Integrating the code to choose the output BITPIX (and quantising floating-point images if necessary) and custom floating-point schemes with the functions to read and write images. More detailed description, from my e-mail: psLib has always read any of the (legal) FITS types, and written a FITS image with the appropriate BITPIX for the image->type.type. Now the output BITPIX can be forced by setting psFits.bitpix to the desired BITPIX. Floating point images being quantised to integer types have BSCALE and BZERO set in such a fashion as to maintain the dynamic range of the image; quantisation after applying the scale and zero includes adding a random number [0,1) to preserve the expectation of the floating-point value. On reading, the BSCALE and BZERO are applied by cfitsio without any assistance by psLib. I've also added Gene's custom 16-bit floats (1S 5E 10M). In anticipation of further flavours down the road and to maintain compatibility, I've labelled this as FLOAT_16_0. An F32 image may be written out with this scheme by setting psFits.floatType = PS_FITS_FLOAT_16_0. The image is converted and written out with BITPIX=16, and the header keyword PSBITPIX added with the custom floating-point scheme name. Such images are automatically converted to F32 on reading (if psFits.conventions.psBitpix is true, which it is by default). One thing to keep in mind is that I'm not sure how NANs and INFs are handled in these processes. These additions are made on the top of the compression feature --- quantised or custom floating-point images may also be compressed (though the custom floating-point images don't gain much from it).
