TODOs

Mechanical / logistical things to fix:

• revamp regression

  • cleaner/tests
  • accept the weird macro situations
    • "skipping due to weird macros", checked by hand so far
    • MPI_Alloc_mem
    • MPI_Win_allocate
    • MPI_Win_allocate_shared
    • MPI_Win_shared_query
  • allow plugin based architecture, find specials
    • find parameters with different names
    • find differing LIS types without whole diffing
  • F08 POLY expressed causes name collision, not important for regression, but for comparison
    • how can we identify the polymorphic F08 bindings? there is no name difference...

• revamp KIND system

  • rename KINDs to reasonable/understandable names
  • simple maps no longer capture everything required
  • required refactor and rethink

• pythonization

  • the function parameter generation only for large variants abuses the postfix parameter
    • this was never the intention of this parameter
    • future rewrite with the KIND map system will need to move POLY ideas into the generators themselves
    • this abuses the POLY kinds to cause multiple function expressions and the postfix interpreted as the POLY flag
  • the POLY expression needs to be moved done to generators
    • this will remove the need for the KIND maps as is
    • replace with a nicer to edit and more understandable system
    • this change will also allow testing POLY expression, not just the generator expression

• chapter feedback

  • state corrections, from prior to 4.x, they were wrong before we just replicated
  • chapter 7 corrections, with changes to prior bindings, deferred until after pythonization
  • pt2pt INDEX -> non-negative
  • generally reduce KINDs, these are a lot of special cases, therefore shows complexity of our type system

• nicer indentation through Latex

  • MPI_Comm_spawn_multiple
    • array_of_commands, array_of_argv, linebreak without indentation
  • MPI_Type_indexed
    • newtype parameter in LIS prototype is split to "new-\ntype"
      • just new line on comma, don't allow line breaking a parameter name
  • MPI_Error_class is split by a two pages table

• header generation

  • constants and functions
  • how do we do predefined functions?
    • they are defined in "english"
    • we just give straight C code?

• lemonspotter

  • integration (partially done) (functions were easy)
  • typing is complicated from current KINDs

• Weird mpifoverloadOnlyInAnnex

  • MPI_ALLOC_MEM
  • MPI_WIN_ALLOCATE
  • MPI_WIN_ALLOCATE_SHARED
  • MPI_WIN_SHARED_QUERY

• MPI_Copy_function and MPI_Delete_function are not found in regression, because they are not a mpibind statement, they are written by hand in verbatim. The F90 version is a binding, but not the C version.

• EXTRA_STATE and EXTRA_STATE2 exist

• C_BUFFER, C_BUFFER2, C_BUFFER3, C_BUFFER4 exist There have no difference other than textual, which is not semantic. Some author wasn't paying attention.

• Also we may want to discuss unifying how we do argv for MPI_Init, MPI_Init_thread, MPI_Comm_spawn and MPI_Comm_spawn_multiple. MPI_Init does ***argv, MPI_Comm_spawn does *argv[]. One of those * is because for MPI_Init it is an inout, but the [] and * is iffy. This refers back to our discussion that we cannot liberally remove [] and put *, so this would be an errata.

• MPI_T_cvar_get_num, using wrong type for num_cvar

  • Not a big deal right now because even though it's the wrong KIND, it renders to the right language type.
  • Factors in with the "revamp the KIND system" issue

Already fixed:

• FIXED MPI_Datarep_extent_function

  • C parameter file_extent does not match name in F08 and F90

• FIXED MPI_Comm_errhandler_function, C is using weird notation, is it wildcard * or pointer *?

• FIXED MPI_File_errhandler_function, C is using weird notation, is it wildcard * or pointer *?

• FIXED MPI_Win_errhandler_function, C is using weird notation, is it wildcard * or pointer *?

• FIXED MPI_F_SYNC, F90 parameters lowercase, shouldn't be an issue due to case insensitive

• RESOLVED all mpit functions with \textoutargs{}...

  • MPI_T_enum_get_info
  • MPI_T_enum_get_item
  • MPI_T_cvar_get_info
  • MPI_T_pvar_get_info
  • MPI_T_category_get_info
  • RESOLVED: Not an issue (MPIT does not get a special function)