WSClean version 3.1 =================== Released 2022-04-01 New/changed dependencies: * To make use of IDG, WSClean 3.1 requires `IDG version 1.0 `_ or newer. * To make use of EveryBeam, WSClean 3.1 requires `EveryBeam version 0.3.1 `_. Summary: :doc:`Facet-based imaging <../facet_based_imaging>` is significantly improved, and now supports applying per-facet solutions (from H5Parm files) and beam corrections (from EveryBeam). Stokes-I-only imaging with IDG imaging is now significantly faster. WSClean 3.1 also adds experimental support for GMRT primary beam correction. Major & visible changes ----------------------- * Support time-direction :ref:`baseline-dependently averaged data from DP3 `. * Reduce disk-access during :doc:`IDG <../image_domain_gridding>` Stokes I imaging by not storing XY/YX to disk, and enable the faster IDG Stokes I mode when possible. * Automatically use the primary beam corrected image (``...-pb.fits``) file when predicting with beam correction. * Cache average beam to disk in IDG mode to reduce memory usage. * Add support for the GMRT beam (scientific validation still in progress). * Add a beam-normalisation-mode option to control EveryBeam's normalisation mode. * Support multiple h5 files in facetting mode. * Introduce ``-pb-grid-size`` and deprecate ``-pb-undersampling``. Bug fixes --------- * Report error when user asks for incorrect channel selection. * Fix certain selection parameters that no longer worked in reorder mode (e.g. even/odd timestep selection). * Correctly normalize reported flux levels for squared channels. * Fix a crash with the ``-reuse-dirty`` option in certain modes. * Fix segmentation fault when using ``-reuse-psf`` with IDG. * Fix masked parallel deconvolution. * Fix beam + H5 solution application in facetting mode. * Fix possible ``-facet-regions`` mismatch with H5Parms. * Support differential lofar beam in facet-based imaging. * Support scalar gain corrections in facet-based imaging. * Do not apply beam or solutions to PSF in facet-based imaging. * Fix unset threadcount value in imageweights class. * Fix IUWT. Because of code clean-ups in Nov 2020, the IUWT deconvolution option was no longer producing correct results. * Fix channel selection issue for non-uniform multi-SPW sets (#79). * Fix channel selection issue for multi-SPW set with no-mf-weighting (#105). * Fix rare crash caused by uninitialized array. * Fix compilation on hurd-i386. Other significant code changes ------------------------------ * WSClean now requires a compiler that supports C++17. * Add support for gcc versions 11 and 12. * Lots of refactoring. * Preparations to split off deconvolution code as a separate reusable module. * Several documentation improvements / updates. * Better support and error reporting for missing git submodules. * Many new unit and system tests.