Changelog

All notable changes to OpenTIMS: the C++ core (opentims++), the Python package (opentimspy) and the R package (opentimsr), which share one version number. R versions carry no pre-release suffix, so 1.2.0b4 was opentimsr 1.2.0.

Each release’s section becomes the notes of its GitHub release, and src/opentimsr/NEWS.md points here, so R’s news() shows it too.

1.2.3 (2026-11-03)

Core

• The shared C++ library installs NOTICE (licences of the bundled SQLite, Zstandard and mio) next to its licence.

Python

• Package metadata gives the licence as the SPDX expression MIT (PEP 639) instead of the full licence text; LICENCE and the new NOTICE (licences of the bundled SQLite, Zstandard and mio) ship in the wheel. Building from source needs scikit-build-core 0.11 or newer.

R

• OpenTIMS() on a folder without analysis.tdf stops with “No analysis.tdf in ” before opening a database connection, instead of RSQLite’s “Could not connect to database” (which also leaked memory; CRAN’s valgrind check would report it).

1.2.2 (2026-10-25)

Core

• Extraction decodes frames in parallel, and decodes a frame requested several times only once.
• Optional lookup tables for Bruker’s m/z and 1/K0 conversion, much faster when converting many peaks; calls into Bruker’s library are serialised.
• Compressed frame headers are read without unaligned loads (undefined behaviour reported by CRAN’s sanitizer checks).
• Errors from parallel decoding keep their message when R loads both libc++ and libstdc++ (CRAN’s fedora-clang check reported “c++ exception (unknown reason)”).
• Repeated frames are extracted correctly in every column, and zstd flag overrides are honoured.
• The public header no longer includes SQLite, so consumers of the shared library need no SQLite headers.

Python

• get_separate_frames() decodes each frame once.
• query() and frame_arrays() reject fractional, negative and out-of-range frame ids instead of truncating them.
• SQLite connections are closed, framesTIC is cached per instance, and invalid input raises exceptions instead of failing asserts.
• Docstrings of the conversion and retention-time functions describe what the code does.
• Requires Python 3.9 or later; pybind11 is no longer a runtime dependency.

R

• Back on CRAN: the README installs it with install.packages("opentimsr").
• New get_separate_frames().
• Frames are decoded straight into R vectors.
• download_bruker_proprietary_code() reuses a library already downloaded (new argument overwrite) and never rewrites one in place: downloading it a second time in a session no longer crashes R (#21).
• Addressed CRAN’s review of 2026-10-06: Title and Description wording, the length(), range() and [ methods are exported, and examples that download Bruker’s library use \donttest{} and write to tempdir().

1.2.1 (2026-10-03)

Core

• Frame extraction is thread-safe; cached frames no longer alias one buffer, and failed extraction joins its workers and restores the threading state.
• Frame ids are no longer assumed to run 1..N.
• Hardened against corrupted data: decompressed sizes are checked before allocation, unexpected dataset contents raise exceptions instead of failing asserts, and slice counters cannot wrap.
• Locale-independent number parsing, without side effects on the global locale.
• Frame counts are parsed as 64-bit on Windows; dlopen’ed SQLite symbols are resolved thread-safely; an SQLite busy timeout lets concurrent readers open the same dataset.
• Corrected intensities are clamped into the uint32 range.
• Conversion failures raise exceptions with Bruker’s diagnostics.
• A C++ test suite, run in CI with sanitizers, Valgrind and fuzzing.

Python

• Frame owners are kept alive while frame references exist.
• query() with _sanitize=False fills user-provided arrays.

R

• Links the system zstd and no longer bundles zstd or SQLite; SQLite access goes through RSQLite.
• Mateusz Łącki is the maintainer.
• Frame numbers and slices are validated with clear errors; range() with by = 0 no longer loops forever.
• OpenTIMS() expands ~ and relative paths, and opens analysis.tdf read-only.
• download_bruker_proprietary_code() stops clearly on platforms without Bruker’s library.
• Released on GitHub as a source tarball and a Windows binary.

1.2.0 (2026-04-22)

• Built-in open-source m/z and 1/K0 converters, selectable per dataset, as an alternative to Bruker’s library.
• Reworked build system: static and shared C++ libraries with CMake and pkg-config files; SQLite loaded from Python’s own module or linked.
• Reworked R bindings with tests.
• Datasets whose frame ids do not start at 1 are supported.
• Asserts replaced by exceptions.

Earlier releases are described only by the git history.