jitter has had many problems with noisy frames, and empty fields too. The main issue was to increase detectability to find some objects worth cross-correlating on. Some improvements have been brought to enhance detection a lot, namely:

• The bright object detection is now performed on the difference between the first two raw frames. By making such a difference, it is easy to get rid of dark current effects, sky background, and so on. This yields much better results than lauching detection on the first sky-subtracted frame.
• Mean and sigma estimations for bright object detections are now done through the computation of the median value, and the average absolute deviation from the median. This is much more robust that way.
• Strategies have been implemented to try all possible combinations to find at least one suitable object. jitter should now try at least 6 different ways of finding objects in the reference frame before giving up (and using the image center or the maximum pixel value).

The configuration file for jitter has been reorganized in sections corresponding to the running process, to make it easier to look for configuration values. It is not really a change in syntax, but rather new sections and new section ordering. You will probably have to re-generate all of your jitter.ini files with jitter -g.

There was a bug in the initial 3.1 version (not the consecutive patched ones) which made it impossible to work on rectangle images. It has been corrected since patchlevel 1.

Floating-point values in FITS headers which are actually describing an integer number were correctly read as floating-point but written back as integers. This may have caused problems with many FITS readers which might then not have identified the keyword correctly. Example:

    Input:
    EPOCH = 2000.0  / Epoch

    Output:
    EPOCH = 2000    / Epoch
	

The warping command suffered from a bug when using the -t or --linear option. When giving a transformation through this option, which is not a strict combination of rotation, translation and scale, warping was not able to compute the size of the output image. This fix solves this problem for very simple linear transforms which are not such combinations, but something more generic will be included in a future eclipse release.

An obvious bug, because it prevented compilation on some platforms... A misuse of the vararg macros leads cc on DEC-Alpha machines to stop compilation in the file pipes/messages.c.