[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: 20050922:DODS/THREDDS confusion




On Feb 24, 2006, at 2:05 PM, David Wojtowicz wrote:

I'm happy to work on it... though I'm having trouble understanding what the STRING_AS_ARRAY stuff is there for. Is this an attempt to fix the problem?
Turning on debugging, I don't see many of its parts being invoked when
strings are encountered (such as in NCConnect::parse_array_dims())

It is an attempt to solve the problem, but it works only for Str --> char[], not that case where there's an array of Strings going to a two dimensional array of characters.

Take a look in the unit-test subdir for some code that will exercise this.

James



-----Original Message-----
From: address@hidden
[mailto:address@hidden] On Behalf Of James Gallagher
Sent: Thursday, February 23, 2006 4:52 PM
To: David Wojtowicz
Cc: 'Ethan Davis'; address@hidden;
address@hidden; 'support-thredds'
Subject: Re: 20050922:DODS/THREDDS confusion

David,

It would be fantastic if you could look into fixing this! If you need
some help, let me know. After the 6th of March I should be able to
really dive into this.

James

On Feb 23, 2006, at 2:35 PM, David Wojtowicz wrote:

I've pinned the problems down somewhat.

 The actual netCDF file that contains the data on the TDS, has
string data
that looks like this (from dncdump on the local file):

---------------------------------
dimensions:
   station = 4818;
   id_len = 4;
variables:
   char station_id(station, id_len) ;
      station_id:long_name = "Station id" ;
      station_id:reference = "sfmetar_sa.tbl" ;
---------------------------------



However, the TDS represents it via DAP as (pertinent bits of DDS
and DAS):


---------------------------------
Dataset {
    String station_id[station = 4818];
}
Attributes {
    station_id {
        String long_name "Station id";
        String reference "sfmetar_sa.tbl";
        DODS {
            Int32 strlen 4;
            String dimName id_len;
        }
    }
}
---------------------------------

So, the two dimensional character array has been converted to a single
dimensional array of strings, whose length is stored in the nested
attribute
"DODS".

When you do dncdump on the dods URL, you get:

---------------------------------
dimensions:
        station = 4818 ;
variables:
        char station_id(station) ;
                station_id:long_name = "Station id" ;
                station_id:reference = "sfmetar_sa.tbl" ;
                station_id:DODS =
dncdump: Attribute not found  (program terminates)
----------------------------------


Dncdump aborts when it gets to the nested attribute, "DODS" because
a nested
attribute is represented as a container and the Attr_container type is
something that the code in Dattr.cc can't deal with.


Now, even if it could deal with nested attributes, there'd still be
the
problem of figuring out what to do with the strings.

I temporarily hacked around the Attr_container problem and had it
pretend it
was an empty character array....i.e.  station_ID:DODS = "" ;
essentially
making it harmless to the rest of the netCDF API.

dncdump ran through then without aborting, but if I asked it to dump
"station_id"  (-v station_id option) it'd get the strings all wrong
because
the second dimension, id_len, isn't there.

So, there are two problems:

 1) Can't deal with nested attributes (i.e. containers)
 2) Can't deal with DAP strings


Ideally the output from the DAP network version should look just
like the
output from the local file version... the differences become
transparent.

This would involve fixing the attribute handling code to not pass
through
the nested attribute containers to the netCDF API (like the Java
version
does).

It would also involve converting the string arrays back to two-
dimensional
character arrays as far as the netCDF API is concerned.


I've already been hacking around in the code and thinking about
this for a
while.   If you'd like I could take a first shot at it.


-----
David Wojtowicz, Sr. Research Programmer, Sysadmin
Dept of Atmospheric Sciences / Computer Services
University of Illinois at Urbana-Champaign
address@hidden  (217) 333-8390

-----Original Message-----
From: James Gallagher [mailto:address@hidden]
Sent: Thursday, February 23, 2006 10:14 AM
To: Ethan Davis
Cc: David Wojtowicz; address@hidden;
address@hidden; 'support-thredds'
Subject: Re: 20050922:DODS/THREDDS confusion

This is in Trac as ticket 320. I'll try to get a fix out by the 6th
of March.

James

On Feb 23, 2006, at 7:02 AM, Ethan Davis wrote:

Hi David,

Yeah, this has been an issue for some time (seems to have started
on this list in 2003, http://www.unidata.ucar.edu/support/help/
MailArchives/dods-tech/msg01910.html).

The underlying issue is that netCDF-3 doesn't understand String, it
only knows char and arrays of char, and OPeNDAP knows String but
not char. So, the mapping between netCDF-3 and OPeNDAP is a bit
thorny. The current nc servers map char arrays to String arrays (so
each String contains one character). The TDS maps char arrays into
Strings. The problem is that the nc client library doesn't know how
to map Strings back into char arrays just from the DDS/DAS because
the length of Strings can vary. (The TDS contains some hints in the
DAS, that is how the netCDF-java library knows how to deal with
them.)

I don't know if this comes up anywhere else. But it will be an
issue for any OPeNDAP dataset that contains String arrays. Do any
of the other servers map things to String arrays?

Sorry I don't actually have any answers but since this is in some
ways an issue for the TDS, I'll look into this some more.

Ethan

--
James Gallagher                jgallagher at opendap.org
OPeNDAP, Inc                   406.723.8663



--
James Gallagher                jgallagher at opendap.org
OPeNDAP, Inc                   406.723.8663



--
James Gallagher                jgallagher at opendap.org
OPeNDAP, Inc                   406.723.8663