WSClean version 3.2 =================== Released 2022-10-21 New/changed dependencies: * To make use of EveryBeam, WSClean 3.2 requires `EveryBeam version 0.4.0 `_. * The requirements on IDG have not been changed; WSClean 3.2 requires `IDG version 1.0 `_ or newer. Summary: Major & visible changes ----------------------- * Add support for direction dependent PSFs in deconvolution (parameter ``-dd-psf-grid``). * Support heterogenously polarized measurement sets (`#130 `_). * Make use of the low-memory interface of EveryBeam. This reduces the amount of memory required for large image runs and additionally reduces the disk IO substantially. * Include an experimental, potentially faster tuned wgridder implementation. * Automatically determine right amount of memory and threads to use in parallel runs. * The various ``-use-...-gridder`` are replaced by a single ``-gridder `` option. * Support spectral fitting when polarizations are joined. * Better automatic determination of number of cores to be used. * Fix point-scale auto-masking on data with missing channels (`Radler #114 `_). * Calculate and save the MFS beam when relevant. * In facetted imaging, replace NaN solutions by zeros. Bug fixes --------- * Fix a crash when using local rms + multiscale. * Fix a flux-scale issue when applying the primary beam on Stokes I imaging runs (see `!465 `_ and `!441 `_). * Solve crash when compiling with ``_GLIBCXX_ASSERTIONS``. * Fix compilation on Macs. * Allow custom HDF5 directory. * Fix multi-polarization imaging without joining pols (`#128 `_). * Allow facets to be defined outside the FOV, and skip these during imaging. * Fix missing FITS file error when faceting without applying beam. * Improve some error messages. * Make sure exceptions are reported while gridding in parallel. * Fix FITS ``LONPOLE`` issue causing incorrect image coordinates at the pole. * Output correct source lists when using ``-shift`` (`#98 `_). * Fixes for GCC-12. Other significant code changes ------------------------------ * Format python files using black. * Documentation improvements. * Separate deconvolution code into separate library: `Radler `_