Subject AMD64+Classic LI-V1.5.3.4870+Ubuntu - Segfault on arithmetic overflow/sting trun
Author wobbleian
I must have a knack for it...continuing my deployment of Ubuntu+AMD64
we have another problem. This time it's not so serious as we shouldn't
have caused the problem in the first place if everyone here was
reading the same development docs!

When we overflow a string we get a segfault, and maybe some other
errors logged. Other incorrect sql conversions go ok:

SQL> create table tmp ( tmp_string varchar(1), tmp_date timestamp,
tmp_integer integer);

SQL> insert into tmp (tmp_date) values ('this is not a date!');
statement failed, SQLCODE = -413

conversion error from string "this is not a date"

SQL> insert into tmp (tmp_integer) values ('not a number');
Statement failed, SQLCODE = -413

conversion error from string "not a number"

SQL> insert into tmp (tmp_string) values ('more than 1 char');
Statement failed, SQLCODE = -902

Unable to complete network request to host "localhost".
-Error reading data from the connection.

In /var/log/messages we mostly get:
fb_inet_server[24921]: segfault at 00000000000000a0 rip
00002aae06c82219 rsp 00007fffa40d4cf0 error 4

and also the occasional:
fb_inet_server[24907] trap invalid opcode rip:2abdc83f61a0
rsp:7fffe2961580 error:0
fb_inet_server[24891] general protection rip:2b203e089295
rsp:7fff6cd58090 error:0

Nothing goes in firebird.log.

This is a different machine to the previous problem, and anyway we
"fixed" that with a fresh build to solve what we thought of as binary
library problems.

I have repeated this on 2 separate, but very similar servers.

Any help would be appreciated, as always. And any further details can
be provided if needed. I have one of the servers sitting on the desk
next to me so can do nearly anything to it if need be - including
hitting it.