v3.9.2 2024-08-19
     * Prevent rare error in the MiniMC framework, which happened when acos was
       evaluated on a value which due to numerical instabilities was ever so
       slightly larger than one.
     * Minor tweaks to MMC parsing and plotting code.

v3.9.1 2024-08-16
     * Add missing _hklobjects.py file which was left out by mistake. Thus
       making the newly introduced info.hklObjects() function actually work.
     * Fix debug compilation of MMC messages.

v3.9.0 2024-08-12
     * This release brings both general improvements and new features, as well
       as a slew of maintenance updates and bug fixes. The first few items below
       will discuss the main highlights.
     * General speedup in a lot of the code. In particular the Fast Fourier
       Transform code, used when expanding phonon density of states to full
       scattering kernels for inelastic modelling, was improved by as much as
       40%. This will make a noticeable difference when initialising many
       materials, and is an improvement independent from the multi-threading
       option mentioned in the previous item.
     * If requested, NCrystal can now initialise materials even faster, by
       performing the initialisation calculations in multiple threads. This can
       for instance be enabled by setting the environment variable
       NCRYSTAL_FACTORY_THREADS to the number of threads you wish to use (e.g. 8
       if you have 8 CPU cores available, or 9999 to have NCrystal pick a number
       suitable for your machine). Alternatively, you can call functions in the
       Python, C, or C++ APIs to enable this:
         NC.enableFactoryThreads() # Python
         NC::FactoryThreadPool::enable(); //C++
         ncrystal_enable_factory_threadpool(9999); //C
         export NCRYSTAL_FACTORY_THREADS=9999 # BASH
     * NCrystal now includes the first iteration of a "Mini Monte Carlo"
       framework (for lack of a better word), which makes it possible to insert
       an NCrystal material into a simple geometric "sample" shape, subject it
       to a source of neutrons, and obtain a spectrum or pattern of the outgoing
       neutrons. Thus, one can easily produce a simplistic diffraction pattern,
       including effects of multiple scattering and both elastic and inelastic
       physics. For now, the focus has been on developing the framework itself,
       rather than preparing a library of interesting shapes, sources, and
       tallies. Thus, the only readily available setup in this release is that
       of a monochromatic pencil beam impinging on a sphere, tallying the
       outgoing angle of the neutrons as they leave that sphere. Nonetheless,
       the results provides a very convenient view of how NCrystal modelling
       translates into a particular pattern -- without the need for an
       additional software package like McStas, Geant4, or OpenMC. The `nctool`
       command will now by default show such scattering pattern, or they can be
       explicitly requested, as shown in this example where 1.8Aa neutrons
       impinge on an aluminium sphere with a diameter of 2cm:
       $> nctool --mc 1.8Aa 2cm Al_sg225.ncmat
       The scope of this "Mini Monte Carlo" framework is of course not to try to
       rival existing applications in terms of ability to simulate complicated
       geometries or beamlines, but rather to provide another convenient way to
       quickly investigate NCrystal materials, as well as potentially serve as a
       way for NCrystal to be used in a regression-loop while analysing actual
       neutron scattering data. For that reason, speed has been a major concern,
       and both multithreading and vectorisation has been utilised to bring the
       time it takes to go from defined NCMAT data to having a high-statistics
       diffraction pattern available, down to the sub-second scale.
     * Internally NCrystal code no longer uses raw `std::cout` or `printf`
       statements. Rather, all messages are now emitted via a centralised
       infrastructure. This most importantly fixes an issue where output from
       the Python and C++ layers could occasionally be emitted in incorrect
       order - now, when using the NCrystal Python API, all output will be
       emitted through Python's own print buffer. Secondly, several NCrystal
       objects now gets a `.dump_str()` method, meaning for instance that
       material dumps can be captured in strings rather than emitted on
       std-output. Finally, this new feature could in principle be used for
       various output redirection purposes (e.g. a GUI application might wish to
       show all output from NCrystal in some dedicated text box).
     * NCrystal has been updated to work with latest versions of Numpy, Clang,
       GCC, Python, Gemmi, and Spglib. Additionally work has started to support
       Intel OneAPI compilers and Microsoft VisualStudio, although more work is
       needed to finalise this. As part of these efforts, the ABI had to be
       broken in some places, since for instance Microsoft VisualStudio did not
       support all the data structures used previously (for instance,
       VisualStudio `std::vector` does not accept move-only objects like
       `std::unique_ptr`).
     * This concludes the highlights in this changelog entry - refer to the
       following items for more detailed information about specific changes.
     * Several speedups of the phonon convolution calculation code was
       implemented, resulting in noticable speed increases (20-40%): the FFT
       phase factor calculation by a factor of two, swap indices were cached,
       and an integer division was rewritten as a bit shift.
     * Reduce luxury in some VDOS integrations slightly for a more reasonable
       amount of calculations, speeding up Info material creation.
     * Add internal utilities to allow usage of multiple threads for material
       loading, allowing for sizeable speedups. This required many changes to
       ensure good behaviour (i.e. using a fixed number of threads only), and no
       dead-locks of various kinds even though the multi-threading can take
       place at various levels during the loading. For now, the number of
       available threads can be controlled with the NCRYSTAL_FACTORY_THREADS
       environment variable, the NCrystal::FactoryThreadPool::enable() C++
       function, the NCrystal.enableFactoryThreads Python function, or the
       ncrystal_enable_factory_threadpool C function.
     * Build NCrystal library with -fno-math-errno (github issue #169).
     * Several non-trivial fixes were implemented to prepare for MS VisualStudio
       and Intel oneapi compilation support. This breaks ABI.
     * Apply -fp-model=precise for intel oneapi compiler.
     * Change C++ DynamicInfoList type from
       std::vector<std::unique_ptr<DynamicInfo>> to typedef
       SmallVector<std::unique_ptr<DynamicInfo>,4>, since VisualStudio
       apparently does not support std::vector's with move-only objects. This in
       principle breaks the ABI and API, but is deemed necessary. A few other
       types in internal non-API headers were changed as well for the same
       reason.
     * Fix build error due to ODR violation in the TopLvlVar type.
     * Removed noexcept from C++ functions where it should not have been added
       (e.g. those used for streaming to std::ostream).
     * Fix cfg-string bug (github issue #182). This only affected multiphase
       cfg-strings in which the first phase contained either a density or
       phasechoice setting, so is not believed to have affected many users.
     * ProcComposition::consumeAndCombine(..) fct is now better at removing null
       components.
     * Adding the infrastructure for better control of NCrystal output, through
       usage of a new setMessageHandler(..) function (related to github issues
       #53 and #122).
     * Rewrote the NCrystal::dump C++ function (which provides printouts/"dumps"
       of material Info objects) to stop using printf entirely and now only use
       C++ output streams, but notably while preserving existing output
       exactly. This makes it possible to capture the dumps into strings and
       streams, rather than printing them out. Along with the general message
       handling changes, this should prevent Python output buffering causing
       incorrect output ordering (seen in e.g. Jupyter notebooks).
     * Python objects with .dump(..) methods gets new .dump_str() methods as
       well, which simply returns the information as a string instead of
       printing it directly.
     * Update cifutils to follow changes in upstream Gemmi project.
     * Updates to reflect upstream API changes in spglib.
     * Stop using numpy.asfarray which was removed in numpy 2.0.0 (github issue
       #185).
     * More helpful error message when loading huge DFT-generated unit cells
       (thousands of atoms, no symmetries) without raising dcutoff (github issue
       #160).
     * DI_VDOSDebye Python objects now also get an analyseVDOS() method, for
       consistency (github issue #174).
     * Update NCMAT documentation to clarify the ordering of sab entries (github
       issue #170).
     * Add convenience .xsect(..) and .macroscopic_xsect(..) methods to
       LoadedMaterial instances in the Python API.
     * Add experimental material loading benchmark utility to nctool. This
       feature is for now kept out of the documentation, but can be tested with:
       "nctool --bench --help".
     * Updated year in copyright notice in all files.
     * Info objects gets a new .hklObjects() method which returns HKL list
       information in a more object oriented and convenient format than the
       existing .hklList(). See issue #164 for more details.

v3.8.2 2024-04-30
     * Add importlib_metadata as python build dependency for Python < 3.8.
     * CMake code no longer strictly disallows mixing of NCRYSTAL_NAMESPACE and
       NCRYSTAL_GEANT4 options.

v3.8.1 2024-04-11
     * Fix rare FPE in lcbragg code (github issue #171).
     * Remove a few faulty noexcept modifiers in (found by cppcheck).
     * Fix NCMatCfg::textData which technically returned a reference to a
       temporary (although this is not believed to have actually caused any
       issues, since the "temporary" came from a longer-lived shared_ptr). This
       was also found with cppcheck.

v3.8.0 2023-12-03
     * Completely obsolete the ncrystal_inspectfile command in favour of the
       nctool command. Trying to invoke ncrystal_inspectfile now results in an
       error message with a reminder to use nctool instead.
     * The default value of the CMake cfg option -DNCRYSTAL_ENABLE_SETUPSH is
       changed from "ON" to "OFF". Additionally, if setting
       -DNCRYSTAL_ENABLE_SETUPSH=ON, the oly files installed are now named
       <PREFIX>/ncrystal_setup.sh and <PREFIX>/ncrystal_unsetup.sh, whereas in
       the past they were named <PREFIX>/setup.sh and <PREFIX>/unsetup.sh. Note
       that in releases v3.7.0 and v3.7.1 both versions of the files were
       available. Consequently, users who were manually compiling NCrystal and relying on
       e.g. <PREFIX>/setup.sh files files will now need
       to explicitly set -DNCRYSTAL_ENABLE_SETUPSH=ON and instead use the
       <PREFIX>/ncrystal_setup.sh file. (github issue #126).
     * NCrystal components and instrument examples for McStas (mcstas.org) are
       no longer shipped with NCrystal, since all NCrystal components and
       instrument examples are now shipped with McStas itself.
     * Add a new NCRYSTAL_NAMESPACE option, which can be used when there is a
       need to be able to use NCrystal header files and libraries from multiple
       NCrystal installations into the same environment. Eventually, it is the
       plan to enable such namespace protection in NCrystal conda-forge and PyPI
       packages, to prevent hard-to-catch errors in case a user ends up having
       both packages installed in the same environment (which can happen when
       people mix "conda install .." and "pip install .."). CMake-based
       downstream projects should not be affected by this, but if you are
       maintaining downstream custom builds which hardcode linker flags like
       "-L/path/to/ncrystal/lib -lNCrystal" you are encouraged to based your
       flags instead on $(ncrystal-config --show libpath). That way you will be
       future-proof and ready to also work with NCrystal installations where a
       namespace has been enabled. See the README file for more details.
     * As a result of the namespace support, all function names in the C-api are
       now actually macros.
     * Adding VERSION and SOVERSION to libNCrystal.so/dylib. For now, the
       VERSION number is simply the full NCrystal release version number, while
       the SOVERSION is the major release number (i.e. "3" for release
       3.8.0). This feature can be disabled with -DNCRYSTAL_ENABLE_SOVERSION=OFF.
     * CMake find_package(NCrystal ...) calls no longer sets NCrystal_DIR to the
       installation prefix, since that could disrupt the usage of multiple
       find_package calls. The prefix directory previously available as the
       CMake variable NCrystal_DIR is thus instead made available under the name
       NCrystal_PREFIX.
     * Update Avogadro's constant to NIST/CODATA2018 value, changing it from
       6.022140857e23/mol to 6.02214076e23/mol (a change of ~0.016ppm).
     * The CMake code now also adds $CONDA_PREFIX/lib to rpath when appropriate.
     * Explicitly forbid # and $ chars in cfg-strings (github issue #153).
     * Fix misleading values in ncrystal-config in some rare cases.
     * Fix for a few (incorrect) compiler warnings.
     * Fix for cmake test in python3 -mNCrystal.test cmake.
     * Warnings emitted from python layer now has right stacklevel.
     * Fix tests for Python 3.12 (a few tests depended on the now
       deprecated distutils modules, cf. PEP-632).

v3.7.1 2023-08-02
     * Update documentation to reflect availability on PyPI.
     * Set version-based policies in NCrystalConfig.cmake to make
       find_package(NCrystal) work when CMake is new enough, but the downstream
       CMake project nevertheless specifies a very old version in their
       cmake_minimum_required statement (cf. github issue/discussion #137).
     * Add NCrystal_LIBRARIES variables for downstream CMake-based projects with
       legacy 2.x-style CMake code. If the GEANT4BINDINGS component is enabled,
       also set G4NCrystal_LIBDIR, G4NCrystal_INCDIR, G4NCrystal_LIBNAME and
       G4NCrystal_LIBRARIES variables, likewise intended for such legacy
       projects. Note of course, that usage of such legacy CMake 2.x-style code
       is discouraged.
     * Include type_traits in NCDefs.hh, to potentially fix some compilation
       errors seen in #137.
     * Fix several minor bugs in NCrystal Python code which were discovered via
       the ruff static code analysis tool.
     * Change signature of all 0-argument functions in C-API in ncrystal.h from
       f() to f(void), since the first form is apparently deprecated since 30
       years and gives issues in clang 14
       (cf. https://github.com/madler/zlib/issues/633 and
       https://github.com/mctools/dgcode/issues/34).
     * For NCrystal builds, change max version in cmake_minimum_required to
       3.27.

v3.7.0 2023-07-14
      * The main feature of this release is that NCrystal now can be optionally
        built and installed using standard Python-centric toolchains. Work on
        this feature was spearheaded by Milan Klausz, and is associated with
        newly added setup.py and pyproject.toml files, using skbuild behind the
        scenes. Additionally, Python modules and script were moved out of the
        ncrystal_python directory and into directories named NCrystal and
        scripts, respectively (solving github issue #127). Finally, existing
        Python modules and the ncrystal-config script were modified to support
        such builds and installation. As a result, it is now possible to do "pip
        install <path-to-ncrystal-src>", and get a fully-fledged NCrystal
        release. The only difference with respect to e.g. a conda installation
        of NCrystal, is that downstream CMake-based projecs won't detect the
        installation, unless the directory reported by "ncrystal-config --show
        cmakedir" is manually added to the CMAKE_PREFIX_PATH.
      * Work is ongoing by Milan Klausz to setup workflows at GitHUB to
        automatically build binary wheels for PyPI, similar to how binary
        packages are now built automatically at conda-forge. Once this is
        completed, it will be possible to install prebuilt NCrystal packages via
        "pip install ncrystal" (solving github issue #93).
      * Note that setup.py maps NCrystal releases with minor number >= 80 to a
        development version lke "3.6.dev80", with the intention of preventing
        "pip install ncrystal" from automatically picking up release candidates.
      * Rewrote the aligned allocator code used internally when implementing
        convenience containers such as SmallVector. This was done in response to
        fix compilation errors with gcc12 (github issue #125). This new version
        avoid the previous incorrect reliance of a non-standard feature of
        std::free, and overall the new version is more portable, as it relies on
        nothing except std::malloc and std::free.
      * Updated the SmallVector::clear method, in order to work around what
        appears to be a false-positive compilation error by gcc12.
      * For completeness, added more functions like ekin2k, constexpr_ekin2k,
        k2ekin, and k2wl functions in the C++ and Python API's.
      * Bugfixes (C++ API): NeutronEnergy::k() now actually returns the
        wavenumber k, and not the squared of k, as it did previously. Also, the
        free-standing ekin2wl(..) function is no longer constexpr (since
        std::sqrt is not actually constexpr), but there is a new
        constexpr_ekin2wl which is.
      * Added new standard testing facilities, for checking an installation for
        presence of commandline scripts (solving github issue #110) and that
        downstream CMake/C++ projects can find and use NCrystal (after getting
        CMAKE_PREFIX_PATH from ncrystal-config). To invoke all tests, run
        "python3 -mNCrystal.test all". For other options, run "python3
        -mNCrystal.test all".
      * Add ncrystal_setup.sh/ncrystal_unsetup.sh files in addition to the
        ambiguously named setup.sh/unsetup.sh files. (github issue #126).

v3.6.82 2023-07-13
      * RC3 for release 3.7.0.

v3.6.81 2023-06-23
      * RC2 for release 3.7.0.

v3.6.80 2023-06-07
      * RC1 for release 3.7.0.

v3.6.1 2023-05-16
      * Tiny update with a quick workaround for gcc12 compilation issues
        (github issue #125).

v3.6.0 2023-05-16
      * This milestone release primarily brings a massive restructuring and
        update of the Python API of NCrystal. The changes should be backwards
        compatible in almost all cases, and the most user-visible changes should
        hopefully be the addition of several new big features, highlighted
        below.
      * Note that along with this release we are also publishing several Jupyter
        notebooks at https://github.com/mctools/ncrystal-notebooks which might
        be a more enjoyable (or at least useful) way to learn the Python API of
        NCrystal. These notebooks will also be used during the "HighNESS
        International School on Thermal Neutron Scattering Kernel Generation"
        hosted at the European Spallation Source in Lund
        (https://indico.esss.lu.se/event/3096/).
      * The new NCMATComposer helper class is added to the Python API. It can be
        used to create new materials in an object-oriented and pythonic manner, as
        opposed to hand-crafting the NCMAT data. It essentially supports all
        features of the underlying NCMAT format, but does so in a much more
        easily accessible manner, and handles a lot of the book-keeping that can
        arise in some complicated materials.
      * A new NCrystal.cifutils submodule brings a CIFLoader class, which behind
        the scene empowers the NCMATComposer class to adopt crystal structures
        from CIF files. Such files can reside both locally or in selected online
        databases. CIF files often contain incomplete information, so be sure to
        consult the Jupyter notebook tutorial linked above for examples of how
        to deal with these. Note that based on this CIF functionality, the
        NCMATComposer now also support ASE (https://wiki.fysik.dtu.dk/ase/)
        objects, or the loading of a variety of input formats via ASE.
      * A new PhononDOSAnalyser class can be used to investigate and cleanup
        phonon density of state (VDOS) curves, and then (with the help of the
        NCMATComposer class) add them to a material. In particular, one of the
        Jupyter notebooks linked above shows how one might go through the steps
        of converting phonon curves produced by Quantum Espresso
        (https://www.quantum-espresso.org/) to define high quality NCMAT data
        for a material.
      * A new NCrystal.plot submodule brings several new plotting utilities,
        that can either be used directly, or via various convenience plotting
        methods added to various classes in the API. In particular, note the
        plot_xsect(..) and plot_xsects(..) function in the submodule, which can
        be used to easily produce cross section plots from a range of inputs
        (NCrystal objects, NCMAT data, cfg-strings, etc.).
      * In general, many objects, both new and old, have been enriched with more
        convenience methods for plotting, dumping, or merely extracting
        data. Note for instance the new .findDynInfo and .findAtomInfo methods
        of the Info objects, or the various .plot_xxx() methods on the
        DynamicsInfo objects.
      * A new global .load() function can be used to load either cfg-strings or
        raw NCMAT data into LoadedMaterials objects. LoadedMaterials objects
        have .info, .scatter, and .absorption properties available for accessing
        all of these physics objects. Loading a material in such a unified
        manner might be more convenient for many use-cases (in particular
        interactive sessions) than the classic .createInfo(..),
        .createScatter(..), and .createAbsorption().
      * Note that while NCrystal in general does not have any required
        dependencies, several of the features above does depend on third-party
        python modules. All these modules are well-maintained open-source
        projects, available both on PyPI (for "pip install xxx") and on
        conda-forge (for "conda install -c conda-forge xxx"). The optional
        dependencies are:
        - "numpy": for various array-based functionality. Although not strictly
          needed, it is recommended to always install this.
        - "matplotlib": for usage of the plotting utilities.
        - "spglib": For producing crystalline materials with NCMATComposer, in
          order to verify (or refine) the structures.
        - "gemmi": For reading CIF data into NCMATComposer.
        - "ase": For loading crystal structures from ASE objects or non-CIF files
          (like those produced by Quantum Espresso) into NCMATComposer.
        -"pymatgen"/"mp_api": Only needed for loading CIF data from the online
          database at materialsproject.org into NCMATComposer.
      * Since release 3.0.0 we have provided the alias "nctool" for the command
        named "ncrystal_inspectfile", and while we will keep providing both
        aliases, we now swap their role and primarily mention the name "nctool"
        in our documentation. There are not immediate plans to remove the
        "ncrystal_inspecfile" command, but users are encouraged to migrate to
        use the "nctool" command instead.
      * Introduce new ncrystal_cif2ncmat command which internally uses the new
        CIF and NCMAT tools from the Python API, and which supports both local
        CIF files or remote online databases. This also obsoletes the
        ncrystal_onlinedb2ncmat command.
      * The rest of this changelog will list a range of other improvements and
        bug fixes that are also a part of this release.
      * Added new utility functions in the NCrystal.atomdata Python module:
        elementNameToZValue, elementZToName, allElementNames, isElementName,
        knownElementNames, and isKnownElement.
      * Added new conversion functions in the NCrystal.constants module in
        addition to the existing. The full list is now wl2ekin, ekin2wl,
        wlsq2ekin, ekin2wlsq, ekin2wlsqinv, ekin2ksq, ekin2k, ksq2ekin,
        wl2k. However, note that only the first two are put in the main module,
        so one must explicitly import NCrystal.constants to access the rest
        (existing users of ekin2ksq, wl2k, wl2ksq functions will therefore have
        to modify their code).
      * Be in general aware that a few specialist functions were moved from the
        main Python module to sub-modules.
      * The AtomData class in the Python API gets new .fmt_atomdb_str(..) and
        .to_atomdb_str() convenience methods, for converting data into a string
        with a format suitable for @ATOMDB sections of NCMAT data. Perhaps more
        noticably, the AtomData class also gets __str__ and __repr__ methods for
        a better interactive experience with the NCrystal.atomDB function.
      * NCrystal.Info objects gets a .factor_macroscopic_xs property, which can
        be used to convert cross sections from (barns/atom) to inverse
        penetration depth (1/cm). This is numerically identical to the
        .numberdensity property, but more explicit.
      * Rename NCrystal.directMultiCreate as NCrystal.directLoad, but keeping
        the old name as an alias for now. Also make it more accepting in terms
        of the kinds of data it accepts. Most users likely use the new
        NCrystal.load(..) function instead though, as it accepts a wider range
        of input types.
      * Prepare Python modules with a special mode for finding libNCrystal.*,
        needed for an hopefully upcoming PyPI package being developed with the
        help of Milan Klausz.
      * Add "version_tuple" variable in the main NCrystal module, so one can do
        "if NCrystal.version_tuple>=(3,6,0)" instead of "if
        NCrystal.version_num>=3006000". But be aware that the
        NCrystal.version_num exists all the way back to NCrystal release 0.9.1,
        so it will for now result in nicer error messages if one checks against
        NCrystal.version_num (at least for the next few years).
      * Allow atomic positions in unit cell with fractional coordinates within
        1e-4 of each other. The previous limit of 1e-2 was too tight for some
        CIF files.
      * Single crystal materials with no hkl entries (due to high dcutoff or
        forced b_coh=0 for atoms) no longer results in an exception from the
        PlaneProvider. Rather, the std plane provider instead now provides an
        empty list of normals, avoiding spurious errors.
      * Modify the .laz/.lau loader to be able to work directly with files
        produced by the cif2hkl programme. Specifically this means being able to
        extract spacegroup and chemical composition from the SPCGRP and TITLE
        fields. It still might not load all files though, since cif2hkl might
        produce small fluctuations between structure-factors for
        symmetry-equivalent planes, which NCrystal won't accept currently.
      * Due to the new improved utilities for interpreting CIF files for
        conversion to NCMAT, a few materials in the data-library received
        corrections. Additionally, several online database entries (in
        particular those at materialprojects.org) appear to have been updated
        with supposedly more accurate content. These changes are naturally
        reflected in the new validation plots at the data library
        page. Consequently, two materials received updates to structures
        (CaSiO3_sg2_Wollastonite.ncmat and KOH_sg4_PotassiumHydroxide.ncmat),
        and for a few other materials the list of relevant online database
        entries (mentioned in comments in the files) was updated. As a comment,
        we observe a general trend for structures loaded from
        materialprojects.org to now better match the equivalent structures
      * The C API function ncrystal_ncmat2json now works with raw NCMAT data
        rather than just ncmat files. It is still an incomplete implementation
        though.
      * Updated README file, including adding a mention about OpenMC support.
      * Modify CMake to make sure downstream C++ projects use at least C++11.
      * Data files embedded in the binary with -DNCRYSTAL_ENABLE_DATA=EMBED will
        no longer be in a compact form stripped for comments. This was a
        mistake, of course the content of the stdlib data files should never
        depend on installation details.
      * Fix bug in ncrystal_ncmat2cpp which could introduce extra random bytes
        at the end of huge (>60KB) files. This affected two files
        (LiquidHeavyWaterD2O_T293.6K.ncmat and LiquidWaterH2O_T293.6K.ncmat) in
        the standard library when building with -DNCRYSTAL_ENABLE_DATA=EMBED,
        like is done in the conda packages.
      * NCScatKnlData now allows a vanishing boundXS, which would earlier
        trigger an exception. Related to this, the standard scattering factory
        has been updated to not initialse SABScatter processes based on such
        vanishing kernels, or if the scattering cross section is vanishing.
        Also Added a few checks elsewhere in the code against sterile elements
        (all scattering lengths 0), ensuring not only a tiny speedup, but also
        that consistency checks did not falsely trigger exceptions.
      * It is now possible to get all usual NCrystal functions in the api Python
        submodule, meaning that "import NCrystal.api as NC" will mostly (apart
        from a bit of package metadata variables) give the same result as
        "import NCrystal as NC" since the __init__.py module also does an "from
        .api import *" import internally. However, this latter import can be
        avoided by setting the environment variable NCRYSTAL_SLIMPYINIT.
      * Many obsoleted functions will now emit a warning upon
        usage. Additionally, setting the environment variable
        NCRYSTAL_NOPYOBSOLETE before importing the NCrystal modules, will remove
        many such obsolete functions entirely. Note that some obsolete functions
        might not yet have been migrated to this new scheme.
      * Better error messages if code like the following causes the Info object
        to be cleaned up before an associated dyninfo object needs it:
            di=NC.createInfo("bla.ncmat").dyninfos[0]
            print(di.vdosData())
        A better solution for the future would be to have DynInfo + AtomInfo
        objects exposed directly as objects in the C-API, so the object
        lifetimes in Python could be directly tied to the C++-side equivalents.
      * Expose raw Gn functions and vdos2knl expansion code to both C and Python
        APIs, and make available on dyninfo objects as well for convenience.
      * Caching of idealised VDOS based on Debye temperatures use slightly more
        digits in the cache keys.

v3.5.81 2023-05-14
      * RC2 for incoming release v3.6.0.

v3.5.80 2023-05-09
      * RC1 for incoming release v3.6.0.

v3.5.1 2022-12-05
      * Redo VDOS integrals for gamma0 and Teff to support much lower
        temperatures (down to 0.001K certainly no issue now).
      * Ensure that the evaluation of the G1 function is now numerically safe,
        even at very low temperatures where the detailed balance factor becomes
        numerically indistinguishable from 0 or infinity (github issue #109).
      * Trim excess zeroes at edges of G1 function.
      * Unify temperature range checks. More specifically, replace asserts which
        were imposing temperature limits inconsistent with those allowed by the
        "temp" cfg-parameter (1e-3K to 1e6K).
      * Make FileListEntry returned by NCrystal.browseFiles sortable.
      * Fix ncrystal_endf2ncmat (github issue #110).

v3.5.0 2022-11-23
      * Migrate names of all CMake parameter options to be prefixed with
        "NCRYSTAL_". Using the old names will mostly keep working for the time
        being, but will cause a warning to be issued (mctools/ncrystal#105).
      * No longer build and install examples by default.
      * We now newer fiddle with CMAKE_BUILD_TYPE if a multi-cfg generator is
        detected.
      * Add option NCRYSTAL_ENABLE_CPACK. If set, a few CPACK-related variables
        will be set and an "include(CPack)" statement is triggered just after
        the "project(..)" statement (mctools/ncrystal#107).
      * Rename CMake target "common" to "ncrystal_common" to avoid clashes when
        project is build as a subproject (mctools/ncrystal#105).
      * Completely ignore the CMake option BUILD_EXTRA which has not been
        supported since release v3.0.0.
      * Minor fix for setup.py file generated from CMake, needed to work with some
        outdated python installations.
      * Add hidden --pytrace option to ncrystal_inspectfile and nctool, for
        running the script under the python trace module.
      * A few fixes for Windows (thanks to Peter Willendrup). In particular,
        never use aligned_alloc on windows.

v3.4.1 2022-10-19
      * Expose utility for completely flattening a material composition to
        Python interface, as a new info.getFlattenedComposition() method. This
        is intended to better support materials with non-natural compositions in
        the OpenMC bindings. Internally, the new method is using the same
        machinery as what is already in place to provide the same functionality
        for the Geant4 bindings.
      * ncrystal_onlinedb2ncmat was not using new unified cmdline script
        machinery introduced in v3.4.0, which has now been corrected.

v3.4.0 2022-10-12
      * Modify location of CMake config files provided by an NCrystal
        installation, to hopefully make it easier for downstream projects to
        locate an existing NCrystal installation (more details on this MCPL
        issue: https://github.com/mctools/mcpl/issues/58).
      * Modify manner in which Geant4 bindings must be activated in downstream
        CMake code. Previously they were enabled via a dedicated
        "find_package(G4NCrystal)" call, but now they are instead provided as an
        optional COMPONENT when requesting the NCrystal package:
        "find_package(NCrystal COMPONENTS GEANT4BINDINGS )" and if succesful the
        downstream binaries must be configured to depend on the G4NCrystal
        interface by adding the NCrystal::G4NCrystal target as a dependency,
        i.e. "target_link_libraries( mybinary NCrystal::G4NCrystal )".  The
        Geant4 bindings will not be enabled unless the GEANT4BINDINGS component
        is explicitly requested.
      * Unify manner in which shipped cmdline scripts look for NCrystal python
        module.

v3.3.1 2022-10-06
      * Fix info.getComposition() for multiphase materials in the Python
        interface.  This issue is not believed to have affected many users, but
        a fix is crucial in order to support multiphase materials in OpenMC
        hooks which are currently being prepared (github issue #102).
      * Minor fix to make Geant4 hooks work with Geant4 release 11. Note that
        this is still not the long awaited major update which makes the hooks
        work with Geant4 in multithreaded mode.
      * Rewrote internal code in NCIter.hh to avoid usage of deprecated
        std::iterator. This avoids compilation warnings on newer platforms
        (github issue #100).
      * Reduce precision from %.15g to %.14g in output of ncrystal_ncmat2hkl to
        get reproducible results across platforms.

v3.3.0 2022-08-23
      * Updates to CMake code which is needed to support creation of
        NCrystal conda packages.
      * ncrystal_inspectfile now has --version and --versionnum arguments.

v3.2.0 2022-08-11
      * Introduce a new quick TextData factory for gas mixtures, supporting
        on-the-fly gas mixtures through cfg-strings like
        "gasmix::0.7xCO2+0.3xAr/2bar/30C".  The format is rather flexible, and
        supports both mass and molar fractions, relative humidity, and has an
        alias for normal air ("gasmix::air"). See more details on the wiki:
        https://github.com/mctools/ncrystal/wiki/Announcement-Release3.2.0
      * Introduce NCMAT v7 which adds a new optional @TEMPERATURE section to the
        format, allowing the default temperature value to be modified from the
        normal 293.15K, or even of locking the value at a certain value
        (attempts at modifying it via cfg variables will then fail). This is
        used by the new gasmix factory to error on the flawed but
        correct-looking "gasmix::CO2/2bar;temp=200K" (which is not the same as
        the correct "gasmix::CO2/2bar/200K"). Future developments might move the
        gas mixture support to the NCMAT format itself (rather than just in the
        pre-processing as now), which would solve the issue in a nicer way.
      * At the C++ level, a new fmt_frac function allows easy printing of
        floating point values which might be simple fractions, a dedicated
        Pressure type was added, and a minor issue with confusing error messages
        from NCMAT data from quick factories was addressed.

v3.1.0 2022-06-29
      * Introduce cfg parameter named "ucnmode" which can be used to enable a
        new dedicated modelling of UCN-production processes. Thus, appending
        ";comp=inelas;ucnmode=only" to a given cfg-string, will result in a
        scattering process which ONLY contains UCN-production
        events. Conversely, appending ";ucnmode=remove" will provide a process
        which contains everything EXCEPT UCN-production events. This splitup is
        intended to facilitate usage of biasing and variance reduction tricks
        when designing VCN or UCN moderators. The UCN threshold defaults to
        300neV, but this can of course be modified. In addition to just
        facilitating a process splitup, the UCN-production process resulting
        from "ucnmode=only" is more precise (essentially exact) than the usual
        S(alpha,beta) treatment - which is rather precise in general, but
        currently contains some flaws when sampling events where the neutron
        loose almost all energy (see also next item).
      * Improve general (i.e. also when not running with special ucnmode)
        scattering-kernel based inelastic physics for the case where the neutron
        looses almost all energy. Specifically, the sampling of S(alpha,beta)
        near the kinematic endpoint is greatly improved. We still have
        additional plans for how to completely remove any artifacts related to
        the numerical processing, if manpower allows at some point in the
        future.
      * Add internal utilities for evaluating S(alpha,beta) kernels at any
        (alpha,beta) point, and add utilities for what is essentially exact
        sampling or integration for cross sections. These utilities are useful
        as they allow us to validate scattering kernel models better, and are
        used directly in the new UCN-production models.
      * Add notice to the ncrystal_preparemcstasdir for how to make it work when
        using NCrystal with the McStas-Union geometry system.
      * CMake now actually installs the "nctool" alias of "ncrystal_inspectfile".
      * S(alpha,beta) kernels hardcoded in .ncmat files will now have any
        null-edges discarded upon loading. This is often needed for kernels
        converted from ENDF files, where the alpha/beta grid is often larger
        than needed.
      * Add PiecewiseLinearFct1D utility class (for some reason we had utilities
        for cubic splines and sampling of piece-wise linear functions, but not
        for simple evaluations of them!).
      * Fix bug when accessing DynInfo of directly specified scattering kernels
        with the Python interface (the DI_ScatKnlDirect was missing a required
        initialisation).
      * The process summaries printed via ncrystal_inspectfile or accessed via
        the Python interface, now supports multiple layers of nested
        processes. This was required to support the new ExcludeUCNScatter
        process, which is implemented as a wrapper of two other processes.
      * Adding experimental and incomplete support for converting raw parsed
        NCMAT data to JSON. This is intended to help avoid the need for
        resorting to custom parsing code, but is clearly an expert-only feature
        as the data is presented as-is, without additional loading.

v3.0.0 2022-04-24
      * The NCrystal v3.0.0 release brings both major new features and technical
        improvements "under the hood", with around 19.000 lines of code changed
        in more than 300 files! The list of specific changes is too long to
        document here, but the following bullet points mention a few highlights
        as well as a select number of very specific technical details. For a
        more detailed and hopefully readable overview, please refer to the
        following page in the wiki:
        https://github.com/mctools/ncrystal/wiki/Announcement-Release3.0.0
        Furthermore, we would like to encourage users to post questions
        concerning the new release (or anything else NCrystal related obviously)
        to the forum at: https://github.com/mctools/ncrystal/discussions
      * Support for multiphase materials. These can be defined in terms of phase
        volume-fractions, either in cfg-strings
        ("phases<0.9*Al_sg225.ncmat&0.1*Be_sg.ncmat>"), in NCMAT data, or a
        combination of those. All interfaces and tools have been updated to
        properly integrate multiphase materials, in a manner which is believed
        to not add complications when working with single-phase materials.
      * Framework-level support for Small Angle Neutron Scattering (SANS)
        physics. The release includes resources to facilitate future development
        of advanced SANS models, and already include a hard-sphere SANS model
        (useful both by itself, and as proof-of-concept for other models).
      * Material densities can now be easily modified. This can be done by
        either specifying a relative scaling,
        "Polyethylene_CH2.ncmat;density=0.9x", or an absolute one:
        "Polyethylene_CH2.ncmat;density=0.88gcm3" (g/cm^3) or
        "Polyethylene_CH2.ncmat;density=0.1perAa3" (atoms/Aa^3).
      * Support for quickly setting up unstructured materials with just a
        cfg-string and without a data file for such simple materials. For
        instance, a He3 gas might be defined via the cfg-string
        "freegas::He/0.17kgm3/He_is_He3". This makes it easy to add materials
        for which the structure is not deemed important (e.g. in a strong
        absorber).
      * The various new features can be combined for quickly settting up
        advanced materials. For instance, here is how to define a low-density
        polyethylene with enriched B4C pellets, even without a dedicated data
        file for B4C:
        "phases<0.99*Polyethylene_CH2.ncmat;density=0.8x&0.01*solid::B4C/2.5gcm3/B_is_0.95_B10_0.05_B11>"
      * Improve documentation of cfg-string parameters, with the code now able
        to provide dynamic documentation, which can be accessed either
        programatically (with the Python/C/C++ API or by running
        ncrystal_inspectfile --doc), or on the new wiki page:
        https://github.com/mctools/ncrystal/wiki/CfgRefDoc
      * Introduce for convenience a new cfg-string parameter "comp", which can
        be used to easily pick out one or more of the components contributing to
        the scattering cross section. For instance, appending ";comp=inelas" to
        any cfg-string (i.e. "Al_sg225.ncmat;comp=inelas"), will disable all
        components *except* inelastic models.
      * Reflection (hkl) planes provided by NCrystal are now grouped according
        to the symmetry-equivalency indicated by the space group. This has the
        benefit of drastically improving both the initialisation time and memory
        usage needed for hkl lists, and additionally makes it easier to use
        reflection lists from NCrystal with other software which often depends
        on such symmetry groupings. Additionally, ondemand loading for quicker
        inspection (e.g. --dump will be almost instantaneous even for the most
        complicated crystal structures), and by default only the 10 planes with
        largest dspacing is shown.
      * Improve and update McStas support:
        - Make NCrystal work with both the McStas v2.7 and McStas v3.0 branches.
        - Add ncrystal_ncmat2hkl cmd-line utility for converting NCMAT files to
          the .laz and .lau files needed by the McStas PowderN and
          Single_crystal components (note that this only includes Bragg
          diffraction, not inelastic/incoherent/SANS/multiphase physics for
          which direct NCrystal usage in the NCrystal_sample component or
          McStas-Union system is needed).
        - Utilities for better/easier integration between NCrystal and the
          McStas-Union system for complex geometries.
      * Remove the few optional files which were not copyrighted by the NCrystal
        developers, ensuring that the entire NCrystal release now has a single
        well-defined license (Apache 2.0). The removed files were used to
        support the semi-obsolete .nxs file format, and will still be available
        as a separately distributed plugin if really needed.
      * Stop supporting Python 3.5. Oldest supported Python version is now
        Python 3.6.
      * Introduce NCMAT v6 which adds a new optional @OTHERPHASES section to the
        format, allowing secondary phases to be defined via volume fractions and
        cfg-strings.
      * Remove NCMatCfg::isPolyCrystal which used to return the opposite of
        isSingleCrystal(). This was obviously misleading for non-crystalline
        materials.
      * Vector::normalise() now does nothing if the squared magnitude is already
        within 2 ULP's of unity. This is to avoid repeated calls to this
        function from inducing slight changes due to numerical inaccuracies in
        the normalisation process.
      * Add cyclical dependency detection in NCrystal factories to exit with
        clear error instead of deadlocking. This is in particular needed to
        handle cyclic dependencies in multiphase materials in which a phase is
        wrongly defined by recursively referring back to the top level material.
      * Type safe AtomData::captureXS in C++.
      * Add utilities for encoding data in JSON format, which for now is used to
        enable conversion of MatCfg objects as JSON objects. On the python side
        this means that a cfg-string can be decoded into a Python dictionary
        using the new decodeCfg function (a new encodeCfg function re-encodes
        this into a cfg-string).
      * Fix bug where scattering kernels generated on-the-fly from VDOS curves
        with vdoslux values less than 3 (the default) were subsequently treated
        to the "beta-thickening" procedure introduced in release v2.6.1. This
        procedure is now, as intended, only applied to externally provided
        kernels, since the on-the-fly generated ones already have an optimised
        grid layout.
      * ncrystal_inspectfile gets --energy/--liny/--logy/--xrange options to
        control more aspects of the generated plots.
      * Introduce an alias, nctool, for the ncrystal_inspectfile command (this
        is done in anticipation of a future actual renaming of the command).
      * Fix misleading @ATOMDB examples in NCMAT documentation where cohSL
        numbers for Si and Al were a factor of 10 off from realistic numbers,
        which might confuse users to think they have to use a unit of 0.1fm
        rather than fm for such numbers (github issue #77).

v2.9.93 2022-04-21
      * Release candidate #3 for release 3.0.0.
      * Detailed ChangeLog to come.

v2.9.92 2022-04-10
      * Release candidate #2 for release 3.0.0.
      * Detailed ChangeLog to come.

v2.9.91 2022-03-02
      * Release candidate #1 for release 3.0.0.
      * Detailed ChangeLog to come.

v2.7.3 2021-09-09
       * Add missing import statement in ncrystal_hfg2ncmat (github issue #71).
       * C++ API gets ekin2ksq and ksq2ekin utility functions.
       * Cleanup the PointwiseDist utility class and add new sampleBelow and
         commulIntegral utility functions
       * Add new internal IofQHelper utility class which can be used to
         implement elastic physics described via an I(Q) curve
       * Fix some obscure -Wshadow warnings on some rare platforms when some
         local variable named s or m clashes with global variables of the same
         name from Geant4.

v2.7.2 2021-07-08
       * Various compilation fixes, affecting only users of Geant4-hooks with
         the most recent Geant4 release (10.07.p02) or users with the rather
         outdated gcc 4.8.5 (github issue #69).

v2.7.1 2021-06-17
       * Fix single crystal alignment bug which was inadvertantly introduced
         among the many technical changes in release v2.5.0, essentially
         swapping the meaning of @crys: and @crys_hkl: in the parts of the cfg
         strings defining the crystal orientation.

v2.7.0 2021-05-31
      * Introduce NCMAT v5 which two updates for the format, both meant to
        facilitate modelling of amorphous (i.e. non-crystalline) solids.
        Firstly, Debye temperatures can now be specified directly in @DYNINFO
        sections of vdosdebye type, meaning that such sections can now also be
        used in non-crystalline materials where @DEBYETEMPERATURE sections are
        disallowed. Secondly, a new optional @STATEOFMATTER section can be used
        to explicitly indicate the state of matter if needed. If @STATEOFMATTER
        is not specified, materials with crystal structure or vdos curves are
        assumed to be solids, otherwise the state will be undefined. Amorphous
        solids will currently get both inelastic and elastic physics determined
        from the VDOS curves - with coherent components approximated via the
        incoherent approximation.
      * Improve HKL plane initialisation code to make it 200% faster (i.e. a
        factor of 3 speedup)! This was mostly achieved by using our own custom
        sin+cos implementations, which are much faster than the standard ones,
        but with no impact on precision. As a result, the heuristics picking a
        default dcutoff value were updated so that structures with more than 40
        atoms per unit cell now defaults to dcutoff=0.2Aa, as opposed to
        dcutoff=0.25Aa before (the default value for other materials remain
        dcutoff=0.1Aa).
      * Add ncrystal_hfg2ncmat script for convenient production of NCMAT files
        for hydrogen rich amorphous materials, based on concept and data from
        Romanelli et. al. (2021), https://dx.doi.org/10.1088/1361-648X/abfc13
      * Add ncrystal_onlinedb2ncmat script for creating materials with crystal
        structure taken from online databases (materialprojects.org or
        Crystallography Open Database). The script also has options for copying
        material dynamics from existing .ncmat files, and for verifying that
        NCMAT files containing comments containing specific links to entries in
        these databases, actually contain those structure (by overlaying Bragg
        cross section curves). Internally the script uses the Python Materials
        Genomics (pymatgen) Python module to access the online databases.
      * Add ncrystal_verifyatompos script for verifying that the list of
        @ATOMPOSITIONS in an .ncmat file is consistent with the indicated
        @SPACEGROUP number. Internally the script uses the Atomic Simulation
        Environment (ASE) Python module to handle the symmetries.
      * Add 16 new amorphous materials. Most are created with the
        ncrystal_hfg2ncmat script: Epoxy_Araldite506_C18H20O3 Kapton_C22H10N2O5
        Nylon11_C11H21NO Nylon12_C12H23NO Nylon610_C16H30N2O2
        Nylon66or6_C12H22N2O2 PEEK_C19H12O3 Polycarbonate_C16O3H14
        Polyester_C10H8O4 Polylactide_C3H4O2 Polypropylene_C3H6 Polystyrene_C8H8
        PVC_C2H3Cl Rubber_C5H8. The material Polyethylene_CH2.ncmat is not
        created with the ncrystal_hfg2ncmat script (although it could have
        been), but is instead based on phonon DOS curves provided by Kemal Ramic
        from the ESS Spallation Physics group, (based on modelling with Phonopy,
        VASP and OCLIMAX software). Finally, AcrylicGlass_C5O2H8.ncmat
        (a.k.a. "Plexiglass" or "Lucite"), is based on a VDOS curve used in
        ENDF/B-VIII.0 due to Petersen, et. al. (NCSU)., as it gave a slightly
        better fit to experimental data.
      * Add CaSiO3_sg2_Wollastonite.ncmat. The main motivation for adding this
        is that our library should contain at least one triclinic material, to
        make sure we exercise all code paths. For now, the material does not
        have VDOS curves and the dynamics are described via the Debye model.
      * Thanks to work by K. Ramic and J. I. Marquez Damian from the ESS
        Spallation Physics group under the HighNESS EU project, a massive number
        (64) of new crystalline materials are added to the data library, which
        along with the 16 new amorphous materials brings the number of material
        files shipped with NCrystal to a whooping total of 131!  Additionally,
        VDOS curves were added for 4 existing materials so that now only two
        crystalline materials (gamma-calcium and wollastonite) in the entire
        library have dynamics based purely on the less accurate Debye model. The
        64 new crystalline materials are mostly polyatomic materials of interest
        to the wider community of nuclear scientists, neutronics, and neutron
        scattering instrumentation. All these materials come fully described
        with both crystal structures and phonon VDOS curves, and have been
        carefully validated. Crystal structures are mostly taken from the
        Crystallography Open Database and the Materials Project, with most
        structures verified by verification against multiple DB entries (using
        the ncrystal_onlinedb2ncmat script). The VDOS curves are mostly based on
        calculations with Phonopy and OCLIMAX software with input files from the
        Phonon database at Kyoto university
        (http://phonondb.mtl.kyoto-u.ac.jp/).  VDOS curves for two materials,
        magnesium hydride (MgH2) and magnesium deuteride (MgD2), were provided
        by Davide Campi and Marco Bernasconi from University of Milano-Bicocca,
        based on DFPT calculations with the Quantum-ESPRESSO package.  All
        materials were validated against total cross section curves where
        available, VDOS curves were verified to be in agreement with
        experimentally measured heat capacity curves, predicted material
        densities were verified to be in accordance with known densities, and
        crystal symmetries were verified with the ncrystal_verifyatompos script.
      * Files for a few existing materials (SiLu2O5, UO2, SiO2) were renamed for
        consistency, and Crystallography Open Database reference numbers were
        added for all existing crystalline files.
      * Fixed bug where some hkl planes slightly above the d-spacing cutoff were
        left out of the final calculated list. Cubic, tetragonal, and
        orthorombic space groups were not affected by this bug, and since our
        default d-spacing cut off value is rather low, the effect was in any
        case unlikely to have affected many users.
      * Fixed bug where some hkl planes could very rarely be left out in
        crystals containing elements with negative coherent scattering
        lengths. It is not believed to be likely that this bug has affected any
        users, but a few of the new materials would have been affected.
      * Updated calcium scattering lengths due to improved measurements,
        Palomino et al. (2019), DOI:10.1016/j.nima.2019.02.072. This was
        implemented as a modification of b_coh for natural Ca and the dominant
        isotope Ca40, increasing the values of both b_coh parameters by roughly
        23%.
      * NCrystal.directMultiCreate now works with NCrystal.TextData objects.
      * Improve wrapping and indentation in code produced by the Python
        formatVectorForNCMAT utility function.
      * Add --stdout option to ncrystal_vdos2ncmat for easier usage in scripts.
      * Small fix for NCMAT parser which previously allowed multiple entries in
        @DENSITY section.
      * Add unofficial mechanism for removing certain contributions when
        expanding a VDOS to a scattering kernel. This is primarily intended to
        allow people working on plugins with detailed coherent physics to avoid
        double-counting by making it possible to exclude components otherwise
        created based on the incoherent approximation. The mechanism is
        implemented via the @CUSTOM_ support in NCMAT data, and it should be
        understood that files with such sections are always considered "under
        development", and unlikely to be suitable for general usage.
      * Expose VDOS analysis utilities from NCVDOSEval.hh to Python and C
        interfaces. In Python this is available both as a global analyseVDOS
        function, and as a member function in DI_VDOS objects. Return value is a
        dictionary with quantities derived from the VDOS curve: Debye
        temperature, mean-squared-displacements, effective temperature, etc.
      * Rename Python crossSectionNonOriented method to crossSectionIsotropic
        for consistency, keeping crossSectionNonOriented as a backwards
        compatible alias.
      * Fix python NCrystal.atomDB method where some situations of querying for
        unknown isotopes would always result in an exception even when called
        with throwOnErrors=False (github issue #65).
      * Type safe VDOSEval::elementMassAMU in C++.
      * Conversion from shared_obj<T> to pointer T* in C++ is made explicit to
        avoid footgun "const Info* info = NC::createInfo(..)". The footgun is
        still available for references (const Info& info = ...), but that is a
        necessary tradeoff since the implicit conversion to a reference is so
        useful when passing shared objects to functions taking reference
        parameters.

v2.6.1 2021-05-03
      * Workaround for numerical artifacts in inelastic cross sections arising
        from low-granularity beta-grids. In practice this affected materials
        which either used hard-coded low-granularity scattering kernel, or those
        with phonon VDOS curves that had large empty regions (often the case for
        hydrogen VDOS curves in materials with C-H bounds). The workaround for
        external low-granularity kernels consists of a simple "thickening" of
        the beta-grid, while the workaround for VDOS-based kernels consists in
        modifying the beta-grid generation code to ensure points are inserted
        into any large gaps in the single-phonon region.

v2.6.0 2021-04-05
      * Introduce NCMAT v4 which introduces a few new features, as described in
        detail in the ncmat_doc.md file. The most important new feature relies
        on the fact that it is possible to estimate atomic
        mean-squared-displacements directly from a phonon density of state
        distribution (VDOS), meaning that it is now allowed (and actually
        recommended) to skip @DEBYETEMPERATURE's for crystalline materials where
        atoms have VDOS curves available). This is not only useful, but improves
        low-temperature reliability of the Debye-Waller factors used in the
        elastic scattering components. This development was helped by Kemal
        Ramic and Jose Ignacio Marquez Damian from the ESS Spallation Physics
        group, and more details are available in the ncrystal GitHub issue
        #58. The other new feature in NCMAT v4 is a new optional syntax for the
        @CELL section, which allows more compact specification for certain
        crystal systems (in particular cubic systems).
      * The vast majority of crystalline materials in the NCrystal data library
        were updated to take advantage of the new features, and in addition the
        number of VDOS curves was increased dramatically, meaning that the
        dynamics in almost all crystalline materials are now described using
        actual VDOS curves rather than just Debye temperatures.
      * Missing VDOS curves for all elements in beryllium-fluoride (BeF2),
        aragonite (CaCO3), cuprite (Cu2O), periclase (MgO), and quartz (SiO2)
        were provided by Kemal Ramic from the ESS Spallation Physics group,
        based on modelling with Phonopy, VASP and OCLIMAX software.
      * Missing VDOS curves for 14 mono-atomic materials were added (fixing
        github issue #39): Au, Ba, Cr, Mg, Mo, Na, Nb, Pd, Pt, Rb, Sc, Sn, Y,
        and Zn. Refer to comments in the individual NCMAT files for references
        concerning the origin of these curves.
      * Missing VDOS curves were added to uranium oxide (UO2), taken from
        DOI:10.1103/PhysRevB.102.134312. Note that UO2 NCMAT files with VDOS
        tuned for specific temperatures are available (refer to comments in the
        UO2 NCMAT file for details).
      * All existing files with VDOS have been update to NCMAT v4, and had
        @DEBYETEMPERATURE sections removed in order to benefit from the better
        performance of getting the atomic displacements from the VDOS.
      * Added new material: gamma-Iron (Fe_sg225_Iron-gamma.ncmat) with VDOS
        curve. At the same time, the not particularly useful beta-Iron file was
        removed, so the NCrystal data library now contains both alpha- and
        gamma-iron, both with VDOS curves.
      * Expose debyeIsotropicMSD and debyeTempFromIsotropicMSD functions from
        NCDebyeMSD.hh header to C and Python, allowing conversions between
        atomic mean-squared-displacement and Debye temperature.
      * Improve robustness of the function debyeTempFromIsotropicMSD. Previously
        the internally deployed root-finding algorithm could fail under some
        scenarios (e.g. non-ultra cold rubidium).
      * Various improvements for ncrystal_vdos2ncmat.
      * Compilation fix for pre-Catalina OSX where the aligned_alloc function is
        apparently missing.
      * Fix bug where early calls to removeAllDataSources could get reverted by
        subsequent plugin loading.
      * Before defining macros in public headers, check if present already and
        if so undefine.

v2.5.81 2021-04-01
      * RC2 for incoming release v2.6.0.

v2.5.80 2021-03-25
      * RC1 for incoming release v2.6.0.

v2.5.0 2021-03-21
      * The NCrystal v2.5.0 release brings major technical improvements to
        NCrystal "under the hood", with around 15.000 lines of code changed in
        more than 200 files! Although massive, the changes are mostly technical
        in nature and mainly affects the C++ API, so people using NCrystal in
        another way (e.g. through Python, C, Geant4, McStas, ANTS2, ...), should
        still find that everything keeps working as before - although perhaps
        with slightly improved performance and with a few new useful function
        having been added to the Python/C API's.
      * The list of specific changes is too long to document here, but we
        have tried to provide a high-level overview of the developments
        in the wiki, so please find additional information at:
        https://github.com/mctools/ncrystal/wiki/Announcement-Release2.5.0

v2.4.81 2021-03-16
      * RC2 for incoming release v2.5.0.

v2.4.80 2021-03-15
      * RC1 for incoming release v2.5.0.

v2.4.0 2020-12-17
      * Properly support delivery of NCMAT data files from plugins. For clarity,
        and to avoid name clashes, such data files must follow a clear naming
        scheme "ncplugin-<PLUGINNAME>_*.ncmat". The mechanism for making the
        data files available will depend on how the plugin is used, and is of
        course automatic. For a static plugin, built into a given NCrystal
        installation, the data file will be installed along with all of
        NCrystal's own data files (either as physical files or compiled into the
        NCrystal library, depending on the NCrystal configuration). For a
        dynamic plugin, built separately and later loaded by adding the plugin's
        shared library to NCRYSTAL_PLUGIN_LIST, the data file will be compiled
        into the plugin library. Finally, for plugin developers, the data files
        will be made available via a symlink, ensuring that it is possible to
        directly edit the data file during development.
      * Support (on unix) relative paths in NCRYSTAL_PLUGIN_LIST.
      * Add option to ncrystal_ncmat2cpp for injecting include statements in
        generated code.
      * Plugin developers can now also see their test .ncmat files in the list
        provided by ncrystal_inspectfile --browse.
      * Export more information to client cmake projects.

v2.3.1 2020-12-17
      * Small bugfix fixing the loading of dynamic plugins.
      * Ensure plugins are loaded before file searching is carried out. This
        allows plugins to register in-memory files.

v2.3.0 2020-12-11
      * Replace span class with a fully home-grown solution. This means that all
        core NCrystal code is once more fully available under the primary
        license (the Apache 2.0 license).
      * Cfg parameters scatfactory and absnfactory now supports exclusion of
        specific named factories. This is used to (by default) exclude current
        factory from consideration when globalCreateScatter/Absorption is
        invoked from a factory (github issue #51).
      * Replace the '$(ncrystal-config --setup)' idiom with one which works also
        when entries in PATH contains spaces: 'eval $(ncrystal-config --setup)'
        (github isse #52).
      * Rename internal radius variable in McStas sample, to avoid clashes with
        component parameter (apparently this gave issues for McStas 3.0).

v2.2.2 2020-12-11
      * Bugfix for system-wide installations. Input files were
        accidentally opened in RW mode rather than just readonly, causing
        permission errors when user invoking NCrystal did not have
        write-permission to the input file. Thanks to the McStas crew for
        noticing this.

v2.2.1 2020-12-02
      * Bugfix restoring support for .nxs/.laz/.lau files which was accidentally
        disabled in v2.2.0.
      * Modify CMake code slightly to avoid warnings or unset CMAKE_BUILD_TYPE
        under certain conditions where NCrystal is included as a sub-project.

v2.2.0 2020-12-02
      * Support plugins (github issue #50)! Such plugins can be externally
        developed (for instance on GitHub), and either loaded dynamically at
        run-time by NCrystal or statically compiled into the NCrystal library
        (see the next item). How to use or develop plugins is described in
        more detail at https://github.com/mctools/ncrystal/wiki/Plugins .
      * Major revamp and modernisation of CMake configuration code. Client code
        can now find NCrystal via find_package statements (github issue #45),
        which makes an NCrystal::NCrystal library target available. Geant4-code
        needing the Geant4-NCrystal hooks must instead search for G4NCrystal
        with find_package and add a dependency on the G4NCrystal::G4NCrystal
        target. The new CMake code also facilitates static inclusion of
        externally developed plugins (if for some reason this is preferred to
        dynamic loading of the plugins). NCrystal now requires CMake 3.10.
        A side-effect of the modernised CMake code is that installation
        directories are affected somewhat.
      * Add ncrystal-config script which can be queried for details about a
        given installationn, or even used to setup the user environment by
        typing $(./path/to/ncrystal/installation/bin/ncrystal-config --setup) in
        a shell. The installed setup.sh file now simply wraps such a call.
        A side-effect of the work on ncrystal-config and CMake is that NCrystal
        installations are in general more relocatable and should continue to
        function even if the installation directory is moved.
      * Officially stop support of Python2 (github issue #48). Python2 (and many
        major Python projects) had EOL January 2020, and PyPI (aka pip install)
        has announced end of support for Python2 by January 2021. Users needing
        to use NCrystal with Python2 will have to stick with NCrystal release
        v2.1.1.
      * Add support for NCRYSTAL_DATA_PATH variable, checked after
        NCRYSTAL_DATADIR env. var and NCRYSTAL_DATADIR preprocessor definition
        (github issue #49).
      * Add ability to browse a list of available .ncmat files (of course only
        those in the current working directory or search path). Can for instance
        be accessed via "ncrystal_inspectfile --browse" (github issue #5). The
        same tool can also --extract content of available files, and display
        available plugins with --plugins.
      * Fix Python interface cleanup-order bug where __del__ method could refer
        to an already removed ctypes module at programme shutdown.
      * Fix silly bug where W was left out of the alphabet in a few places.
      * Fix big affecting printing of MatCfg objects with in-memory datafiles.
      * Remove obsolete functions naturalElementDB and naturalElementDBByName
        from Python interface (use the atomDB function instead), and update the
        ncrystal_endf2ncmat script to work with new atomDB interface.
      * Modernise C++ code, preventing raw new/delete usage and associated
        memory management woes. RCHolder objects can be automatically converted
        when possible (to const ptr or base class ptr), and add makeRC / castRC
        / static_castRC functions, similar to std::make_unique and
        std::make_shared, but for instantiating classes derived from RCBase
        directly into an RCHolder object. Apply new features throughout
        codebase, and fix a slew of compilation warnings (mostly for clang12 or
        gcc 4.8).
      * Add DynLoader utility class for dynamic loading of shared libraries and
        access to contained functions (used for dynamic plugin loading).
      * Add common utilities for reading environment variables.
      * Add ncglob function for file globbing.
      * Add FactoryBase::combineScatterObjects helper method to encapsulate
        the workarounds still needed in custom scatter factories.

v2.1.1 2020-11-07
      * Reverse minor change in NCDefs.hh, which meant that client code which
        was including public NCrystal C++ headers was prevented from using constants
        like M_PI and friends (of course, usage of such constants is not
        portable, and thus not really recommended). Release 2.1.1 is otherwise
        completely identical to release 2.1.0.

v2.1.0 2020-11-06
      * Introduce the NCMAT v3 format, which supports @ATOMDB (github issue #21)
        and @CUSTOM_ sections (github issue #42). As usual, the file
        docs/ncmat_doc.md documents the new format in detail.
      * With the @ATOMDB section (and the related configuration parameter named
        atomdb, documented in NCMatCfg.hh), it is now possible to configure
        atomic compositions in a highly flexible manner. Thus, both enriched
        elements and impurities/dopants are now supported, and physics constants
        such as neutron scattering lengths can be overridden from the data file
        -- but most users will likely find plenty of materials in the internal
        database. This internal database has now been expanded with 261 isotopes
        in addition to the natural elements previously included (note that the
        constants for some natural elements have received slight tweaks for
        consistency).
      * The @CUSTOM sections provides a way to embed custom data in NCMAT files
        and having it end up on the Info object (accessible from
        C/C++/Python). This is mainly intended for people working on extending
        NCrystal with custom or experimental physics models, who need a
        mechanism for providing the extra data.  Note that by default NCrystal
        emits warnings when loading NCMAT files with @CUSTOM_ sections. This is
        because files with such sections are by their very nature associated
        with custom code development, and therefore can not be expected to
        provide consistent physics if shared with a wider community.  It is
        possible to silence these warnings, via the C++ API or an environment
        variable.
      * Support for using in-memory NCMAT data (github issue #20). This makes
        it possible to initialise materials from dynamic string-data in
        C++/Python/C, rather than needing on-disk files.
      * Add script ncrystal_ncmat2cpp which can be used to convert NCMAT files
        to C++ code which embeds the data directly in a binary, using the the
        new feature for in-memory NCMAT data.
      * Add CMake installation option for embedding the shipped .ncmat data
        files into the NCrystal shared library. Enable via CMake configuration
        flag -DEMBED_DATA=ON. The files are still accessible via their usual
        names in NCrystal cfg strings, e.g. createScatter("Al_sg225.ncmat") will
        still work, even if there will be no physical Al_sg225.ncmat file
        installed.
      * Added advanced C++ example which shows how it is possible to customize
        NCrystal and add a new physics model. The example also use the new
        features for in-memory files and @CUSTOM_ section in .ncmat files.
        The example is in the file examples/ncrystal_example_customphysics.cc
        (which compiles to the executable bin/ncrystal_example_customphysics
        when installing via CMake).
      * The internal database of elements and isotopes is now accessible also in
        the C and Python interfaces. In Python, one can iterate through all
        entries using the new module-level function named iterateAtomDB, or
        extract specific entries using the new module-level function named
        atomDB. The latter accepts either Z and A values, or strings such as
        "Al", "He3", etc.
      * The Python-API is made slightly more convenient, by equipping the
        Process and Scatter classes with new methods named xsect and
        genscat. These methods optionally accepts neutron wavelengths (by the wl
        parameter), and will delegate the calls to oriented and non-oriented
        functions depending on whether the parameter named direction was provided.
      * Various other improvements were done for the Python interface as well:
        AtomInfo, Compositions, and all Debye temperature methods are now
        accessible as well, there are now functionality for efficiently
        generating large amount of oriented scatterings, and finally all classes
        and methods now have proper doc-strings.
      * Various improvements were carried out in the C++ code: removed obsolete
        __cplusplus >= 201103L precompiler checks, added more string utility
        functions, made wl2ekin and ekin2wl constexpr, made createInfo(..)
        multi-thread safe, added NCRYSTAL_API macro to new classes and functions
        (including those introduced in v2.0.0). Remove custom UniquePtr class in
        favour of std::unique_ptr from C++11. Improve RCHolder class (*,->
        overloading, move semantics).
      * Fix radian->degree conversion in the plot of scatter angles versus
        wavelength provided by ncrystal_inspectfile (it was multiplying by 57
        instead of 180/pi which is 0.5% off).
      * Support the rare use-case of wanting HKL structure factors calculated
        with Debye-Waller factors forced to unity (github issue #22). Setting
        the NCRYSTAL_FILLHKL_FORCEUNITDEBYEWALLERFACTOR environment variable
        will now result in just this.
      * Parsing of NCMAT v1 files now more strictly adheres to the
        specification, and does not allow comments except as full-line comments
        before the first data section.
      * Info dumps will now detect and display simple fractions in atomic
        positions (e.g. "1/3" instead of "0.666667"). This only happens for
        denominators <=64 and for exact matches at full double-precision.
      * Internal NCrystal header files are now exposed in the internal sub-
        directory (github issue #46) with include statements like #include
        "NCrystal/internal/NCMath.hh" . These are meant to facilitate
        experimental development of new features, and are not guaranteed to
        present a stable API. They also do not in general contain NCRYSTAL_API
        statements needed for symbol visibility in certain builds.

v2.0.0 2020-08-13
      * Milestone 2.0.0 release!
      * The main feature of this release is that it finally brings proper
        state-of-the-art inelastic physics to NCrystal. Additionally, also
        incoherent-elastic scattering was cleaned up and made self-consistent,
        so NCrystal v2.0.0 can truly be said to be the release in which NCrystal
        finally moves beyond its former focus on Bragg diffraction. Additionally
        and also for the first time, NCrystal can now support the modelling of
        certain non-crystalline materials (e.g. liquids).
      * The actual changes between NCrystal v1.0.0 and v2.0.0 took place in ~500
        commits touching several hundreds of files, and are far too vast to list
        here in detail. However, the release announcement page for NCrystal
        2.0.0 attempts to provide a short summary of user-facing changes
        concerning physics capabilities, the materials data library, and the
        syntax of NCrystal configuration strings:
        https://github.com/mctools/ncrystal/wiki/Announcement-Release2.0.0
      * A few technical details warrent highlighting here though as they
        potentially impact deployment: Firstly, NCrystal now requires C++11 or
        later for compilation, the older standards C++98 and C++03 are no longer
        supported. Secondly, in order to support C++11, the standard NCrystal
        CMakeLists.txt file now requires CMake 3.1.0 or later (as opposed to the
        previous requirement of CMake 3.0.0). Finally, in order to backport
        std::span from C++20 for internal usage in NCrystal code, a single file
        (NCSpan.hh) is for now covered under the Boost Software License (the
        license text is included in the file NCSpan.hh itself). This license is
        even more permissive than the Apache 2.0 license which still covers the
        rest of the core NCrystal code, and as such this should not add any new
        licensing burdens to users.

v1.99.1 2020-08-11
      * Release candidate number 1 for the new major release v2.0.0. The only
        difference between this release and the release v2.0.0 is in a few
        pieces of documentation and changed wording in an error message. Thus,
        for all practical purposes it is identical to release v2.0.0.

v1.0.0 2019-07-05
      * Milestone 1.0.0 release!
      * Introduce new optional modelling of more realistic energy transfers in
        inelastic scattering which can be enabled by setting
        bkgd=phonondebye:modeldeltae. This feature is still to be considered
        highly experimental and unvalidated.
      * C-buffer length fixes for gcc 8.2.0.
      * Prefer importlib.import_module rather than exec in ncrystal_inspectfile

v0.9.19 2019-02-19
      * Default value of sccutoff parameter is moved slightly down, from 0.5Aa
        to 0.4Aa and to make the effect less complicated, fsquared values are no
        longer taken into account. The effect of the change is very small,
        influencing the single crystal mosaicity distributions slightly for
        neutrons with wavelengths below 1Aa.

v0.9.18 2019-01-24
      * SPACEGROUP field is now optional in .ncmat files (but is always provided
        in all official ones).
      * The loadNCMAT function now accepts std::string for convenience.

v0.9.17 2019-01-11
      * Change source for Quartz data file (results in small numerical changes only).
      * Add new datafile for Sodalite.
      * Updated year in copyright notice in all files.

v0.9.16 2018-12-18
      * Fix small mistake preventing compilation of McStas component.

v0.9.15 2018-12-18
      * Add three new materials, expanding coverage of data library: Uraninite,
        Aragonite and SiLu2O5.
      * Lower default dcutoff value for most .ncmat files to 0.1Aa, in response
        to validation plots of Uraninite.
      * Updated year in copyright notice in all files.
      * Laz/Lau file loader now deals correctly with missing lattice_b/lattice_c
        parameters, and performs sanity checks in all cases.
      * Provide convenience functions/methods to Info objects, providing
        d-spacing of a given hkl value.
      * FillHKL use numerically stable summation when calculating structure factors.
      * Add NCrystal version printout from the NCrystal_sample component in
        McStas. Also sanity check the header file and library versions found.
      * Add NCDebyeMSD utilities for getting mean squared atomic displacements
        in the Debye model and use in .ncmat factory (to replace isotropic
        Debye/Glauber model everywhere for consistency).
      * For cosmetic reasons, avoid signed negative zero in atomic positions.
      * Minor updates to .ncmat files (mostly cosmetic or rounding of numbers).

v0.9.14 2018-10-17
      * Minor change in Geant4 interface to support Geant4 10.4 and later releases.

v0.9.13 2018-09-27
      * Stop shipping hand-written .nxs files. Instead, attempt to auto-generate
        .nxs files from all shipped .ncmat files, and where possible ship
        those. This is done in order to ensure greater consistency and a lower
        burden of maintenance moving forward. During this process a few issues
        were found with Wyckoff positions in a few .nxs files, which should now
        be fixed by the new procedure.
      * Changed dcutoff for crystals with complex unit cells (like yttrium
        oxide) from 0.4Aa to 0.25Aa when loading from .ncmat. For all other
        .ncmat files in our data library, the value stays at 0.15Aa.
      * Fix LCBragg error in case of the double degenerate case where both
        neutron and lcaxis are parallel to (0,0,1).
      * Faster fillHKL function, in particular for very small dcutoff values.
      * The fillHKL function now supports the NCRYSTAL_FILLHKL_IGNOREFSQCUT
        environment variable, which can be used to disable the fsquarecut entirely.
      * Increased numerical stability in structure factor summation during
        PCBragg initialisation.
      * Phase out usage of non-standard constants like M_PI.
      * Increase MC overlay safety factor slightly for LCBragg scatter generation.
      * Improve colors and linestyles in ncrystal_inspectfile script.
      * LCBraggRef models use numerically stable summation.

v0.9.12 2018-08-28
      * Fix GaussMos cache-polution bug, where planes with similar d-spacing but
        different Fsquared could get wrong Q-factor in single crystals with large
        mosaicity.
      * Fix uninitialised variable in NCMatCfg.cc, affecting only floating point
        parameters not supporting units.
      * Allow re-seeding of the RandXRSR fallback RNG.
      * Add McStas example instrument file.
      * More checks for invalid geometry parameters in the McStas component.

v0.9.11 2018-08-18
      * Mark API functions and classes in an attempt to support Windows DLL
        builds and UNIX builds with -fvisibility=hidden (see github issue #17).
      * Add missing include in NCPCBragg.cc.

v0.9.10 2018-08-15
      * This release represents a large amount of work in many areas, and in
        particular focusing on aspects of Bragg diffraction in single crystals.
      * The Gaussian mosaicity modelling code was completely reimplemented,
        going back to the fundamental equations and making sure everything is
        carried out in a consistent and precise manner. Cross-section evaluation
        automatically selects between efficient numerical Romberg integrations
        and evaluation of new improved closed-form expressions, which takes the
        relevant spherically geometric into account. The code thus now
        gracefully handles everything from backscattering to forward scattering
        scenarios, and a large dynamic range of mosaicities, covering at least a
        range of 0.01arcseconds to tens of degrees.  Despite this, computational
        speed is also improved, thanks to various enhancements and in particular
        faster code for searching through the available plane normals. A new
        parameter, mosprec, can be used to tune the tradeoff between precision
        and speed. The default value of 1e-3 is likely adequate for almost all
        users.
      * Introducing a completely new and precise modelling of layered crystals,
        with the LCBragg class. This class, primarily intended to be used for
        modelling of Pyrolythic Graphite, by default takes the rotational
        aspects of such crystal into account using an from-first-principles
        approach to figuring out which sets of normals will contribute to the
        scattering cross-sections, and then finding the exact contribution
        through efficient numerical integrals. A crystal is considered to be a
        layered crystal if it has the lcaxis vector parameter set. Additionally,
        the lcmode parameter can be used to select slower reference parameters,
        which simply sample an SCBragg single-crystal instance many times, in
        different rotated systems.
      * Huge speedup when modelling low-wavelength (<1Aa) neutrons in single
        crystals: Approximate the very large number of very weak and
        uninteresting scatter planes at 2*d-spacings < 1Aa with an isotropic
        mosaicity distribution, which is a rather appropriate approximation due
        to the very large number of very weak planes involved. The cutoff value
        of dspacing=0.5Aa can be modified with the new sccutoff parameter.
      * Bragg diffraction in powders and polycrystals was never slow, but the
        PCbragg class was nonetheless rewritten to become even faster by working
        on energies directly, avoiding internal conversions to wavelength at
        each call.
      * Embedded an NCRYSTALMATCFG[lcaxis=0,0,1] statement in the data file
        C_sg194_pyrolytic_graphite.ncmat, so that it will by default be modelled
        as a layered crystal with the correct rotation axis, which is likely
        what almost all users will want.
      * Calling generateScattering where cross-sections are vanishing now
        generally leads to nothing happening (i.e. scatter_angle = 0 and
        delta_ekin=0) rather than the previous fallback of isotropic elastic
        scattering.
      * Retire RandSimple and introduce instead RandXRSR implementing the
        xoroshiro128+ generator. This means that the fall-back RNG option
        shipped with NCrystal is now not only fast, but also fully suitable for
        scientific work.
      * Fix bug in the NCrystal McStas component which resulted in wrong
        attenuation factors being calculated for the default absorption mode.
      * Changed the conversion constants used in the McStas component in order
        to make unit conversion issues in the NCrystal-McStas interface less
        likely to produce undesired imprecision. Further discussions with McStas
        devs will be needed in order to completely address the issue.
      * Code creating lattice rotation matrices was updated in order to avoid
        small rounding errors in the generated normals in most cases.
      * More sanity checks and input pruning of atomic positions and lattice
        structure loaded from input files. Also improved robustness of .nxs file
        loading.
      * Remove spurious ref-count increase in CalcBase::setRandomGenerator
        (thanks to A. Morozov for the report).
      * Fix windows builds (see github issue #17). Many thanks to A. Morozov for
        a detailed report.
      * Many general infrastructure improvements: Refactor random sampling
        algorithms from CalcBase objects (a necessary step for future enhanced
        MT support). Refactor orientation code from SCBragg (now used by both
        SCBragg and LCBragg). Refactor .ncmat loading code to disentangle the
        parsing code from the code. Lots of new utilities: Cubic splines,
        Romberg integration, root finding, derivative estimation. Single header
        NCDefs.hh providing all ubiquitous definitions and infrastructure
        related to memory handling and error reporting, and also handles
        includes which must get special treatment due to differences between
        platforms or C++ versions. Add UniquePtr (similar to C++11's
        unique_ptr). Add PlaneProvider (supported in PCBragg, SCBragg, LCBragg),
        for customizing which planes goes in which process. Avoid bind1st usage
        as it is removed in C++17. Custom functions for evaluating sine and
        cosine of angles in various ranges - between 5-10 times faster than the
        standard functions at 15 significant digits. Also add fast
        approximations for atan and exp. Performed careful changes to ncmin and
        ncmax functions with large performance implications (surprisingly,
        std::fmax generates very inefficient machine code:
        https://godbolt.org/g/Dxy52A ). Add helper class for efficiently
        generating sine and cosine values for all grid-points in a grid. PCBragg
        supports more custom constructors.

v0.9.9 2018-03-26
      * Expose loadNCMAT function to client C++ code, making it possible to work
        with .ncmat files completely independent from the factory
        infrastructure.
      * Cleanup internal .ncmat loading code, splitting out fillHKL to a new
        utility file and renaming old loader code to be purely a parser.
      * Allow Wyckoff positions and mean-squared-displacement parameters on
        NCInfo.
      * Rename standard factories to stdnxs, stdncmat, stdlaz, stdscat and
        stdabs respectively.
      * Allow factory-specific parameters to be passed along to info
        factories. Thus, expandhkl which was until now a global NCMatCfg
        parameter, is now a parameter specific to the stdncmat factory.  The
        stdnxs factory gets two flags: mcstaslike and fixpolyatom. The former
        makes nxs-provided bkgd curves use same composition as in the McStas nxs
        component (the default is to compose as in NXSG4).
      * Remove polyatom fix in wrapped nxslib code, so bkgd=external xs curves
        for .nxs files are now completely as in NXSG4, apart from the fact that
        our version contains fixes for crashes (reported upstream).
      * Rename NonOrientedScatter to IsotropicScatter.
      * Fix compilation error under some compilers in NCRCBase.hh.

v0.9.8 2018-03-19
      * Decouple PhononDebye background cross-section code from .ncmat format,
        making it the new default also for .nxs files. For .nxs files, it is
        still possible to select the old behaviour of using curves provided by
        nxslib itself by configuring those files with "bkgd=external".  A side
        benefit of this splitout can in some cases result in significant faster
        initialisation times when working with .ncmat files.
      * Major cleanup and renaming of configuration variables, in preparation of
        the 1.0.0 release milestone. The renames are: "mosaicity"->"mos",
        "orientationprimary"->"dir1", "orientationsecondary"->"dir2",
        "orientationtolerance"->"dirtol", "packingfactor"->"packfact",
        "dcutoffupper"->"dcutoffup", "scatterbkgdmodel"->"bkgd",
        "scatterfactory"->"scatfactory, "absorptionfactory"->"absnfactory" and
        "skippbragg" becomes "bragg" but with opposite logic, e.g. "bragg=0"
        disables Bragg diffraction. Removed entirely was the variables
        braggonly (use "bkgd=none" or "bkgd=0" instead now) and
        "nphonon". Experts can now instead tune parameters for specific
        background models by embedding them in the "bkgd" parameter value
        itself, e.g. "bkgd=phonondebye:nphonon@20".
      * Temporarily disable single-crystal speedup code while we investigate
        further, due to concerns that it is too aggresive (see github issue #13).

v0.9.7 2018-03-06
      * Don't produce floating point exception when loading .ncmat files with H,
        Li, Ti, V or Mn (github issue #8).
      * McStas component now actually applies requested packingfactor for powder
        materials (github issue #9).

v0.9.6 2018-02-07
      * Fix visualisation of box-shaped sample in McStas component.
      * Rename McStas component parameter multscatt to multscat.

v0.9.5 2018-02-05
      * Add support for python3 (github issue #6).

v0.9.4 2018-01-31
      * Correct version numbers.

v0.9.3 2018-01-31
      * NCrystal_sample.comp (McStas interface): Fix link flags on OSX
        (github issue #3) and compilation warnings (github issue #4).

v0.9.2 2018-01-23
      * Fix cross-section and material composition bugs in Geant4
        interface. Only affects Geant4 users.
      * CMakeLists.txt: Always use absolute rpath (needed on OSX).
      * Fix gcc7 compilation.

v0.9.1 2017-08-30
      * Proper behaviour of python interface if Numpy is absent.

v0.9.0 2017-08-30
      * First github based public release of NCrystal.
