Skip to content

Latest commit

 

History

History
418 lines (379 loc) · 32.8 KB

CHANGELOG.md

File metadata and controls

418 lines (379 loc) · 32.8 KB

preCICE Change Log

All notable changes to this project will be documented in this file. For future plans, see our Roadmap.

2.3.0

  • Added isMeshConnectivityRequired(meshID) to the SolverInterface API. This is useful to generate connectivity information only if required by preCICE.
  • Added a configuration option to enable experiemental API functions.
  • Added bindings to doxygen.
  • Added check of user-written data to be finite.
  • Added convergence information for linear solvers of PETSc-based RBF mappings. Reports of converged solvers are logged as DEBUG, stopped solvers as WARNING and diverged solvers as ERROR messages.
  • Added experimental support for direct mesh access of meshes.
  • Added fmtlib for more readable and maintainable message formatting.
  • Added support for 3D meshes in ScaleByAreaAction.
  • Added the Rank of partitions in VTK exports.
  • Added the associated data name when printing convergence information.
  • Added the constraint scaled-consistent to mapping methods, which scales a consistent mapping such that the surface integral is equal on both sides of the interface.
  • Added the experimental API functions setMeshAccessRegion and getMeshVerticesAndIDs.
  • Added the xml configuration option '<use-mesh ... direct-access=true />' for this feature.
  • Adopted the Contributor Covenant code of conduct (see CODE_OF_CONDUCT.md).
  • Change m2n:mpi to use the more efficient single-ports implementation. To use the old implementation, use m2n:mpi-mulitple-ports.
  • Changed edge and triangle normals to be computed on-demand.
  • Changed formatting scripts to use git ls-files. This simplifies handling of edge cases.
  • Changed linear solvers of PETSc-based RBF mappings now issue a warning when they reach the maximum iterations. This used to result in an ERROR.
  • Changed the convergence measures INFO print to a fixed width scientific format. (issue #975)
  • Changed the default build type of the library from static to shared.
  • Changed the precice log files to print a fixed width scientific notation using 8 digits after the comma for floating point numbers and a fixed width for integer values. (issue #975)
  • Changed the watch-point calculation algorithm to use already built index trees.
  • Deprecated the flip-normals attribute of meshes. This is not functional anymore.
  • Deprecated vertex normals in the vertexCallback() of the python actions. preCICE will pass None if the normal parameter is defined.
  • Fixed FindNumPy using fallback of find_path.
  • Fixed FindPETSc to also find PETSc if the include directory is in CPATH, which occured with some versions of pkg-config.
  • Fixed socket communication to fail without a network connection.
  • Fixed wrong error in tightly converging QN coupling. (Issue #976)
  • Implement a simplified interface to query index trees of meshes.
  • Improved memory consumption and data locality of vertices.
  • Improved mesh memory usage of vertices by ~41% and of connectivity by ~55%.
  • Improved usabilty and readability of logging macros and assertions.
  • Introduced package time and moved functionality related to extrapolation there.
  • Migrated PRECICE_XXX macros to fmtlib.
  • Modifying release PR templates according to restructured tutorials.
  • Removed limitation of multi-coupling scheme for coupling topologies without a central participant. A non-centric controlling participant still needs to run in serial.
  • Removed obsolete query package functionality. Interpolation is refactored into mapping package.
  • Removed the FASTEST fortran bindings.
  • Removed vertex normals from the vtk exporters.
  • Replaced prettyprint with fmtlib.

2.2.1

  • Fixed a bug leading to a freeze when using sync-mode with lazy indexing.
  • Fixed empty received partitions for filtering on slaves.
  • Fixed gather-scatter communication deadlock with empty master ranks.

2.2.0

  • Added a file sink to the test runner, which additionally writes the test log output to the file test.log.
  • Added a verbose log file test.debug.log to tests. Submitting this file alongside bug reports will significantly simplify debugging the tests.
  • Added check for user-defined python actions.
  • Added check to ensure advance() isn't called with invalid timestep size.
  • Added error message for incorrectly configured acceleration data in a serial implicit coupling scheme.
  • Added missing documentation to XML tags and attributes.
  • Added support for BUILD_TESTING, which may be used to disable the compilation and execution of the tests.
  • Added support for escaped characters in XML.
  • Added the build configuration (Release/Debug) to the preCICE startup statement and inform the user when Debug an Trace logs are not available.
  • Added watch-integral functionality to, for instance, compute total force, total stress, or flow rate at coupling mesh.
  • Changed an error to a warning when all sub-vectors of the residual-sum preconditioner are numerically zero.
  • Changed the CMake FindPETSc module to a robust wrapper based on pkg-config. The new method is more robust and simplifies the compilation of preCICE in e.g. SLURM jobs.
  • Changed the log format for tests. They now contain the participant name and are |-separated.
  • Clarified the wording of errors messages in data access functions. They now refer to vertex IDs instead of indices.
  • Deprecated API method getMeshIDs. Call getMeshID for specific mesh names instead. This method will be removed in a future release.
  • Extended configuration and repartitioning to allow the user to define multiple mappings from and/or to the same mesh.
  • Fixed MacOS compilation and test errors.
  • Fixed a bug in ReceivedPartition which led to problems when coupling multiple participants.
  • Fixed a bug in the configuration of Aitken underrelexation.
  • Fixed an issue when running the tests with Intel MPI.
  • Fixed an occasional issue solving the system matrix in PETSc-based RBF mappings.
  • Fixed boost log_level issues on MacOS.
  • Fixed compilation error emitted by intel compilers.
  • Fixed cryptic assertion for forgetting the max-iterations tag. Now max-iterations is enforced in the configuration.
  • Fixed indexing bug in solverdummies.
  • Fixed input checks for data access functions.
  • Fixed interleaved assertion output.
  • Fixed parameter type of operator"" in TestContext, which was a system-dependant error.
  • Fixed parsing error on systems without locales installed. This fixes issues when running preCICE in minimal docker containers.
  • Fixed syntax of the Fortran function precicef_get_mesh_vertex_size_, which lead to incorrect name de-mangling.
  • Fixed the data type and precision in exported VTK files.
  • Fixed two wrong assertions in QR factorization, which did not allow meshes with only a single partition.
  • Improved checks of configuration related to data access.
  • Improved compiler compatibility of assertion.
  • Improved the error message for not exchanging data over the same mesh used for convergence measures.
  • Increased the minimum required C++ version from 11 to 14. This was triggered by Boost.Geometry increasing their minimum version to C++14 in Boost 1.75.
  • Removed obsolete trigger-solver-plot config option.

2.1.1

  • Fixed a compilation error emitted by intel compilers when compiling tests.
  • Fixed an error message for unique acceleration subtags.
  • Fixed parsing error on systems without locales installed. This fixes issues when running preCICE in minimal docker containers.
  • Fixed preCICE erroneously expects cyclic communicator for IMVJ with restart.
  • Fixed system-dependent compilation error when compiling tests.
  • Updated the PKGBUILD for the Arch Linux User Repository to be compatible with preCICE v2.

2.1.0

  • Added MPI version detection and print it during the CMake configuration.
  • Added a new action to sum up data values, action:summation.
  • Added check for only allowing watchpoints on a provided meshes.
  • Added check for the coordinates of configured watchpoints. They have to match the dimensionality of the <solver-interface>.
  • Added check to prevent <use-mesh> from the same participant.
  • Added control flow checks to C bingings.
  • Added many tests for the communication abstraction.
  • Added option to make a convergence measure strict. It has to converge then and leads to a premature simulation stop if not.
  • Added parallel support for Eigen RBF mapping.
  • Added platform-specific defaults of the loopback interface name to the network attribute of socket connections.
  • Added reset of written data 0 in advance() to simplify detection of missing write data calls.
  • Added several checks to prevent that false API usage leads to complicated quasi-Newton assertions.
  • Added support for planar quads based on triangulation, needed, for instance for nearest-projection mapping. Quads will be split along the shortest diagonal to minimize the mapping error. Note that this triangulation will be visible in configured mesh exports.
  • Added tests which build solverdummies and tests which run them in various configurations.
  • Added tooling to prevent changelog conflicts.
  • Added warning when configuring m2n as mpi when preCICE is compiled with OpenMPI. This is known to cause connection issues.
  • Allowed multiple convergence measures of same coupling data.
  • Changed com::MPIDirectcommunication to work only for Master-Slave connections.
  • Changed test setup to a simpler and consistent version using the new testing::TestContext. Tests now require to run on 4 MPI ranks. They will still compile but not run when MPICommunication=OFF.
  • Changed the CMake FindNumPy module to only consider information based on the selected python interpreter.
  • Changed the minimum required PETSc version to 3.12, which delivers consistent results across platforms.
  • Disabled tests based on MPIPorts and MPISinglePorts when using Open MPI.
  • Enabled RBF-based tests in partiton unit-tests and serial integration tests.
  • Extended iteration logging by total and dropped quasi-Newton columns.
  • Extended title headers of convergence and iteration files by measure abbreviation.
  • Fixed MPIPorts and MPISinglePorts not always closing ports.
  • Fixed SocketCommunication setting up a port and writing connection info even if there are no requesters.
  • Fixed compatibility with Boost 1.73.0.
  • Fixed memory leaks and hanging communication when not calling precice::SolverInterface::finalize().
  • Fixed memory leaks in C bindings.
  • Fixed memory leaks in Fortran bindings in case of missing call toX_finalize().
  • Fixed memory leaks when using petrbf mappings in some conditions.
  • Fixed occasional errors when using PETRBF with a preCICE-managed MPI Communicator.
  • Fixed over-subscription error when executing the tests with recent versions of Open MPI on less than 4 physical cores.
  • Fixed silent truncation of numeric values in the config with C locales using decimal comma. The parser now always uses the locale en_US.UTF-8.
  • Fixed target test_install requiring CMake version 1.13.
  • Fixed value semantics of precice::SolverInterface.
  • Improved all error messages.
  • Improved readability of relative convergence measure INFO logs.
  • Refactored com::Communication handling of rank adjustments.
  • Refactored cplscheme::BaseCouplingScheme and derived classes. Introduce cplscheme::BiCouplingScheme.
  • Refactored mesh::BoundingBox into seperate class.
  • Removed m2n:mpi-single, which never worked outside tests.
  • Removed convergence file logging for min-iterations convergence measure.
  • Removed deprecated and untested Manifold Mapping. API functions hasToEvaluateSurrogateModel and hasToEvaluateFineModel remain as nop stubs.
  • Removed logging of average convergence rates.
  • Removed support for CPP and CPPFLAGS environment variables.
  • Renamed a few functions of the Fortran bindings for consistency with the C++ API. This does not break backwards compatibility, but the previous functions are now considered deprecated and will be removed in v3.
  • Removes the summary of event timing when preCICE finalizes and writes it to the file precice-PARTICIPANT-events-summary.log instead.
  • Restricted the configuration of WatchPoints to provided meshes.
  • Split multi-setup integration tests into multiple single-setup tests.

2.0.2

  • Fixed a critical bug in the testing framework.
  • Fixed a critical bug in the partitioning for geometric filter set to on-master in <use-mesh> tags. The default configuration is on-slave.

2.0.1

  • Fixed broken pkg-config file in some cases due to CMake GNUInstallDirs.
  • Fixed system-dependent error when displaying mapping distance information for empty partitions.
  • Improved RBF error messages by clarifying them and giving a hint to the user on what to do next.

2.0.0

  • Added CMake build type fallback to Debug in case it wasn't provided.
  • Added CMake check for C++11 library conformance. This is especially helpful when using Intel Compilers.
  • Added CMake options to enable native bindings PRECICE_ENABLE_C, PRECICE_ENABLE_FORTRAN (on by default).
  • Added examples/ to installation and package
  • Added a generator for markdown references binprecice md.
  • Added caching to the CMake library validation
  • Added directional directory level for file-based connection exchange. For each connection, there is now a directory in precice-run of the form Accepter-Requestor.
  • Added distance statistics of nearest-neighbour and nearest-projection mappings between mesh pairs as debug output.
  • Added grouped tests by module to CTest.
  • Added information to the log of the first written Data values.
  • Added log statements to the connection information file writers and listeners including full paths.
  • Added step to remove the connection directories after connecting the slaves. precice-run will be empty after a successful run.
  • Added support for python 3 in python actions.
  • Added the mesh name to the information used to generate connection information files, which is required for the two-level initialization.
  • Changed CMake to always validate dependencies. Set PRECICE_ALWAYS_VALIDATE_LIBS=NO to disable this behaviour.
  • Changed the internal handling of meshes by removing sub-meshes, the type hierarchy based on mesh::PropertyContainer, and the obsolete mesh::Group and mesh::Merge. This improves memory consumption, dramatically reduces allocations and improves locality when traversing primitives.
  • Changed unit tests to run only on MPI Rank 0.
  • Completely removed server mode. Now, the only supported parallelization concept is the peer-to-peer master-slave mode.
  • Disabled the installation of the test binary and files by default.
  • Dropped official python2 support for python bindings (remove tests).
  • Removed all experimental python bindings precice, precice-future, PySolverInterface.
  • Fixed a bug in the XML parser which did not correctly checked tag occurrence.
  • Fixed a bug in the XML parser which lead to ignored error messages from libxml2.
  • Fixed the Debian package generation by using GNUInstallDirs, providing a correct changelog and SOVERSION, as well as generating a package name including the SOVERSION.
  • Improved efficiency of nearest projection mapping of matching meshes using lazy generation of index trees.
  • Introduced preCICE-MATLAB bindings (precice#494, precice#580) and provided them in precice/matlab-bindings.
  • Merged the SolverInterface::configure() into the SolverInterface constructors. They now have a second parameter for the configuration file.
  • Moved Fortan 2003 bindings (src/precice/bindings/f2003) and solverdummy (tools/solverdummy/f2003) to a separate repository.
  • Refactored and made two-level initialization configurable through "use-two-level-init" in m2n.
  • Refactored the XML documentation generation out of the xml::XMLAttribute and xml::XMLTag classes into xml/Printer.[c/h]pp.
  • Released finalized version of python bindings in independent repository: precice/python-bindings. Package is named pyprecice and supports the preCICE version >= 2.0.0.
  • Removed MeshHandle from API and replace use in integration tests by SolverInterfaceImpl::mesh().
  • Removed an unnecessary assertion in getMeshVertexIDsFromPositions().
  • Removed deprecated SCons.
  • Removed deprecated HierarchicalAitkenAcceleration.
  • Removed deprecated ModifyCoordinatesAction.
  • Removed deprecated voxel queries in src/query/.
  • Removed packaging files specific to Ubuntu 18.04 as it is covered by CPack.
  • Renamed CMake variables MPI, PETSC, PYTHON to PRECICE_MPICommunication, PRECICE_PETScMapping, PRECICE_PythonActions
  • Replaced geometric filter option "filter-first" and "broadcast-filter" by "on-master" and "on-slaves", respectively, to generalize to two-level initialization.
  • Restructured tools and bindings:
    • Moved developer tools to tools/.
    • Moved user tools to extras/.
    • Moved native bindings to extras/bindings/.
  • Simplify parallel configuration:
    • Automatically add master:mpi-single for parallel participant if necessary.
    • No longer require gather-scatter distribution type for a m2n with at least one serial participant.
    • Automatically choose suitable RBF implementation based on whether preCICE was built with PETSc and whether the participant is serial or parallel.
  • Sorted out the different meaning of timestep and time window:
    • Renamed API function isTimestepComplete to isTimeWindowComplete.
    • Renamed C bindings function precicec_isCouplingTimestepComplete to precicec_isTimeWindowComplete.
    • Renamed cplscheme configuration option timestep-length to time-window-size.
    • Renamed cplscheme configuration option max-timesteps to max-time-windows.
    • Renamed acceleration configuration option timesteps-reused to time-windows-reused.
    • Renamed acceleration configuration option reused-timesteps-at-restart to reused-time-windows-at-restart.
    • Renamed export configuration option timestep-interval to every-n-time-windows.

1.6.1

  • Fixed a platform-dependent issue with boost.

1.6.0

  • Added CMake target to uninstall the project.
  • Added INFO log output of the connection build-up during the instantiation. This dramatically simplifies the identification of connection-related problems.
  • Added precice::getVersionInformation() to the interface. This returns a semicolon separated list of information containing the version, configuration, compiler and flags used to compile the library.
  • Added additional checks for wrong user input in SolverInterface.
  • Added basic index validation of user input to the SolverInterface
  • Added events for mapping internals such as the computation of the mapping and the index-generation.
  • Added more "pythonic" python bindings precice_future. More: here and here
  • Added overrides for dependencies, allowing users to force CMake to use a given version. Read more
  • Added support for the CPP and CPPFLAGS environment variables. Read more
  • Added support for the xsdk default mode. Read more
  • Added the CMake variable PRECICE_CTEST_MPI_FLAGS which can be used to pass additional flags to mpirun. A common use-case is to enable oversubscribing on machines with only 2 cores by passing --oversubscribe.
  • Added the possibility to pass a custom MPI_Comm to preCICE via a new SolverInterface constructor. The passed communicator is then used as the global internal communicator. The user has to ensure that the mpi implementations of preCICE and the caller code are consistent and compatible.
  • Added the result of git describe --tags --dirty to the library, which is now displayed during the configuration of preCICE. This allows quickly check what commit you are actually using and whether there were local changes.
  • Added tests for python bindings using a mocked preCICE C++ Interface. Read more
  • Added validation of dependencies in the CMake script.
  • Changed the connection publishing to a hash-based approach. This method is faster, more robust and NFS-friendly. The files are rooted in the folder precice-run, deleting this folder resolves most connection problems.
  • Changed the errors in the configuration stage to throw std::runtime_error.
  • Changed the id management from std::set to boost::container::flat_set. This reduces the peak memory consumption of meshes by about 9%.
  • Changed the log output for filtered vertices. Log level INFO prints the total number of filtered vertices. The detailed information was moved to DEBUG.
  • Changed the scope of the generated version.[ch]pp files to precice/impl. This prevents collisions with other random version headers.
  • Changed and optimized the index generation for an individual speed up of 10-20x.
  • Changed and optimized some complexity issues in the generation of communication maps for an individual speedup of around 1000x for bigger meshes.
  • Changed and optimized some complexity issues in NearestNeighbour::tagFirstRound leading to an individual speedup of around 100x.
  • Deprecated the python bindings precice, which will be removed in preCICE Version 2.0.0. If you still want to use them, please install precice and precice_future. Our recommendation, if you want to use the new bindings: Use import precice_future as precice.
  • Fixed a bug in the python bindings which ignored the memory layout of numpy array. We now use numpy.ascontiguousarray to guarantee a C-compatible layout of data structures. Read more
  • Fixed a major memory issue due to excessive logger instantiations. This reduced peak memory consumption and allocation count of meshes by 50%.
  • Fixed compatibility with Eigen versions >3.3.7.
  • Fixed macro namespace by using the PRECICE_ prefix. This prevents collisions with foreign macros.
  • Fixed the index-based version of the nearest projection mapping and reintegrated it.
  • Fixed wrong user input in Triangle creation to trigger assertions instead of a comprehensive user error.
  • Fixed a bug in MPI initialization when using PETRBF mappings.
  • Fixed a warning message in ReceivedPartition displaying a wrong vertex count.

1.5.2

  • Fixed faulty communication in master:sockets leading to crashes.

1.5.1

  • Fixed the exposure of boost::asio implementation details leading to version incompatibilities
  • Fixed the CMake linkage of Boost to become compatible with the generated CMake config of Boost 1.70.0

1.5.0

  • Added CMake alias precice::precice which mimics the namespaced library name after calling find_package(precice). This allows seamless use as a subproject.
  • Added state-awareness to the SolverInterface to keep track of the Mesh states. This results more informative error messages for mesh related functions.
  • Added faster PetRadialBasisFctMapping tagging using RTree queries.
  • Added potentially missing signal SIGXCPU.
  • Added sanitization of IDs to the SolverInterface and harden it against misconfiguration using a consistent mechanism to express requirements.
  • Added support for using preCICE directly from the binary tree using precice_DIR in other projects.
  • Added target test_install, which can be used to test the install. It configures builds and runs the cpp solverdummy using the installed preCICE library.
  • Added the ability to send a mesh to multiple participants.
  • Added the attribute Participant to the logger, which is now available in format and filter expressions.
  • Added the automatic creation of target directories, when exporting meshes with <export:vtk ...>.
  • Added the data name to the log output of convergence measures.
  • Added the generation of a pkg-config file for preCICE, which is usable after installing.
  • Added the generation of code coverage reports.
  • Changed the tested build-system in CI from SCons to CMake.
  • Fixed a critical bug in the nearest projection mapping rooting in the variant index tree. We decided to roll-back to the old implementation until there is a fix.
  • Fixed PETSc complaining about unused arguments when running tests.
  • Fixed bindings/fortran to use precice::SolverInterface instead of precice::impl::SolverInterfaceImpl
  • Fixed a potential integer overflow when parsing huge numbers in the configuration code.
  • Fixed a potential overflow in the computation of the side length of bounding boxes.
  • Fixed bug in preconditioner when we have convergence in the first iteration.
  • Fixed integration tests resetting the log-level by calling configure.
  • Fixed missing includes for Eigen
  • Fixed overuse of std::endl in code, file generation should be noticeably faster now.
  • Fixed the constness of passed pointer arguments as well as some member functions of the SolverInterface.
  • Fixed the default safety factor for nearest projection mapping by increasing it from 0.1 to 0.5.
  • Fixed Python bindings now offer all preCICE API functions (except get_mesh_handle)
  • Fixed CMake now properly detects PETSc on systems with a directory-scoped MPI installation (include/openmpi/mpi.h) when the compiler was not set to the MPI compiler wrapper.
  • Refactored Mesh::computeState into two logical units, improving the trace information.
  • Refactored computePartitions using the stl.
  • Refactored logger which dramatically reduces compile times.
  • Refactored the M2N handling code, which improves the available trace information for issues related to the M2N initialization.
  • Removed #395 API functions nameConfiguration(), dataDisplacements(), dataForces(), dataVelocities(), actionPlotOutput(), exportVTK(), exportAll() from CPP API and all language bindings.
  • Removed ancient PRECIE_NO_SOCKETS definition.
  • Removed dead configuration code.
  • Sending data between participants is now fully asynchronous. This is relevant in one-way coupling scenarios, where the sending participant doesn't need to wait for the receiving one.

1.4.1

  • Bug in re-partitioning fixed, occured for OpenFOAM and empty ranks in parallel.

1.4.0

  • The python modules are now proper packages tracking dependencies etc.
  • Fix CMake now importable from binary directory.
  • The Python module for the preCICE bindings PySolverInterface is renamed to precice. This change does not break old code. Please refer to src/precice/bindings/python/README.md for more information.
  • Add a pkg-config for preCICE (libprecice.pc).
  • Use the Boost stacktrace library for cross-platform stacktrace printing. This requires Boost 1.65.1.
  • Added explicit linking to libdl for boost::stacktrace.
  • Reimplemented the internals of the nearest-projection mapping to significantly reduce its initialization time.
  • The EventTimings now do a time normalization among all ranks, i.e., the first event is considered to happen at t=0, all other events are adapted thereto.
  • The old CSV format of the EventTimings log files, split among two files was replaced by a single file, in structured JSON format.
  • Fixed memory leaks in the xml::XMLAttributes and xml::Validator*.
  • Removed the xml::Validator* classes and replaced them in xml::XMLAttribute with a set of "options".
  • Made xml::XMLAttribute and xml::XMLTag chainable.
  • Added manpages for binprecice and testprecice.
  • Fixed memory leaks in mesh::Mesh.
  • Fixed mapping classes not flushing the underlying caches on clear().
  • Fixed format of version logging (first preCICE log message).
  • Added tolerance of iterations in quasi-Newton tests.
  • CMake overhaul:
    • Converted to target-based system: precice, testprecice, binprecice
    • New options:
      • PRECICE_Packages (default ON) to configure CPack,
      • PRECICE_InstallTest (default ON) to configure installation of tests.
        This includes the binary testprecice and necessary files. Use PREFIX/share/precice as PRECICE_ROOT.
    • Moved CMake files from tools/cmake-modules to cmake/ (general scripts) and cmake/modules (find modules).
    • Migrated from file-globing to explicit source/interface/test-file lists.
      Use tools/updateSourceFiles.py from project-root to update all necessary files.
    • install target installs:
      • the library PREFIX/lib.
      • the binaries PREFIX/bin and their manfiles into PREFIX/share/man/man1.
      • the CMake configuration files into PREFIX/lib/cmake/precice.
      • the pkg-config configuration files into PREFIX/lib/pkgconfig
      • the necessary files to run testprecice into PREFIX/share/precice. Use this as PRECICE_ROOT on installed system.
    • CTest definition of tests run in isolated working directories:
      • precice.Base for the base test suite
      • precice.MPI2 run on 2 MPI ranks
      • precice.MPI4 run on 4 MPI ranks
    • CPack configuration of target package to generate binary debian, tar and zip packages.
    • Added CMakeLists.txt to tools/solverdummy/cpp. It is an example of how to link to preCICE with CMake.
    • Extended the displayed information when configuring.
  • Extended updateSourceFiles.py to verify the sources using git ls-files --full-name if available.
  • Fixed the io::VTKXMLExporter not to write VertexNormals.
  • Improved the user-friendliness of the tests.
    • make test will run all tests.
    • make test_base only a unproblematic base-set.
    • A timeout will kill hanging tests.
    • All tests sets run in isolated working directories.
  • Added an (experimental) Fortran 2003 solver dummy.

1.3.0

  • Update of build procedure for python bindings (see precice/src/bindings/python/README.md for instructions). Note: you do not have to add PySolverInterface.so to PYTHONPATH manually anymore, if you want to use it in your adapter. Python should be able to find it automatically.
  • Make naming of log files consistent, following the pattern precice-SOLVERNAME-logtype.log, example: precice-FLUID-eventTimings.log
  • Enable boost.geometry based preallocation. Speeds up initialization of PetRBF based mapping.
  • Actions can now specify a MeshRequirement, such as the ScaleByAreaAction.
  • Many events have been reworked and are now uniformly named.
  • There is a syncMode for events (for detailed performance measurements), configurable and off by default.

1.2.0

  • Make polynomial=separate the default setting for PetRBF.
  • Removed ExportVRML functionality
  • Build system:
    • Make python=off default.
  • Building with Conda:
    • The helper scripts are now placed in the directory tools/conda_building. All the terms refering to Anaconda have been changed to Conda.
  • Sending data exchange is now fully asynchronous, so that the sending participant never waits for the receiving one.
  • Rename src/precice/adapters to src/precice/bindings
  • adding libprefix option in scons build process to allow for non-standard library paths

1.1.1

  • Fix SConstruct symlink build target failing when using lowercase build (debug, release) names.

1.1.0

  • Build system:

    • Remove the staticlib and bin from the default targets to reduce the building time and storage requirements.
    • Change build types to mixed case, i.e. Debug and Release. Old versions are retained for backward compatibility.
    • Make mpicxx default setting for compiler.
    • Experiemental support for building with Conda, see tools/anaconda_building
    • Use NumPy to figure out Python include paths.
    • Search for PETSc in more paths
    • Add experimental CMake build control files.
  • Add a job command file for the SuperMUC HPC system in tools.

  • compileAndTest.py Change the -b option to -t, do not crash when ./tests do no exist, make mpirun command configurable

  • Use libxml2 for XML parsing, this makes libxml2-dev a dependency.

  • Update EventTimings framework.

  • Add python script to plot Events on a timeline.

  • PETSc RBF mapping now supports conservative mapping with a separated polynomial

  • Converted all tests to the new, boost test based, unit testing framework.

  • Removed the tarch legacy library.

  • Use boost::signal2 for implement observer pattern for the Mesh class.

  • Add contributer guidelines.

1.0.3

  • Fix compilation for boost 1.66, see issue #93.

1.0.2

  • Fix bug in the mesh repartitioning for plane-like coupling interfaces and small gaps between both sides.

1.0.1

  • Fix compilation issue with the python interface.