flint -gtxs -Sabc *.fwhere it is assumed that the default directory contains all FORTRAN source files involved. This command line causes files to be created in the directory with the names of the form "abc.tre", "abc.xrf", "abc.stt", and "abc.lnt" for calling tree, cross-reference, statistics, and lint-style error checking. By using *.f with all source files in the directory, flint is able to check for consistency of calling lists and COMMON blocks across all routines involved. The files generated by flint can be inspected with a text editor.
When deblending produces different results in different bands, fragments of point sources are generally obtained which are neither complete nor reliable representations of the sources involved. Some standard fixup for this situation is very desirable, but it is not clear what this should be. Clearly, recourse to the original raw frame data would allow optimal extractions to be performed, but multi-passing the raw data or processing all bands simultaneously in PROPhot are not even close to being viable options because of hardware limitations and the relatively infrequent occurrence of deblending processing.
If possible, it would be desirable to control the deblending algorithm so that consistent results are obtained in all bands. This does not appear to be possible, however, since any thresholds will always have events occurring randomly above or below them on different occasions, and the geometrical details of flux falling on pixels will not be the same in all bands (the varying plate scale alone will cause this). Even eliminating the deblending code would not prevent sources from being resolved in some bands and combined in others.
It was suggested that whatever corrective action is to be applied be implemented in BANDMERGE, since that is the first time that point source extractions in different bands are brought together. BANDMERGE will know when a single source in one band has close multiple matches in another, and PROPhot provides a blending flag that tags the products of this processing. BANDMERGE could theoretically recombine separate pieces to match single objects in other bands (applying some splitting algorithm to break up single sources to match multiple objects in another band does not appear feasible). Another action could be to bandmerge the best fragment, write the others to a special file, and set a pointer in the bandmerged output source record to indicate where to find the additional pieces (a nonzero pointer would also serve to flag the condition). A variation of this idea would be to output a recombined entry and add pointers to all the original separate fragments (actually, a single pointer to an entry in the "fragment file" would suffice, with the other entries accessible via a linked list, with the last entry indicated by a zero pointer); this would allow users to ignore the fragment file easily if they wished to do so, leaving them with the best available recombined entries.
Another possibility is for GALWORKS to apply templates to the coadded images at indicated locations to extract multiple sources where a single one was produced by PROPhot in the given band but multiple objects were found in another band. A problem with this is that it does not match identically what PROPhot does to extract sources, since PROPhot uses the single frames and corresponding PSF templates. Such templates would not be appropriate for use in the coadded images. For GALWORKS to do anything at all is suggested because of the fact that band filling is to be done in this subsystem.
No decision can be made without science team input, but the discussion served to define the problem and several possible approaches to its solution. In the meantime, the BANDMERGE SDS will contain a lien stating that it does not attempt any corrective action for cases of this sort. The lien will give visibility to the problem until such time as a decision is made by the project about how to handle it.