standard error handling for malloc() broken for user root and group wheel

[Bruce Evans]

On Thu, 19 Feb 2004, Poul-Henning Kamp wrote:

[The problem turned out to be mainly controlling both undefined behaviour
and standard behaviour with the same flag.]

> Think if the 'A' option as userland SIGSEGV.  (peter@ actually suggested
> that I make it kill(getpid(), SIGSEGV) before calling abort in order
> to let a process install a handler for it.

Programs can already catch SIGABRT, and they probably need to be familiar
with malloc's internals to trap its errors sort of safely, so it is
reasonable to expect them to know what to trap.

> I'm currently testing this patch:

This would fix the original problem and some others.

Bruce