WSClean version 2.6
===================

Released 2018-06-11

Summary: A "lots of little improvements" update, with no major changes at this time. Interesting new features are some very specific cleaning modes, like linked polarizations and parallel cleaning.

Major & visible changes:

* Support for "parallell cleaning" (option '``-parallel-clean``'), which splits up images using a path search algorithm, and allows faster and clean and larger images.
* Polarization-linked cleaning ('``-link-polarizations``'), which allows cleaning a set polarizations using information from another set of polarizations (e.g. clean QUV only using I information).
* Support for calculating AARTFAAC primary beam.
* Tapering is now taken into account in the initial value of the PSF, thereby improving fit stability on tapered runs (SF #134, reported by P. Serra)
* Circular and non-circular Gaussian fitter improvements (related to SF #134).
* Added option '``-beam-fitting-size``' to allow changing the fitting box size (useful for MSSS).
* Added option '``-divide-channels-by-gaps``', which changes how bandwidth is splitted in multiple-output-channels mode.
* Adding option '``-pb-undersampling``' to turn off or change undersampling.
* Added parameter '``-lofar-centroids``', which takes centroid information into account. This was an experiment, and seemed not to be very useful (it is very slow and does not improve accuracy).
* WSClean can now grid only the even or odd timesteps (parameters '``-even-timesteps``' and '``-odd-timesteps``'). Useful for certain experiments, e.g. estimating noise value (SF #133).
* Option '``-no-normalize-for-weighting``' was removed, as keywords can be used to scale images appropriately.
* Accept 3d fits masks that have velocity instead of frequency as 3rd axis (reported by M. Ramatsoku).
* Provide an error message when automasking threshold <= auto threshold.
* Removing ``-force-dynamic-join`` option, as it is no longer necessary to specify it.
* IDG command line parameters have been added to the command line help.

Bugfixes / technical changes:

* CMake findcasacore script updated to fix unresolved symbols with certain Casacore versions.
* Model column was not updated when using '``-continue``' (reported by R. van Weeren and G. Di Gennaro).
* One-off error in calculating of nr scans, and reporting proper error when requesting more intervals than available (SF #131, reported by S. Bourke)
* When writing model data back to MS, progress bar did not actually show progress.
* Threading and mutexes are now done with (C++11) the stdlib. This decreases the dependency on Boosts.
* Fixed gcc 7.3 compilation: `pow10 was removed from gcc 7.3 (glibc 2.27) <https://abi-laboratory.pro/tracker/changelog/glibc/2.27/log.html>`_.
* Fixed unit weighting to not include flagged/unavailable visibilities during imaging when used.
* Preparations for IDG A-term correction: no longer store visibilities preweighted, pass full weighting to IDG.
* Do output errors in quiet mode.
* Don't allow rectangular images when IDG is enabled.
* Improved readability of non-verbose mode a bit; be less verbose for sets with denormal phase centre and removed some other technical output.
* Reported nr of major iterations was one too high (on cmdline as well as in fits hdr).
* Invalid floating point cmd line parameters are reported as error instead of interpreting them as zero.
* Independent QU imaging with multiple channels would not clean U: weight was not initialized (reported by R. van Weeren).
* Also fixing bad keyword 'WSCMINOR' in dirty image of second polarization.
* Lofar beam correction with only-I imaging causes segfault (Identified by A. Shulevski).