(release-notes)=

# Release Notes

## v0.4.0 (Unreleased)

## v0.3.0 (25 June 2025)

Thanks to the contributors to this release: Benoît Bovy, Adrien Delsalle,
and Wolfgang Bangerth.

### Highlights

- A new HEALPix global grid for running simulations at the planetary scale (not
  available on Windows, for pre-built Python packages only available on
  conda-forge).
- Flow kernels: user-defined functions as an easy, flexible and efficient way to
  implement custom algorithms to apply on the flow graph. It is currently available
  for the Python bindings only. User-defined function, written in pure-Python, are
  jit-compiled with Numba, therefore making their application very fast.

### Breaking changes

- Updated the minimum required version of ``xtensor`` and ``xtensor-python`` to
  0.26 and 0.28, respectively ({pull}`174`, {pull}`182`). Upstream changes in
  the header file structure make older versions incompatible.
- Refactored the C++ flexible container system so alternative container libraries
  like Eigen may be used instead of Xtensor ({pull}`155`). Note that this is work
  in progress, Fastscapelib still heavily relies on Xtensor. As a consequence, the
  the C++ grid class templates have been slightly refactored and the type aliases
  have been removed.

### New features

- Added flow kernels and parallel execution of kernels and/or flow routers
  ({pull}`157`, {pull}`185`, {pull}`186`, {pull}`187`).
- Added full fletched Pixi configuration for installing the dependencies,
  building and installing the library as well as running common development
  tasks (tests, documentation, etc.) on all major platforms (Linux, MacOS and
  Windows) ({pull}`170`).
- Added HEALPix grid ({pull}`162`).

## v0.2.2 (5 June 2024)

### Bug fixes

- Fixed and refactored some benchmarks ({pull}`152`).
- Fixed CMake installed C++ header files ({issue}`158`, {pull}`159`).

## v0.2.1 (20 October 2023)

### Bug fixes

- Fixed ``pflood_sink_resolver`` that was "flooding too much" ({issue}`145`,
  {pull}`146`).
- Fixed ``RuntimeWarning`` (invalid value in cast) issued when calling
  ``FlowGraph.basins()`` in Python ({pull}`147`).
- Python bindings: fixed Python interpreter crash (segfault) when accessing data
  members of the implementation of a graph snapshot via ``FlowGraph.impl()``
  ({pull}`148`).

## v0.2.0 (9 October 2023)

A complete re-write of Fastscapelib with brand new features and API (note: the
API of the previous version has been mostly removed). Some highlights:

- A flexible grid system, including 1D profile grid, 2D raster grid (with
  support of diagonals vs. non-diagonals connectivity) and 2D triangular mesh
- Full support of looped (periodic) boundary conditions for uniform rectangular
  grids
- Flexible flow routing using a "flow graph" and "flow operators"
- Support for both single direction and multiple direction flow (grid-agnostic
  implementation)
- Efficient resolution of closed depressions in the topography while routing the
  flow paths, based either on explicit computation of a graph of basins or on
  the priority flood algorithm (grid-agnostic implementation)
- The current flow graph implementation graph based on fixed-size arrays is
  extensible to alternative representations (e.g., sparse matrix, linked-lists)
- First-class C++ and Python APIs
- Detailed documentation (examples, user-guide, API reference, etc.)

Thanks to the contributors to this release: Benoît Bovy, Adrien Delsalle,
Guillaume Cordonnier and Hannah Sophia Davies. Thanks also to Johan Mabille and
QuantStack (https://quantstack.net/) for their contribution and advice on the
design of the library.

## v0.1.3 (5 November 2018)

- Update to xtensor latest version (0.18) ({issue}`35`).

## v0.1.2 (23 August 2018)

- More robust and cleaner version management ({issue}`32`).
- Support older versions of apple-clang ({issue}`30`, {issue}`33`).

## v0.1.1 (22 August 2018)

- Fix version ({issue}`29`).

## v0.1.0 (21 August 2018)

Initial release.