> Tatsuo Ishii <t-ishii@sra.co.jp> writes:
> > For the regression test, I got 7 failures, most of them seem harmless,
> > the only concern I have is bit test though.
>
> Most of the diffs derive from what I recall to be a known SunOS problem,
> that strtol fails to notice overflow. A value that should be rejected
> is getting inserted into int4_tbl (mod 2^32 of course).
>
> The bit test diffs seem to indicate that bit_cmp is messed up. That
> depends on memcmp. I seem to recall something about memcmp not being
> 8-bit-clean on SunOS ... does that ring a bell with anyone?
Good point. From the man page of memcmp(3) on this machine:
BUGS memcmp() uses native character comparison, which is signed on some machines and unsigned on other
machines. Thus the sign of the value returned when one of the characters has its high-order bit set is
implementation-dependent.
--
Tatsuo Ishii