Branch: refs/heads/master
  Home:   https://github.com/Starlink/starlink
  Commit: 733c47147dca2363fe77862cacd24e6b76ba97f8
      https://github.com/Starlink/starlink/commit/733c47147dca2363fe77862cacd24e6b76ba97f8
  Author: David Berry <[log in to unmask]>
  Date:   2014-03-31 (Mon, 31 Mar 2014)

  Changed paths:
    M applications/smurf/libsmf/smf_iteratemap.c

  Log Message:
  -----------
  smurf: Prevent mapchange being dominated by bad edge pixels

Pixels round the edge of the map may contain very few samples, and so
their variances can be widely inaccurate. So exclude pixels with low hits
from the estimate of the normalised map change (which is the basis of the
convergence test). "low hits" is anything under 10% of the mean hits per
pixel across the map.


  Commit: 3a26bd66c98e6054f1288f2a240251a013dbb809
      https://github.com/Starlink/starlink/commit/3a26bd66c98e6054f1288f2a240251a013dbb809
  Author: David Berry <[log in to unmask]>
  Date:   2014-03-31 (Mon, 31 Mar 2014)

  Changed paths:
    M applications/smurf/libsmf/smf_jsatile.c

  Log Message:
  -----------
  smurf: Correct pixel bounds for JSA tiles in the split HPX facet

The split facet uses RA=12H as the reference point, and so the pixel
bounds of tiles in this facet need to be corrected to refer to the
expected NDF origin at RA=0h.


  Commit: 9f7306d73d5f24f2b33c6bcecc0f742d79236059
      https://github.com/Starlink/starlink/commit/9f7306d73d5f24f2b33c6bcecc0f742d79236059
  Author: David Berry <[log in to unmask]>
  Date:   2014-03-31 (Mon, 31 Mar 2014)

  Changed paths:
    M applications/smurf/libsmf/Makefile.am
    M applications/smurf/libsmf/smf.h.source
    M applications/smurf/libsmf/smf_mapbounds.c
    A applications/smurf/libsmf/smf_offsets.c
    M applications/smurf/libsmurf/smurf_makemap.c

  Log Message:
  -----------
  smurf: Correct mapbounds fast-mode so that it can be used for JSA maps

The previous version assumed that north was upwards, which it isn't for
JSA. For instance, try

% makemap in=/star/share/smurf/s4a\* crota=45 \
    config=/star/share/smurf/dimmconfig.lis

The issue was that the extrema time slices were found by looking at the
bounding box in (ra,dec) offset, but if the pixel axes are not aligned
with (Ra,Dec), then that's not the right box to use. Now, the boresight
positions are transformed into output map pixel coords, and the bounding
box in pixel coords is used to find the extrema time slices.

In addition, there may be some time gain due to the fact that the new
scheme uses astTran2 to transform all time slice positions simultaneously,
rather than making a separate call to astTran2 for each time slice.


Compare: https://github.com/Starlink/starlink/compare/0761d1556592...9f7306d73d5f