[BUG] a doozy (resend)

From: Squeak <squeak_at_xirr.com>
Date: Sun, 10 Oct 1999 15:21:27 -0400 (EDT)

[I hope this is the first time this message hit the list, last time
there were some errors that were unclear as to whether the message was
sent]

ln(499*ln(2))^(1/3) = 1.801 ish (verified on a calculator)
ln(551*ln(2))^(1/3) = 1.812 ish (verified)
5.9^(1/3) = 1.807 ish (verified)

ln(500*ln(2))^(1/3) = crash/hang .. gdb'ing shows caught in Loop() but
with the stack smashed beyond recognition.
ln(550*ln(2))^(1/3) = crash/hang

ln(500*ln(2)) = 5.94 ish
ln(550*ln(2)) = 5.85 ish ... i don't really see that these should be
special numbers...

(ln(500)+ln(ln(2)))^(1/3) also crashes .... also if you use 2/3's
3/4's 1/4 2/4 are all good ... 1,2,3,4/5 are good 2/6 crashes, 3/9
crashes all 1,2,3,4,5,6/7 are fine....

Tested on genius-0.4.4, genius-0.4.5 genius-0.4.6 and current cvs...

Very repeatable (100%)

I'm uhh mystified so no patch...

One thing that might have caused the prob ... it appears that raising it
to the 1/3 is causing a call to mympw_pow_z where it spends quite a bit of
time... dunno really tho...

If anyone is out there, could you confirm the bug? I have a amdk6-2 using
egcs-1.1.2 (for older geniuses) and gcc-2.95.1 glibc-2.1.2 and
binutils-2.9.5.0.12 and binutils-2.9.1.0.23 ...
Received on Sun Oct 10 1999 - 11:52:15 CDT

This archive was generated by hypermail 2.2.0 : Sun Apr 17 2011 - 21:00:02 CDT