1. 16 Jan, 2018 1 commit
  2. 06 Oct, 2017 1 commit
  3. 07 Dec, 2016 1 commit
  4. 08 Mar, 2016 2 commits
  5. 18 Jan, 2016 1 commit
  6. 03 Dec, 2015 1 commit
  7. 29 Oct, 2015 1 commit
  8. 16 Oct, 2015 1 commit
  9. 15 Oct, 2015 2 commits
  10. 13 Sep, 2015 1 commit
  11. 30 Jun, 2015 1 commit
  12. 29 Jun, 2015 1 commit
  13. 29 May, 2015 1 commit
    • Christophe Geuzaine's avatar
      · 3f96520c
      Christophe Geuzaine authored
      WARNING: potentially dangerous changes to the way nonlinear systems are assembled and solved:
      
      1) the computation of the residual DofData->res = b-Ax has been moved from SolveJac to GenerateJac
      
      2) the calculation of the full Jacobian matrix has also been moved from SolveJac to GenerateJac
      
      3) DofData->Jac now stores the full Jacobian (as soon as GenerateJac is done), instead of only the 
         JacNL terms ; DofData->A is not perturbed anymore
      
      This only applies if Flag_Only is not set.
      
      Performance used to be a problem with this approach in the past (PETSc 3.3) ; it seems OK now.
      
      Please test and report any breakage.
      
      Christophe & Kevin.
      
      3f96520c
  14. 14 May, 2015 1 commit
  15. 13 May, 2015 1 commit
  16. 12 May, 2015 2 commits
  17. 11 May, 2015 1 commit
  18. 07 May, 2015 1 commit
  19. 05 May, 2015 1 commit
  20. 14 Feb, 2015 1 commit
  21. 02 Nov, 2014 1 commit
  22. 15 Jul, 2014 1 commit
  23. 12 Jun, 2014 1 commit
  24. 10 Apr, 2014 1 commit
  25. 12 Feb, 2014 1 commit
  26. 04 Dec, 2013 1 commit
  27. 03 Dec, 2013 1 commit
    • Christophe Geuzaine's avatar
      Changed behaviour of DefineConstant & co for better symmetry between GetDP/Gmsh · 7616c416
      Christophe Geuzaine authored
      and Python onelab clients.
      
      WARNING WARNING WARNING: This is a major change -- all our onelab-enabled .geo
      and .pro files will need to be (slightly...) modified.
      
      What's new:
      
      1. The name of a onelab variable (in the onelab database) is no more constructed
         from the name of the corresponding GetDP/Gmsh variable. One now needs to
         specify the onelab name explicitely, using the "Name" attribute. The "Name"
         is the actual name of the parameter in the onelab database, i.e., it also
         includes the path.
         
         This makes the "Path" attribute obsolete (it has no effect anymore). The
         "Legend" attribute can still be used (and it can be useful in edge cases,
         e.g. when you want a "/" in the name of a onelab paramater), but in most
         cases it's not necessary.
      
      2. When a DefineConstant[] & co is used and no Name is given
         (e.g. DefineConstant[a=2]), no onelab parameter is created. This allows to
         provide default values to internal parameters without polluting the database.
      
      Why did we change?
      
      1. The new syntax matches what we do in Python, where specifying a name is
         mandatory (there's no way around this in Python, as onelab cannot guess the
         name of a Python variable to which a onelab parameter value will be assigned).
         
         The change will prevent common mistakes where two parameters with the same
         label actually correspond to 2 different onelab parameters, due to a change
         in local getdp/gmsh variable name (which would change the onelab name
         automatically)
      
      2. The new syntax allows to nicely decouple onelab parameters from internal
         variables with default values, that we don't want in the onelab database.
      
      
      7616c416
  28. 04 Oct, 2013 1 commit
  29. 01 Sep, 2013 1 commit
  30. 31 Aug, 2013 1 commit