Add a diag_send_complete call

[uramirez8707]

Add a diag_send_complete call. This is needed for the new diag manager. It is backwards compatible with the old diag manager.

[bensonr]

@rem1776 - with Niki's comment about sub-stepping in the ocean component, will this fix be sufficient?

[rem1776]

@rem1776 - with Niki's comment about sub-stepping in the ocean component, will this fix be sufficient?

I am not sure. I was under the impression this call would be needed regardless, and other updates would go to the ocean component to handle the sub-stepping if needed.

@uramirez8707 @nikizadehgfdl What do you guys think?

[bensonr]

@rem1776 - what's the behavior for diag_send_complete when I call it twice in a row with no new send_data calls in between? This is where we might end up with trouble by having it in both the component and the coupler driver.

[uramirez8707]

@rem1776 - what's the behavior for diag_send_complete when I call it twice in a row with no new send_data calls in between? This is where we might end up with trouble by having it in both the component and the coupler driver.

That will not be good.
https://github.com/NOAA-GFDL/FMS/blob/bb6de937f70a08a440f5e63b8553b047c1921509/diag_manager/fms_diag_object.F90#L688-L689

[rem1776]

@rem1776 - what's the behavior for diag_send_complete when I call it twice in a row with no new send_data calls in between? This is where we might end up with trouble by having it in both the component and the coupler driver.

That will not be good. https://github.com/NOAA-GFDL/FMS/blob/bb6de937f70a08a440f5e63b8553b047c1921509/diag_manager/fms_diag_object.F90#L688-L689

Yeah it'd be incrementing the time step along with running the reduction an extra time.

[uramirez8707]

this%current_model_time should be modified inside send_data instead, if there is a new time

[rem1776]

this%current_model_time should be modified inside send_data instead, if there is a new time

I could try to add that in and then have it ignore any un-needed reductions from extra calls, but would that solve the issues with the ocean model substeps?

[uramirez8707]

This is going to prevent the math from being done twice: NOAA-GFDL/FMS@efbb9f0

The ocean model substeps still need to call diag_send_complete. If they don't the code will crash with something like: https://github.com/NOAA-GFDL/FMS/blob/bb6de937f70a08a440f5e63b8553b047c1921509/diag_manager/fms_diag_file_object.F90#L1204-L1206

[thomas-robinson]

Is this going to add on to the model time? I don't think the math needs to be done in fms_diag_send_complete in the ocean model because there isn't any threading. The math should all be done in accept_data

[uramirez8707]

no it doesn't add to the model time. It just updates the model_time with whatever was passed in send_data.

That commit also sets math_needs_to_be_done back to .false. after done doing the math. Otherwise if you call diag_send_complete twice in a row with no new send_data calls in between the math is going to be done twice.