Request for information on stack usage

[John D. Shortridge]

Dear gprolog users,

I have a question which I hope someone may be able to help with. I am using gprolog 1.2.16 under Linux to decode meteorological messages, and a wonderful tool it is. However, in scaling up from test volumes of data to real-life volumes I have run into a problem with stack sizes. For example, my program will process 4000 records OK, but then if I duplicate those 4000 records and try to process the resulting 8000 I will get a fatal error due to local stack overflow. Then if I adjust LOCALSZ and try again I get a problem with global stack overflow. Then if I adjust GLOBALSZ I can process 8000 records OK, but when I duplicate them again and try to process 16000 it's back to local stack overflow. Clearly this is a problem which should be solved properly rather than worked around.

The program is written in C and Prolog, with the C program simply reading individual records from a flat file and passing each individual message to Prolog for decoding. In general terms I would have thought that each invocation of Prolog would start with a clean slate, but obviously that isn't the case and some resource is being chewed up behind the scenes.

In the archive, http://mail.gnu.org/archive/html/bug-prolog/2001-11/msg00005.html seems to describe my problem more or less exactly, but unfortunately following "Thread next" just leads to a joke about a blonde :-(.

I'd be grateful for any general guidance anyone can offer on what I'm doing wrong.

John

(John Shortridge,
Computing Support Unit,
National Climate Centre,
Australian Bureau of Meteorology,
GPO Box 1289K,
Melbourne, Vic 3001
Australia)