TODO.list Tom Henderson 5/19/06 WRF-ESMF TODO LIST: WRF-ESMF is still new and experimental and has many limitations. Many of these are mentioned in the task list that follows. "X" == "DONE" The "ESMF" referred to below is version 2.2.0rp1. + Design and Implementation: - Get rid of masses of "DEBUG" code. - Fix ESMF+RSL_LITE. Currently, only ESMF+RSL works. Patch and memory extents differ between RSL and RSL_LITE, see /users/hender/Tasks/WRF_ESMF/RSL_LITE_broken/README for details. This is almost certainly the underlying cause of the problem given the evil hackery in the current implementation of io_esmf. - Finish sequence diagram for the coupling interactions in /users/hender/Tasks/WRF_ESMF/SequenceDiagram.txt ... - Upgrade SST component so it can run on a subset of processors. This requires breaking the domdesc dependence for SST so SST can be run on fewer processes. SST will need to call wrf_dm_patch_domain(), probably modified to avoid dependence on TYPE(domain) and certainly modified to avoid dependence on head_grid. - Upgrade external/io_esmf/ so more data types are supported. - Upgrade external/io_esmf/ so 3D arrays are supported (just loop over "K" at first with hard-coded names like "U_k_1", "U_k_2", etc.). Then switch to ESMF for this once it can do it. (Can ESMF do this now? Check.) - Extend external/io_esmf/ (etc.) so more than one I/O stream can be used for ESMF coupling. This would involve nested import and export states. - Exchange staggered 2D grids and test-validate (x, y, and z staggerings). Current hacked implementation should be able to handle this for horizontal staggerings, but need to test in case bugs are lurking... - Upgrade external/io_esmf/ so metadata is exchanged correctly. Right now, metadata is ignored. This would allow sharing of metadata and simplify implementation of components that were aware of WRF metadata. Ultimately, CF would be the right thing here... - Fix output formatting so DOMAIN_TIME_TEST works the same way both with and without ESMF. - Fix ext_esmf_ioclose so it actually destroys ESMF objects again. This is #ifdef'd out at the moment due to difficulties with destruction. (At present, ESMF requires users to deallocate anything they allocate, but provides no introspection so a user can figure out if who allocated something. This leads to obfuscation of implementations...) X Make WRF-ESMF build recognize ESMF environment variables? Could remove stanzas from arch/configure.defaults and avoid lots of duplication and documentation. - If possible, upgrade WRF ESMF component to get number of MPI tasks from VM instead of from MPI. This may be tricky or even impractical due to startup issues... - CF conventions: * Numerous tasks here, not really a "WRF-ESMF" issue, but related... Will there be CF conventions for startTime, stopTime, timeStep, couplingInterval? - Other ESMF bugs/issues: * Rip out the current hacks for grid creation in external/io_esmf/ once ESMF can support WRF map projections. Use the "sieve" method of io_mcel. This will take a bit of work... * Restore ESMF_LogErr calls once ESMF fixes them so they no longer truncate all of our messages (trivial for us...). * Ask when ESMF_LogErr will allow us to specify LUN. This will make it easier for us to use this utility. * Upgrade external/io_esmf/ to avoid manual CICO once ESMF can mimic WRF memory layouts. * Index ordering is hard-coded. Fix this once ESMF supports it and connect to "MemoryOrder" argument. * Take a look at my list of reported ESMF bugs and add any I haven't reported. * Calls to ESMF_LogSet in main/wrf_ESMFApp.F cause core dumps -- why? * Calls to ESMF_ArraySpecSet() in external/io_esmf/ quietly do the wrong thing. Why? Once they work, uncomment this code to avoid per-data-type duplication of hard-coded alternative (gaaak). See string DOESNOTWORK. * ESMF cannot handle "extra" rows-columns implied by horizontal staggering for regional models like WRF. Upgrade once ESMF supports this. * Re-connect "Stagger" argument to ESMF "horzrelloc" argument. Uncomment this code in external/io_esmf/. * Until then, maybe use nested ESMF_States to handle it. This is ugly though because there are no conventions for such a thing so it will introduce dependencies with other components. Balaji's recent proposal to standardize grid metadata within CF would be useful if we go this way. * Why can I specify start indices in the call to ESMF_GridDistribute[Vector]() but not in the call to ESMF_GridSet() ? * Remove implementation of WRFU_TimeIntervalDIVQuot in external/io_esmf/module_esmf_extensions.F90 once ESMF supports this directly. * Lack of sane (or any) iterators for ESMF_State and other container classes leads to nasty user code... * Lack of adherence to Orthodox Canonical Form leads to really nasty user code (so much for "best practices"...) * Many others... - Misc. bugs/issues: * Investigate odd debug prints from med_hist_out, output_wrf, input_wrf med_hist_out : opening sstout_d01_000000 for writing. 86 I think the "86" should be "0" output_wrf: fid,filestate = 2 65 I think the "65" should be "0" input_wrf: fid,filestate = 3 804370368 I think the "804370368" should be something else... * Find out why one new message now appears in stdout (weird). This happens with a "no-esmf" build and with an "esmf" build. INPUT LANDUSE = USGS LANDUSE TYPE = USGS FOUND 24 CATEGORIES 2 SEASONS WATER CATEGORY = 16 SNOW CATEGORY = 24 * Replace hand-coded checks of "rc" with use of new ESMF error handler to streamline code. See /loquat2/hender/Tasks/WRF_ESMF/ESMF_FieldFromUserEx.F90 See /loquat2/hender/Tasks/WRF_ESMF/UserCodeMod.F90 * See if "#ifndef ESMFIO" hack can be removed from share/input_wrf.F. * Verify that findSymbol.rb generates module_symbols.F90 as claimed in the comment block in external/io_esmf/module_symbols_util.F90. Update comments if needed. * Upgrade findSymbol.rb to track all usage starting at @old_module and exclude any symbols that are not found in the use-tree. * Better yet, get rid of external/*/module_symbols_util.F90 altogether as encapsulation moves to module_domain. * Rip out WRF_CHEM from set_timekeeping.F and test... + Upgrade to ESMF 2.2.2r+ - Add an ESMF_LOG_NONE option to the "defaultLogType" argument of ESMF_Initialize() to turn off ESMF's default error logging. - Add an ESMF_KEEPMPI option to the "terminationflag" argument of ESMF_Finalize() to allow ESMF finalization without MPI shutdown. Then clean up the current shutdown mess. + Documentation: - General description of and guidelines for coupling WRF with another ESMF component. X Description of WRF-CPL-SST "demo" X Description of SST and CPL demo components * Sequence diagram of component interactions X Limitations of demo * How to extend the demo + Testing - Add WRF-CPL-SST tests to regtest.csh. * Generate jun01 sst warming data set and work with Dave to add new "ESMF_SST" tests to regtest.csh.