Error in writing and reading cell based data for restart

[Sambit Das]

Hi all,

I have written the following failing minimal example (also attached) where I create a single element parallel distributed triangulation and try to write and read a double. 

using namespace dealii;

int main (int argc, char *argv[])

{

  Utilities::MPI::MPI_InitFinalize mpi_initialization(argc, argv);

  const double L=20;

  parallel::distributed::Triangulation<3> triangulation(MPI_COMM_WORLD);

  GridGenerator::hyper_cube (triangulation, -L, L);

  unsigned int offset = triangulation.register_data_attach

          (sizeof(double),

   [&](const typename dealii::parallel::distributed::Triangulation<3>::cell_iterator &cell,

       const typename dealii::parallel::distributed::Triangulation<3>::CellStatus status,

       void * data) -> void

       {

     double* dataStore = reinterpret_cast<double*>(data);

     *dataStore=0.0;

       }

   );

  std::cout<< "offset=" << offset << std::endl;

  std::string filename="triangulationChk";

  triangulation.save(filename.c_str());

  triangulation.load(filename.c_str());

  triangulation.notify_ready_to_unpack

      (offset,[&](const typename dealii::parallel::distributed::Triangulation<3>::cell_iterator &cell,

        const typename dealii::parallel::distributed::Triangulation<3>::CellStatus status,

        const void * data) -> void

        {

        }

       );

}

I get the following error:

An error occurred in line <3236> of file </home/vikramg/DFT-FE-softwares/softwareCentos/dealiiDev/dealii/source/distributed/tria.cc> in function

    void dealii::parallel::distributed::Triangulation<dim, spacedim>::notify_ready_to_unpack(unsigned int, const std::function<void (const dealii::Triangulation<dim, spacedim>::cell_iterator &, dealii::Triangulation<dim, spacedim>::CellStatus, const void *)> &) [with int dim = 3, int spacedim = 3]

The violated condition was: 

    offset < sizeof(CellStatus)+attached_data_size

Additional information: 

    invalid offset in notify_ready_to_unpack()

I am wondering if I am missing something in my implementation.

Thanks,

Sambit

[Sambit Das]

An error occurred in line <3236> of file </home/vikramg/DFT-FE-softwares/softwareCentos/dealiiDev/dealii/source/distributed/tria.cc> in function

    void dealii::parallel::distributed::Triangulation<dim, spacedim>::notify_ready_to_unpack(unsigned int, const std::function<void (const dealii::Triangulation<dim, spacedim>::cell_iterator &, dealii::Triangulation<dim, spacedim>::CellStatus, const void *)> &) [with int dim = 3, int spacedim = 3]

The violated condition was: 

    offset < sizeof(CellStatus)+attached_data_size

Additional information: 

    invalid offset in notify_ready_to_unpack()

I tried to debug this using ddt debugger- the reason the above condition is violated is due to attached_data_size=0, which is strange as the triangulationChk.info file mentions one object to be attached:

 triangulationChk.info 

version nproc attached_bytes n_attached_objs n_coarse_cells

2 1 12 1 1 

[Wolfgang Bangerth]

> I get the following error:
>

> /An error occurred in line <3236> of file
> </home/vikramg/DFT-FE-softwares/softwareCentos/dealiiDev/dealii/source/distributed/tria.cc>
> in function/
> / void dealii::parallel::distributed::Triangulation<dim,

> spacedim>::notify_ready_to_unpack(unsigned int, const std::function<void
> (const dealii::Triangulation<dim, spacedim>::cell_iterator &,
> dealii::Triangulation<dim, spacedim>::CellStatus, const void *)> &) [with int

> dim = 3, int spacedim = 3]/
> /The violated condition was: /
> / offset < sizeof(CellStatus)+attached_data_size/
> /Additional information: /
> / invalid offset in notify_ready_to_unpack()/
> /
> /

> I am wondering if I am missing something in my implementation.

Nothing obvious at least. I copied the issue here:
https://github.com/dealii/dealii/issues/6223

I don't recall why we reset the size to zero. The intention was probably that
we don't ever call `load()` in the same execution as we called `save()`
before. We may be calling `save()` many times, of course, in the course of
long-running programs, but call `load()` only at the beginning of a restart.
Consequently, this may simply be a bug that we have just never encountered, or
there may have been a reason to do it that way -- I can't recall.

I imagine you can work around this is you simply call `register_data_attach()`
again before calling `load()`.

Best
W.

--
------------------------------------------------------------------------
Wolfgang Bangerth email: bang...@colostate.edu
www: http://www.math.colostate.edu/~bangerth/

[Sambit Das]

Prof. Bangerth, Thank you for your reply and creating the issue.

I imagine you can work around this is you simply call `register_data_attach()`
again before calling `load()`.

I checked that calling  `register_data_attach()`again before calling `load()` 

still triggers the same assert, but calling `register_data_attach()` again just after 

calling `load()`  and before calling 'notify_ready_to_unpack` works- I 

am able to read and print the correct cell data from the file.

Best,

Sambit