Fixed

• Fixed invalid data type when gsd.hoomd wrote */group data chunks (#541).

Added:

• gsd.hoomd schema now supports float64 and float32 values. Choose the desired precision with the precision argument to open (#495).

Changed:

• Frame.validate no longer modifies its contents in place by default. Pass inplace=True to recover the previous behavior (#495).
• New files created by gsd 5.0 are schema version 2.0 and may not be readable by older software (#495).

Removed:

• SIGTERM handler (#501).
• Removed state/* data chunks from the hoomd schema (#532).

Fixed:

• gsd.hoomd.read_log no longer iterates over the file more than necessary (#459).
• Debug log entries are no longer duplicated (#459).

Added:

• gsd.hoomd.read_log now accepts glob_pattern to restrict the returned data (#459).
• debug attribute for gsd.fl.GSDFile, which toggles debug logging for performance-sensitive methods (#459).
• Provide linux aarch64 binary wheels on PyPI.

• Build with abi3 for Python >=3.11 (#459).
• Remove support for Python <3.11 following NEP29 (#459).

Fixed:

• Type hints for HoomdTrajectory._read_frame and downstream methods are now correct (#417).
• Sync changes to disk on macOS (#429).
• Sync index entries to disk before returning from flush (#429).
• Work around macOS bug that mysteriously allocated 16 MB of extra disk blocks beyond the end of the file (#429).

Changed:

• No longer test with clang 10, 11, or 12 (#422).
• [breaking] end_frame, find_chunk, find_matching_chunk_names, and read_chunk no longer implicitly call flush. The caller must manually call flush (or close the file) before chunks can be read. This behavior extends to the high level gsd.hoomd Python API (#429).
• Set the default write buffer to 1 MB (#429).

Removed:

• index_entries_to_buffer (#429).

Fixed:

• Make NumPy 2.0 requirement optional (#405).

Fixed:

• Prevent downgrading file versions when opening in write mode.

Added:

• GSD_TYPE_CHARACTER chunk type for string data - valid in file layer versions 2.1 and later (#391).

Changed:

• The log dictionary in a hoomd frame now accepts string values. read_log correspondingly produces a NumPy array of variable length strings (#391).
• Require NumPy >= 2.0 (#391).

Added:

• Support Python 3.13 (#388).

Removed:

• Support for Python 3.9 (#388).

Changed:

• Improved build instructions (#376).