| 1 |
|
| 2 |
API Discussions: |
| 3 |
================ |
| 4 |
|
| 5 |
As discussed in our group meeting of 2003-12-17 (AJA, CNH, JMC, AM, |
| 6 |
PH, EH3), the NetCDF interface should resemble the following FORTRAN |
| 7 |
subroutines: |
| 8 |
|
| 9 |
1) "stubs" of the form: MNC_WV_[G|L]2D_R[S|L] () |
| 10 |
|
| 11 |
2) MNC_INIT_VGRID('V_GRID_TYPE', nx, ny, nz, zc, zg) |
| 12 |
|
| 13 |
3) MNC_INIT_HGRID('H_GRID_TYPE', nx, ny, xc, yc, xg, yg) |
| 14 |
|
| 15 |
4) MNC_INIT_VAR('file', 'Vname', 'Vunits', 'H_GTYPE', 'V_GTYPE', PREC, FillVal) |
| 16 |
|
| 17 |
5) MNC_WRITE_VAR('file', 'Vname', var, bi, bj, myThid) |
| 18 |
|
| 19 |
This is a reasonable start but its inflexible since it isn't easily |
| 20 |
generalized to grids with dimensions other than [2,3,4] or grids with |
| 21 |
non-horizontal orientations (eg. vertical slices). |
| 22 |
|
| 23 |
|
| 24 |
Generalizing what we would like to write as "variables defined on 1-D |
| 25 |
to n-D grids", one can imagine a small number of objects containing |
| 26 |
all the relevant information: |
| 27 |
|
| 28 |
a dimension: [ name, size, units ] |
| 29 |
a grid: [ name, 1+ dim-ref ] |
| 30 |
a variable: [ name, units, *1* grid-ref, data ] |
| 31 |
an attribute: [ name, units, data ] |
| 32 |
a NetCDF file: [ name, 0+ attr, 0+ grid-ref, 0+ var-ref ] |
| 33 |
|
| 34 |
which can then be manipulated (created, associated, destroyed, etc.) |
| 35 |
using a simple interface such as: |
| 36 |
|
| 37 |
MNC_INIT( myThid ) |
| 38 |
|
| 39 |
MNC_FILE_CREATE( myThid, fname ) |
| 40 |
MNC_FILE_OPEN( myThid, fname, itype ) |
| 41 |
MNC_FILE_ADD_ATTR_STR( myThid, fname, atname, sval ) |
| 42 |
MNC_FILE_ADD_ATTR_DBL( myThid, fname, atname, len, dval ) |
| 43 |
MNC_FILE_ADD_ATTR_REAL(myThid, fname, atname, len, rval ) |
| 44 |
MNC_FILE_ADD_ATTR_INT( myThid, fname, atname, len, ival ) |
| 45 |
MNC_FILE_ADD_ATTR_ANY( myThid, fname, atname, atype, cs,len,dv,rv,iv ) |
| 46 |
... |
| 47 |
MNC_FILE_READ_HEADER( myThid, fname ) |
| 48 |
|
| 49 |
MNC_DIM_INIT( myThid, fname, dname, dlen ) |
| 50 |
|
| 51 |
MNC_GRID_INIT( myThid, fname, gname, ndim, dnames ) |
| 52 |
|
| 53 |
MNC_VAR_INIT_DBL( myThid, fname, gname, vname, units ) |
| 54 |
MNC_VAR_INIT_REAL( myThid, fname, gname, vname, units ) |
| 55 |
MNC_VAR_INIT_INT( myThid, fname, gname, vname, units ) |
| 56 |
MNC_VAR_INIT_ANY( myThid, fname, gname, vname, units, type ) |
| 57 |
MNC_VAR_ADD_ATTR_STR( myThid, fname, vname, atname, sval ) |
| 58 |
MNC_VAR_ADD_ATTR_DBL( myThid, fname, vname, atname, nv, dval ) |
| 59 |
MNC_VAR_ADD_ATTR_REAL( myThid, fname, vname, atname, nv, rval ) |
| 60 |
MNC_VAR_ADD_ATTR_INT( myThid, fname, vname, atname, nv, ival ) |
| 61 |
MNC_VAR_ADD_ATTR_ANY( myThid, fname, vname, atname, atype, cs,len,dv,rv,iv ) |
| 62 |
MNC_VAR_WRITE_DBL( myThid, fname, vname, var ) |
| 63 |
MNC_VAR_WRITE_REAL( myThid, fname, vname, var ) |
| 64 |
MNC_VAR_WRITE_INT( myThid, fname, vname, var ) |
| 65 |
MNC_VAR_WRITE_ANY( myThid, fname, vname, vtype, dv, rv, iv ) |
| 66 |
... |
| 67 |
MNC_VAR_READ( myThid, fname, vname, var ) |
| 68 |
|
| 69 |
MNC_FILE_SYNC( myThid, fname ) |
| 70 |
MNC_FILE_CLOSE( myThid, fname ) |
| 71 |
|
| 72 |
|
| 73 |
Heres a further "convenience wrapper" written on top of the above UI: |
| 74 |
|
| 75 |
MNC_CW_INIT( myThid, Gtype, Htype, Hsub, Vtype, Ttype, wHalo ) |
| 76 |
|
| 77 |
with pre-defined - xy - - n |
| 78 |
combinations: U x c t y |
| 79 |
'Cen_xy_c_t_Hn' V y i |
| 80 |
'U_xy_i_t_Hn', Cen |
| 81 |
'Cor_x_-_-_Hy' Cor |
| 82 |
|
| 83 |
MNC_CW_WRITE( myThid,myIter, filebn,bi,bj, Gtype, RX, vname, var ) |
| 84 |
MNC_CW_READ( myThid,myIter, filebn,bi,bj, Gtype, RX, vname, var ) |
| 85 |
|
| 86 |
|
| 87 |
|
| 88 |
To-Do: |
| 89 |
====== |
| 90 |
|
| 91 |
1) NAMING -- We should (as much as possible) try to name variables so |
| 92 |
that they are in agreement with the CF naming conventions. |
| 93 |
|
| 94 |
2) UNITS -- as with the names we need to follow conventions |
| 95 |
|
| 96 |
3) AM described her "diags" (or "myDiags" or "mDiags") interface |
| 97 |
which should use MNC for output. The data storage idea is similar |
| 98 |
to the MNC tables-of-indicies approach but also includes one huge |
| 99 |
double-precision "accumulator" to hold all the temporary values |
| 100 |
(eg. partial sums for averages, current max/mins): |
| 101 |
|
| 102 |
vname ( ni ) |
| 103 |
vlen ( ni ) |
| 104 |
vind ( ni ) ------+ |
| 105 |
| |
| 106 |
|
| 107 |
vij_diag ( i, j, [...] ) w/ lat/lon indicies |
| 108 |
vgl_diag ( [...] ) wo/ lat/lon indicies (global) |
| 109 |
|
| 110 |
4) CNH pointed out that grid interpolation needs to be handled |
| 111 |
"on-the-fly" since pre-processing would result in overly large |
| 112 |
input files. We need an interpolation API... |
| 113 |
|
| 114 |
5) From the group meeting on 2004/01/21, we need to define |
| 115 |
"sub-grids" corresponding to: |
| 116 |
|
| 117 |
var_name HGRID VGRID TIME |
| 118 |
=========================================== |
| 119 |
u U_xy r_c t |
| 120 |
eta T_xy - t |
| 121 |
rac T_xy - - |
| 122 |
|
| 123 |
And write a convenience wrapper so that users can write variables |
| 124 |
using just two function calls. JMC and I worked out the |
| 125 |
following: |
| 126 |
|
| 127 |
|
Parent Directory
| 
Revision Log
| 
Revision Graph