Changes in version 2

Version 2 of pynbody is the first ever release to break full backwards compatibility. Most code for pynbody 1.6 will also work with pynbody 2.0, but there are some changes that may require you to modify code. This document summarises the most important changes.

Rationale

The changes in pynbody 2.0 are designed to make the codebase more maintainable. The essential principle was to remove rarely-used code that was adding complication – to do less, but do it better. Breaking changes are never pleasant, but in its 15-year history, pynbody has never had a major version increment like this. We hope that the changes will make the codebase more sustainable in the long term.

Key changes

• By far the biggest improvement is the documentation, including tutorials; these have been completely overhauled and updated.

• The halo subpackage has been completely re-implemented, for greater consistency across formats. For many simple uses, old code will continue to work. However, please read the change notice in halo for important changes. This especially affects certain AHF catalogues which used to be given an inconsistent halo numbering scheme. Backwards compatibility can be achieved there by passing the correct flag. There is also now a format-independent way to describe parent/child relationships in catalogs, which is described in the Halos in Pynbody tutorial.

• The plot subpackage has been streamlined, with a number of trivial routines removed, and options to save files etc removed. These are much better handled by matplotlib directly. See the plot documentation for details.

• The analysis package has been rationalised, with unnecessary routines removed (including the pkdgrav_cosmo module). Major improvements have been made to luminosity and ionfrac to bring their estimates up to date and explain more clearly how to introduce new tables for interpolation. The profile module has been improved. The decomp module has been renamed to morphology.

• The sph subpackage has been overhauled to improve efficiency and robustness of SPH operations. The implementation of healpix rendering has been improved, by using our own healpix pixelization implementation that enables faster, parallel rendering. KD-tree operations are now found in kdtree and KD-tree information can be accessed directly from python as a numpy record array (e.g. for saving or manipulating KD-trees directly).

• The array subpackage now publishes a number of functions that were previously hidden, specifically enabling the use of shared memory in complex analysis pipelines across multiple processors. For more information see Writing your own parallel code.

• The implementation of bridges, and particularly halo matching, has been improved. Methods match_catalog() and fuzzy_match_catalog() have been deprecated in favour of more robust alternatives match_halos() and fuzzy_match_halos().

• There is improved support for various flavours of GadgetHDF files, including Swift, Arepo and TNG.

• The transformation subpackage has been overhauled, with improved consistency e.g. if one loads data after a transformation has been applied (it will now be correctly transformed).

• Test data is now stored as a series of separate tars on Zenodo, rather than as a single large tarball. Accessing test data can be accomplished automatically via the pynbody.test_utils function. When running tests, the required data will be downloaded automatically if needed.

• A number of external dependencies have changed; this should be taken care of automatically by pip. Numpy 2.0 is now supported.