Boban Acimovic was kind enough to give me access to a Solaris 8 system
to track down a reproducible server crash. What I find is that strxfrm
is buggy on that system. Given locale is_IS.ISO8859-1, the call
strxfrm(<ptr>, "pg_amop_opc_strategy_index", 58)
was observed to scribble on 108 bytes of memory at <ptr>, not the 58
that it was allowed to. This naturally led to death and destruction
upon next use of the adjacent data structures.
I don't know yet whether this is a known/repaired problem, or whether
it occurs in any locales besides Icelandic. But I thought I'd give
the list a heads-up. If anyone recognizes this bug, more info would
be appreciated.
regards, tom lane