Thread: reproducible database crash with simple sql command on postgres 8.3.1

hi all,

first i create a table
create table regions (id integer, name varchar);

then i want to set a default value for a column, e.g.
alter table regions alter column name set default 'bavaria';

at this point crashes the database with the message

PANIK: ERROR_STACK_SIZE exceeded

the rest of the message is unfortunately in german then i have to
restart the postgres-service manually this error is very easy
reproducible at my environment

is this a problem of my installation or can anybody reproduce this error
at his environment

thanks and greetings from munich
christian

p.s.     os windows xp, sp2
    postgres 8.3.1, visual c++, build 1400

No problems here.


create table regions (id integer, name varchar);
alter table regions alter column name set default 'bavaria';


PostgreSQL 8.3.1 vc++ build 1400 xp sp2

# - Memory -

shared_buffers = 32MB            # min 128kB or max_connections*16kB
                    # (change requires restart)
#temp_buffers = 8MB            # min 800kB
temp_buffers = 8MB            # min 800kB

#max_prepared_transactions = 5        # can be 0 or more
                    # (change requires restart)
# Note:  Increasing max_prepared_transactions costs ~600 bytes of shared memory
# per transaction slot, plus lock space (see max_locks_per_transaction).
#work_mem = 1MB                # min 64kB
work_mem = 2MB                # min 64kB
#maintenance_work_mem = 16MB        # min 1MB
maintenance_work_mem = 32MB        # min 1MB
max_stack_depth = 2MB            # min 100kB




cheers


Josep

<Christian.Strobl@dlr.de> writes:
> PANIK: ERROR_STACK_SIZE exceeded

> the rest of the message is unfortunately in german then i have to
> restart the postgres-service manually this error is very easy
> reproducible at my environment

What this typically means is that you've got an encoding/locale
configuration problem.  The thing is trying to report some error
or other, and it discovers that the error text is badly encoded
according to the current database encoding, and it tries to
report that, only *that* error text is badly encoded too, so
it tries to report that, getting another error, etc, until the
error recovery stack overflows.

The hope of getting rid of this scenario is one reason why we locked
down encoding/locale combinations in 8.3, so I'm disappointed that
you're still able to make it happen.  You didn't say what settings
you are using though.  Could we see

    show lc_messages;
    show lc_ctype;
    show server_encoding;
    show client_encoding;

            regards, tom lane