Re: [bufrtables] More on table versions

Hi Milan:

Sorry I was unclear about the checksum idea. The checksum is on the tables, not 
on the data. The intention is only to ensure that the coder and the decoder use 
the same tables. You would have to have a central place where tables can be 
registered, and a coder would have to make sure that their tables were 
registered. The registration center (eg WMO) would generate the checksum, and 
the coder would put it into the BUFR message. A decoder would read the message, 
get the checksum, see if they already had those tables locally, and if not, 
request them from the registration center using the checksum.

My fear is that, without something like that, after a few years BUFR messages 
will no longer be reliably decoded. Or they can only be decoded by the original 
software that generated them, which becomes harder and harder to maintain.

Regards,
John

Milan Dragosavac wrote:
Hi John,

Unfortunately something like this can not be done because of various reasons. The tables are loaded based on the information in section 1 and expansion of data descriptors can start. Then one can find if really local entries are used. Checksum can not help because of possible compression and in this case the size of data vary. The Ecmwf experience
show that you have to know anyway what data you want to process and than
before you start operational usage links are created in advance if needed. It is not too bad situation. In our preprocessing we repack all the data and use usually one version of the tables for all data afterwards.

Regards

Milan
Milan Dragosavac

ECMWF
Shinfield Park, Reading, Berkshire, RG2 9AX, UK

Tel: (+44 118) 949 9403
Fax: (+44 118) 986 9450
Telex: 847908 ECMWF G
E-mail: milan.dragosavac@xxxxxxxxx