IO Issues

Most of the IO issues are discussed elsewhere in this document. Data flows from a data source to an SDRecord through an SDIterator and data flows in the other direction as well, when the data source allows it.

• Single Dish FITS - there are two possible paths here. An SD-FITS file can be converted to and from a MS or an SDIterator can read an SD-FITS file directly. Obviously the first path is required but we believe that users will also want the second path, especially on output. Certainly for sufficiently large data sets the conversion step would be extremely time consuming and there might simply not be enough data for two copies of the entire data set. This requires random access FITS classes which can be used to build a FITS storage manager so that an SD-FIT file can appear as a MS.
• Aips++ Image - certainly all images can be viewed as a collection of SDRecords where the x-axis is any of the axes in the image. It should be possible to do just that. It should also eventually be possible to construct an SDIterator that makes slices through an image at an arbitrary orientation to the axes of the image.
• ASCII output - users will want to write out SDRecords into ASCII files. If they want this as an output format, then they will need to be able to convert these into a MeasurementSet or SD-FITS to be used as an iterator. Since this type of data set is very likely to be small, it is not unreasonable that a conversion be done if this is to be used as a data source.
• UniPOPS SDD format - this is essentially the same issue as SD-FITS. We must support the path through a MeasurementSet but users will appreciate a more direct path. We currently have a storage manager for SDD for the old, flat, MeasurementSet so in principle a MeasurementSet could be constructed from an SDD file.
• For most (all?) of the above, we need to be able to write to an existing file (append only in some cases or replace if possible).