********************************
* New style since January 2012 *
********************************

- In FieldCopier and FieldMultiInstrumentCopier, provide a way to store
  index, pos, start, etc. for the new one. Maybe with some 'magic' in
  read_granule or the actual readers?
- Get new-fields when reading long period? Or force to store first?
- For FieldCopier, be smart in long_name if possible
- single 'CPR' super-dataset that knows in what actual datasets fields are
  and reads the fields accordingly. Maybe in the same way can implement
  different associated datasets as super-dataset?
- Add some kind of checksum to check that read collocs are consistent with
  collocs from the object as it was stored, a hash of some kind?
- in CollocatedDataset.read, more advanced info-passing for associated-datasets where
  fields to read are from, e.g. full structure with info per
  field rather than cell array limiting associated-datasets to
  check in the first place
- For a Collapser, access to the AssociatedDatasets could be useful even
  for global limitators?
- the CachedData objects should also check if the ASsociatedDatasets are
  the same

- Port to new NetCDF interface

- Unify strrep_variables and pos2re, also make more flexible, redesign?

Memory issue: sometimes too many collocations with too many fields, all in
double, needs solution.

Fundamental solution: split over several files per day
Postponing the problem: more memory-efficient
    -> don't always double
    -> don't duplicate/copy data

- More portable way of collocating collocated datasets: get the output
  from CollocatedDataset.read(...) for one day, put this in a structure as
  expected by .reader(...), then make this a dynamic dataset

- when finished, in old-style add small functions using new-style

- handle properly when sibling is missing but core exists
- write proper error message if duration is missing
- write proper error message if input fieldcopier is bollocks
- default granule duration?

Fix:
    - mhs/amsub
    - bugs
    
BUGS:
    - if the FieldCopier no prim. fields are asked, but prim. sibling
      are, failing to read primary therefore data1.lat empty
    - Why is cmoiwp starting time sometimes a minute later than the
      corresponding MHS starting time?

*********************************
* Old style before January 2012 *
*********************************

DOCUMENTATION:

- It appears the distinction between C_I/C_START is not entirely clear.
  Document more clearly 1) how to get to the right datafile 2) how to get
  to the right profile within this datafile

BUGS

- Sometimes, MSPPS starts the minute after the corresponding
  MHS/AMSUB-granule. In the current implementation, near the end of
  colloc_process_data_cpr_mhs, this is not accounted for. When this happens,
  the MSPPS data is missed.
- Bug-checking: sometimes H_INT and B_INT are both within 900 seconds, but
  H_TIME-B_TIME is ~one day; error in hirs time calculation? Or artefact of
  something from a long time ago? Can't reproduce.
- HIRSCS, use scanline numbers for double-checking, rather than unique(...,
  rows), that may be incomplete. At least check whether the latter is or isn't
  incomplete.
- In the collocations, I often add one second to the starting time so that it
  doesn't get truncated to the wrong minute. This is stupid, I should have
  rounded in the first place, and this should be corrected *in the
  collocations*.

WARTS
- Currently, when processing CloudSat/MHS-collocations, I read the files for
  AMSUA and HIRS twice: once when getting collocation info, once when getting
  collocation data. This is not good.
- In both collocate_date and collocate_granule, I predefine 'overlap', 'data'
  and 'meandata' as empty matrices (0x0). This should actually be 0xwidth, so
  that data as stored in the collocation files can always be vertically
  concatenated, and I can get rid of the 'isempty'-test in collocation_read.
- Optimise reading: collocate_date simply loops through every granule and does
  collocate_granule, that means if two granules from sensor A collocate with
  the same granule from sensor B (which they usually do), this one is read
  twice. Bad, particularly for AVHRR where reading takes a long time.
- When collocation poes/mhs or poes/amsub, mhs are amsub are really aliases.
  This is now accomodated by a special case if-statement in collocate_granule.
  This is ugly and needs to go.
- In colloc_read_collocation_cpr_poes.m actually the meandata is read, just
  because that's what I want when collocating with AVHRR; such a special case
  is ugly and needs to go.
- In colloc_process_data_cpr_mhs, move MSPPS reading to a specialised reading
  function (add to colloc_config and all that) and use read_granule to read.

WISHES
- Add aliases for abbreviated satellite pairs, so that noaa15/N15 are the same
  (particularly useful for hirscs).
- Don't let CPR collocate with more than one MHS; only the closest one?
- Don't let AVHRR collocate with more than one MHS; only the closest one?
- Add many, many sat/sensor pairs to the firstline-database and to
  read_granule.
- Generalise the firstline-database. Add to atmlab-data?
- (From Salomon, Ajil) When searching for collocations, permit the ability to
  pass on fields to return on-the-fly.
- Add an extensive testing suite.