From: Wolfgang Bangerth Date: Sun, 6 Dec 2015 17:46:32 +0000 (-0600) Subject: Clean up some changelog entries. X-Git-Tag: v8.4.0-rc2~178^2~1 X-Git-Url: https://gitweb.dealii.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=69c3138a89000152aef6553842686e633402bf4c;p=dealii.git Clean up some changelog entries. --- diff --git a/doc/news/changes.h b/doc/news/changes.h index 633c2d82bb..08359ac86b 100644 --- a/doc/news/changes.h +++ b/doc/news/changes.h @@ -38,14 +38,14 @@ inconvenience this causes.

    -
  1. Changed: The function Utilities::trim now removes general +
  2. Changed: The function Utilities::trim() now removes general white space characters, such as '\\r' and '\\n', as well as space characters.
    (David Wells, 2015/12/05)
  3. -
  4. Removed: Previously deprecated global instance multithread_info of +
  5. Removed: The previously deprecated global instance multithread_info of MultithreadInfo has been removed (all members of this class are static so there is no reason to use/create an instance). The deprecated MultithreadInfo::n_cpus member also got removed in favor of @@ -57,7 +57,7 @@ inconvenience this causes. update_support_points, update_support_jacobians, and update_support_inverse_jacobians have been removed. update_support_points was deprecated in 2013 and has not done - anything in a long time (see the commit message for more information). The + anything in a long time. The other two appeared in 2007 and were never implemented.
    (David Wells, 2015/09/16) @@ -93,9 +93,9 @@ inconvenience this causes. (Matthias Maier, 2015/09/14 - 2015/09/17)
  6. -
  7. Removed: Tensor as well as Point no +
  8. Removed: The Tensor and Point classes no longer have a constructor taking a boolean argument. Those were replaced - by a default constructor will always initialize underlying values with + by a default constructor that will always initialize underlying values with zero.
    (Matthias Maier, 2015/09/07) @@ -157,7 +157,9 @@ inconvenience this causes. to compute on every cell. Consequently, these two functions have been removed and have been replaced by Mapping::requires_update_flags().
    - (Wolfgang Bangerth, 2015/07/20-2015/08/13) + A similar change has been applied to the FiniteElement class. +
    + (Wolfgang Bangerth, 2015/07/20-2015/12/01)
  9. Changed: The function DoFRenumbering::random() now produces different @@ -207,9 +209,10 @@ inconvenience this causes.
  10. Changed: The functionality to distribute cells across processes according to a vector of cell weights that was passed in a call to - parallel::distributed::Triangulation::repartition() + parallel::distributed::Triangulation::repartition() was replaced by a cell-wise signal. This signal is called during - execute_coarsening_and_refinement and repartition + parallel::distributed::Triangulation::execute_coarsening_and_refinement() and + parallel::distributed::Triangulation::repartition() if any function is connected to it. It allows to connect a function that takes the current cell iterator and a status argument that indicates whether this cell will be refined, coarsened or remains unchanged and returns a @@ -220,7 +223,7 @@ inconvenience this causes. (Rene Gassmoeller, 2015/11/02)
  11. -
  12. New: 2nd derivatives are implemented for polynomials_BDM in 3D. +
  13. New: 2nd derivatives are implemented for PolynomialsBDM in 3D.
    (Alistair Bentley, 2015/10/27)
  14. @@ -239,20 +242,22 @@ inconvenience this causes. (Lei Qiao, 2015/10/22) -
  15. New: Triangulation::ghost_owners() returns the set of MPI ranks of the - ghost cells. Similarly ::level_ghost_owners() for level ghosts. +
  16. New: parallel::distributed::Triangulation::ghost_owners() + returns the set of MPI ranks of the ghost cells. Similarly + parallel::distributed::Triangulation::level_ghost_owners() for level + ghosts.
    (Timo Heister, 2015/09/30)
  17. -
  18. New: FunctionParser now supports pow(a,b). +
  19. New: FunctionParser now supports pow(a,b).
    (Timo Heister, 2015/09/30)
  20. -
  21. Improved: The interface to all deal.II type solvers and +
  22. Improved: The interfaces to all deal.II type solvers and preconditioners have been updated such that they function as expected - with the LinearOperator class and its associated functions (i.e. + with the LinearOperator class and its associated functions (i.e., linear_operator(), transpose_operator() and inverse_operator()). These preconditioners can now be wrapped as a LinearOperator, facilitating the construction of approximate matrix inverses such as in @@ -321,7 +326,7 @@ inconvenience this causes. (Matthias Maier, 2015/09/02)
  23. -
  24. Fixed: VectorTools::integrate_difference for VectorTools::Hdiv_seminorm +
  25. Fixed: VectorTools::integrate_difference() for VectorTools::Hdiv_seminorm was computed incorrectly.
    (Timo Heister, 2015/08/31) @@ -350,7 +355,6 @@ inconvenience this causes. - FunctionTime - Function - TensorFunction -
    Classes with fixed interface that now fully support complex number types (pure template classes without explicit instantiations in the @@ -359,7 +363,7 @@ inconvenience this causes. - PackagedOperation - Tensor
    - (Matthias Maier, 2015/08/25 - XXX) + (Matthias Maier, 2015/08/25)
  26. Fixed: The testsuite now properly supports version constraints for @@ -396,8 +400,8 @@ inconvenience this causes.
  27. New: parallel::shared::Triangulation class which extends Triangulation class to automatically partition triangulation when run - with MPI. Identical functionality between parallel::shared::Triangulation and - parallel::distributed::Triangulation is grouped in the parent class + with MPI. Common functionality between parallel::shared::Triangulation and + parallel::distributed::Triangulation is implemented in the parent class parallel::Triangulation.
    (Denis Davydov, 2015/08/14) @@ -411,8 +415,8 @@ inconvenience this causes. (Jason Sheldon, Wolfgang Bangerth, 2015/08/13)
  28. -
  29. New: implemented the gradient method for - InterpolatedTensorProductGridData +
  30. New: The InterpolatedTensorProductGridData::gradient() function + is now implemented.
    (Daniel Shapero, 2015/08/12)
  31. @@ -459,7 +463,7 @@ inconvenience this causes.
    1. Improved: The method - parallel::distributed::Triangulation::fill_vertices_with_ghost_neighbors + parallel::distributed::Triangulation::fill_vertices_with_ghost_neighbors() that is used for distributing DoFs on parallel triangulations previously exhibited quadratic complexity in the number of coarse grid cells. This has been changed into linear complexity calls (apart from a few issues