>> >This problem occurs on NetBSD 1.5.3.  It seems that current and 1.6
>> >are safe.
>> 	i committed it as the problem will re-appear with gcc 3.1.1, i guess.
>Oh, boy!  gcc has revived its bug!?

	see http://www.jp.netbsd.org/ja/JP/ml/port-powerpc-ja/200207/msg00029.html
	and the thread follows.  gcc 3.1.1 will take advantage of ANSI pointer
	aliasing rules, and we have to use unions to be -O2 safe.

itojun