[fitsbits] Error in the FITS Standard document {External}

[William Pence]

Andreas Korpi-Lagg (MPI) and some of the astropy developers recently alerted me to an apparent incorrect statement in the current FITS Standard document. This occurs in section 10.1.3 of that document in the discussion about compressed arrays:

“The compressed data columns described above may use either the ‘1P’ or ‘1Q’ variable-length array FITS column format if the size of the heap in the compressed FITS file is < 2.1 GB. If the heap is larger than 2.1 GB, then the ‘1Q’ format (which uses 64-bit pointers) must be used.”

That second sentence is not correct  because it is possible to create a legal FITS file where the heap is larger than 2.1 GB using the ‘1P’ column format, for instance if the sum of the 2 pointer values in the P descriptor is greater than 2^31. One could also create a FITS file with an arbitrarily large heap containing extra space that is not used by the P variable-length array column. 

My suggestion is to replace those 2 sentences in the FITS Standard with the following text:

“The compressed data columns described above may use either the ‘1P’ or ‘1Q’ variable-length array FITS column format if the 2 integers in the descriptor have values that are within the range supported by the 32-bit pointers in the ‘1P’ column format.  Otherwise, the ‘1Q’ format which uses 64-bit pointers must be used.”

FWIW, I would recommend always using the 1Q format for variable length arrays.  Then you don’t have to worry about whether the array is too large to fit, and the savings in file size by using the 1P format is usually negligible. 

Bill Pence
NASA/HEASARC Emeritus