.html
http://listmgr.cv.nrao.edu/pipermail/fitsbits/2005-June/date.html
Proposals (2) and (3) are no-brainer "yes" votes, noone expressed any
opposition to these. In particular, (3) will be of interest for pulsar
work in recording high-precision timing parameters.
Proposal (1) is moderately contentious. Protagonists argue from the
perspective of FITS as a data format, drawing attention to the fact that
FITS primary image arrays and image extensions can be viewed as special
cases of a binary table. Since proposal (3) allows a 64-bit image to be
stored in a TFORMn = 'K' binary table column it would be inconsistent
not to allow the same for a primary image array or image extension, thus
requiring BITPIX = 64. Antagonists argue that there are no valid
scientific uses for images with 64-bit integer values. The following
from Bill Pence attempts to answer that:
http://listmgr.cv.nrao.edu/pipermail/fitsbits/2005-June/001667.html
Proposal (4) was discussed on fitsbits mainly in June 2005:
http://listmgr.cv.nrao.edu/pipermail/fitsbits/2005-June/date.html
A number of people expressed opposition to it; reading between the
lines, the most compelling reason seems to be that Fortran doesn't
support unsigned integers. Judge for yourself.
Cheers, Mark
Received on 2005-07-28 11:09:59