Add get_checkpoint_id() function to simulator_access

[danieldouglas92]

This PR creates a function that enables other ASPECT plugins to access the last checkpoint id created by ASPECT. This is something that is currently useful for my postdoc work, and may also be useful to other groups who are coupling ASPECT to external softwares.

[tjhei]

ideally, we would return the current checkpoint id while a checkpoint is being written (I know that this is why you want this function 😄 ). Sadly,

aspect/source/simulator/checkpoint_restart.cc

Line 444 in ba06f0f

last_checkpoint_id = checkpoint_id;

currently happens after we call save:

aspect/source/simulator/checkpoint_restart.cc

Line 381 in ba06f0f

oa << (*this);

But we could update the id before instead. Thoughts?

[tjhei]

Can you please be more specific what this returns if you call this while a checkpoint is being written (inside save inside a plugin)? It might make sense to add a test with a simple plugin that prints this number to the screen.

[danieldouglas92]

Moving the assignment of last_checkpoint_id made it so that when this function is called within an overrided save() the checkpoint that is about be written is returned!

[tjhei]

you have to attach to a signal and print this information during runtime.

[tjhei]

see

aspect/tests/edit_parameters.cc

Line 54 in ba06f0f

simulator_access.get_pcout() << "Signal start_timestep triggered!" << std::endl;

[danieldouglas92]

Is this what was missing?

[tjhei]

No. You will need to attach to a signal and print when the signal fires, otherwise you have no SimulatorAccess.

[danieldouglas92]

Ahhh your other comment hadn't loaded for some reason when I asked. I'll take a look at that example

[danieldouglas92]

@tjhei I'm not sure why the dealii-master tester is failing to compile, but what I have right now shows that the new function gets the checkpoint_id, but the id that gets printed in the test lags behind the actual checkpoint being written because the start_timestep signal obviously happens before checkpointing. I ended up just going with this because it was simpler. I initially tried using the pre_checkpoint_store_user_data signal but this requires a triangulation type to be passed to it which seemed complicated to do in a simple test, plus the signal would still have the same issue given that it happens at line 363 in checkpoint_restart.cc and last_checkpoint_id is updated at line 375.

[danieldouglas92]

To properly show that it's getting the correct checkpoint_id when this function is meant to be used (in an overridden save() function), I guess I would need to create a new signal?

[tjhei]

To properly show that it's getting the correct checkpoint_id when this function is meant to be used (in an overridden save() function), I guess I would need to create a new signal?

You just need any plugin that implements save. Something like:

    template <int dim>
    class MyPostprocessor : public Postprocess::Interface<dim>, public ::aspect::SimulatorAccess<dim>
    {
      public:
        /**
         * Generate graphical output from the current solution.
         */
        virtual
        std::pair<std::string,std::string>
        execute (TableHandler &statistics);

        /**
         * Save the state of this object.
         */
        void save (std::map<std::string, std::string> &status_strings) const override;
    };

    template <int dim>
    std::pair<std::string,std::string>
    MyPostprocessor<dim>::execute (TableHandler &statistics)
    {}

    template <int dim>
    void
    MyPostprocessor<dim>::save ([TableHandler &statistics)](std::map<std::string, std::string> &status_strings)
    {
std::cout << this->get_current_checkpoint_id() << std::endl;
}



namespace aspect
{
  namespace Benchmark
  {
    ASPECT_REGISTER_POSTPROCESSOR(MyPostprocessor,
                                  "MyPostprocessor",
                                  "")

Make sure you add this postprocessor to the list of active ones.

[danieldouglas92]

@tjhei Thanks for helping with setting up this test, this should be good now! It shows the current checkpoint id that is being written in the screen output

[tjhei]

Suggested change

	* Return the ID of the checkpoint that ASPECT is currently writing.
	* Return the ID of the checkpoint that ASPECT is currently writing or was last written.

[tjhei]

delete this file

[danieldouglas92]

Done!