Re: [netcdfgroup] [netCDF #KJX-899747]: Ragged Arrays in Fortran and netcdf4

• To: support-netcdf@xxxxxxxxxxxxxxxx
• Subject: Re: [netcdfgroup] [netCDF #KJX-899747]: Ragged Arrays in Fortran and netcdf4
• From: Roy Mendelssohn <Roy.Mendelssohn@xxxxxxxx>
• Date: Thu, 17 Apr 2008 19:06:38 -0700

Thanks!

-Roy

On Apr 17, 2008, at 6:40 PM, Unidata netCDF Support wrote:

On Apr 10, 2008, at 10:52 AM, Unidata netCDF Support wrote:

Hi All:

I am working on creating a netcdf4 file with the structure we are
trying to achieve.  I have a file with Box10 groups and B0x2

subgroups, works great. Now I am trying to do ragged arrays. Below

is the only fortran example I can find that does ragged arrays.  It
is only writing one,1-dimensional array, so I am unsure what I do
when I am writing an N-dimensional ragged array.   Do I just write
each "column" in a loop - if so how does it know that it is writing
to the same structure?  Or if I can do multiple "columns" at once,

Howdy Roy!

To answer some more of your questions, you define your vlen type, and then create a variable of that type, with whatever dimensions you want, 1D, 2D, 3D, whatever.

Then you call the nf_put_vara() and nf_get_vara() functions to write and read the data. You give them the address of the array of data, whether in fortran or C.

If your fortran compiler's memory map of the data are the same as the C version, everything will just work.

There is one caveat: this is untested. I have never tried this from fortran.

There is one twist: you must free each VLEN separately or you will have memory leaks. There is a function nf_free_vlen() for this.

how do I set that up. More realistic examples would be very helpful,

including an example that writes an N-dimensional ragged array, and
an example that adds one more column to an existing ragged array.
Even just the fortran equivalent of the program tst_vl.c would be
very helpful

I agree with you that there needs to be a realistic example, but my experience is that even more necessary are some basic tests. The fortran version of tst_vl.c, as you say. We are not there yet...

I will add some tests for this after my upcoming beta2 release (which will happen any day now). As for the scientifically realistic example, well, perhaps we should be asking you for that! ;-)

It is clear how to do it in C (I set up an array of structures of the

given form), but not in either version of FORTRAN.

It works like this: the f77 layer is just the a bunch of C wrapper shells around the C API functions, with a little macro magic to convert C-like 0-based indexes into F77-like 1-based indexes, and to reorder dimensions so that the first C dimension will come last, etc.

Then the F90 API is built right on top of those F77 calls. Just look at the F90 code and you will see that most of the functions are just F90 wrappers around the F77 function.

It is the same in Fortran as C, except that the HDF5 build knows about the C structs, but not about the fortran structs. So it guarantees to work from C, but not from fortran. However, if fortran uses the same layout for the data in memory as the C compiler does (including any rules for padding or boundaries, or whatever), then everything will work. (Because everything is happening at the C layer anyway.)

But the code will not be portable, because if another fortran compiler does something different, the data will not look the same in memory.

(This portability problem does not matter once the data are written. HDF5 uses a known system of storage, which is always the same, no matter what platform you are on. So once you get the data into the HDF5 file, it is portable. It's just the code that gets the data into the file is not portable.)

Does this shed any light on your problem space?

Thanks!

Ed


Ticket Details
===================
Ticket ID: KJX-899747
Department: Support netCDF
Priority: Normal
Status: Open

**********************

**********************
Roy Mendelssohn
Supervisory Operations Research Analyst
NOAA/NMFS
Environmental Research Division 
Southwest Fisheries Science Center
1352 Lighthouse Avenue
Pacific Grove, CA 93950-2097

e-mail: Roy.Mendelssohn@xxxxxxxx (Note new e-mail address)
voice: (831)-648-9029
fax: (831)-648-8440
www: http://www.pfeg.noaa.gov/

"Old age and treachery will overcome youth and skill."