[Mitgcm-support] Re: KPP (c30) bug

[mitgcm-support at dev.mitgcm.org]

>> What are the inconsistencies?

I don't have a complete list yet, but they include treatment of
shortwave forcing and package initialization and checking.

The general idea is that package maintainers should be able to
modify packages with minimum modifications to baseline code and
hence without having to molest you and Alistair on a continuous
basis. What you suggest is better than what is actually there
right now but still falls short from what I believe is needed.

1. My preference would be for CPP_OPTIONS.h and its dependents
to be visible to all model subroutines.  For example when using
KPP I want to turn off convection and turn on short-wave forcing.
The way this would be handled right now is by issuing warnings
and aborting at runtime in package_check, which seems a bit
wasteful.  So instead of

    CPP_OPTIONS.h
    #define ALLOW_PKG1
    #define ALLOW_PKG2
    ...
    #define CPP_OPTION1
    #define CPP_OPTION2
    ...

I suggest we use

    CPP_OPTIONS.h
    #define CPP_OPTION1
    #define CPP_OPTION2
    ...
    #include "PKG1_OPTIONS.h"
    #include "PKG2_OPTIONS.h"

It's up to package maintainers to make sure their options do not
interfere with baseline code and with other packages.  For example,
package option files couls be as simple as:

    PKG1_OPTIONS.h
    #define ALLOW_PKG1

But they could also do things like:

    PKGX_OPTIONS.h
    #define ALLOW_PKGX
    #ifdef ALLOW_PKGX
    #undef CPP_OPTIONX
    ...
    #endif

2. PACKAGE_CHECK and PACKAGE_INIT
   Note that file names and subroutine names are spelled differently,
   and that right now these routines are very different from what you
   describe in your message.

   As you suggest it's best to let each package check for
   inconsistencies on its own.  Otherwise you or Alistair will
   have to intervene every time someone tries to change something in
   the package.  Again the idea is to transfer responsibility from
   you and Alistair to individual package maintainers.  I much
   prefer your second option which is to altogether get rid of
   PACKAGE_CHECK.

   The only purpose of PACKAGE_INIT and data.pkg right now is to set
   or unset use_PKGX.  Both KPP_INIT and GMREDI_INIT appear in
   initialise_varia.F.  Again I prefer your option #2, that is, call
   PKGX_INIT from the baseline code and let the  package maintainer
   decide whether he needs (or wants) to define a use_pkgx flag.

To summarize I vote for minimum restrictions and minimum interference with
baseline code, where practical, so that I have to bug you and Alistair as
little as possible.

Dimitris