Compatibility

Pymatgen is a tool used for academic research and is actively developed by a large community of people. As such, releases are frequent, and new features and capabilities are constantly being added.

However, pymatgen is also used as a library by other tools, and as such breaking changes such as the removal or renaming of existing functionality, or substantive changes in the output of existing code, are tried to be kept to a minimum. This is especially true of all classes contained in the pymatgen.core module.

Despite this, it is sometimes necessary to make breaking changes to enable future development, or because other libraries we depend upon might change their own requirements. If a breaking change is causing significant issues, please post on the GitHub Issues page to see if it can be resolved.

Depending on Pymatgen

Pymatgen uses calendar versioning based on a YYYY-MM-DD format. This has generally worked well since changes to core pymatgen functionality that most other codes depend on are rare. There have been only two instances in recent memory that breaking changes have been made.

  • v2021.3.4 - Reorganization of pymatgen into namespace packages, which required the removal of root-level imports.

  • v2019.3.13 - Renaming of Site.species_and_occu to Site.species.

Where at all possible, the pymatgen maintainers try to allow for a reasonable deprecation schedule. For exampple, the Site.species change had a deprecation schedule of about 9 months. However, some changes such as the recent reorganization of pymatgen into namespace packages cannot be easily done via a deprecation schedule.

As a compromise solution, pymatgen has adopted temporary semantic versioning. A v2021.3.5 was released after v2021.3.4 to reverse the changes made, and new versions v2022.0.x were released that contains the breaking change (removal of root imports). We will continue to release critical updates to 2021.x.x versions. This allows end users to continue using 2021.x.x versions without having to deal with the breaking changes. However, it is recommended that users make the move to be compatible with 2022.0.x before Jan 1 2022, during which pymatgen will only support the new namespace architecture and the versioning scheme will go back to calendar versioning.

As the developer of a tool that depends on Pymatgen, you can prevent upgrades of the major Pymatgen version by specifying a version range like pymatgen>=2021.1,<2022 or, more succinctly, using the compatible release operator pymatgen~=2021.1. This will prevent pip (and other package managers) from pulling in Pymatgen versions with breaking changes that may end up breaking your tool.

An even more conservative approach is to pin the Pymatgen dependency to a fixed version, for example pymatgen==2021.3.3. While this will always install the the same version of pymatgen, it can lead to unnecessary dependency conflicts with other tools that depend on (a different version of) Pymatgen.

Minimum Python Version

As a rule of thumb, pymatgen will support whatever versions of Python the latest version of numpy supports (at the time of writing, this is Python 3.7+). You can also check what versions of Python are being tested automatically as part of our continuous integration testing on GitHub. We currently test pymatgen on Mac, Windows and Linux.

Recent Breaking Changes

v2022.0.0

Pymatgen root imports have been removed from v2022.0.0 in preparation for a change to a more modular, extensible architecture that will allow more developers to contribute.

Specifically, the following “convenience imports” have been removed in favor of their canonical import:

from pymatgen import Composition  # now "from pymatgen.core.composition import Composition"
from pymatgen import Lattice  # now "from pymatgen.core.lattice import Lattice"
from pymatgen import SymmOp  # now "from pymatgen.core.operations import SymmOp"
from pymatgen import DummySpecie, DummySpecies, Element, Specie, Species  # now "from pymatgen.core.periodic_table ..."
from pymatgen import PeriodicSite, Site  # now "from pymatgen.core.sites ..."
from pymatgen import IMolecule, IStructure, Molecule, Structure  # now "from pymatgen.core.structure ..."
from pymatgen import ArrayWithUnit, FloatWithUnit, Unit  # now "from pymatgen.core.units ..."
from pymatgen import Orbital, Spin  # now "from pymatgen.electronic_structure.core ..."
from pymatgen import MPRester  # now "from pymatgen.ext.matproj ..."

If your existing code uses from pymatgen import <something>, you will need to make modifications.

The easiest way is to use an IDE to run a Search and Replace. First, replace any from pymatgen import MPRester with from pymatgen.ext.matproj import MPRester. Then, replace from pymatgen import with from pymatgen.core import. Alternatively, if you are using a Mac command line, you can try:

find . -name '*.py' | xargs sed -i "" 's/from pymatgen import MPRester/from pymatgen.ext.matproj import MPRester/g'
find . -name '*.py' | xargs sed -i "" 's/from pymatgen import/from pymatgen.core import/g'

From a Linux command line, you can try:

find . -name '*.py' | xargs sed -i 's/from pymatgen import MPRester/from pymatgen.ext.matproj import MPRester/g'
find . -name '*.py' | xargs sed -i 's/from pymatgen import/from pymatgen.core import/g'

This should resolve most import errors and only a few more modifications may need to be done by hand.

v2021.3.3

The variable pymatgen.SETTINGS has been moved to pymatgen.settings.SETTINGS. Since this is mostly used internally within pymatgen, it is not expected to lead to significant external issues.

v2021.2.8.1

The minimum version of Python was increased from 3.6 to 3.7 following the lead of numpy. However, at this point there are no exclusively Python 3.7+ features used in pymatgen so pymatgen may still be able to be installed manually on Python 3.6 systems, although this usage is not supported.

Support for aconvasp has been removed since the corresponding tests were failing and this module was not being maintained.

v2020.10.20

The band structure plotting functionality, BSPlotter, has been overhauled to allow plotting of multiple band structures. This might cause issues for tools relying on the internal structure of BSPlotter’s plot data.