[coyotos-dev] BigNum issues

[Thomas Stratmann]

There's a number of problems in the shifting functions.

vu_rshift_digits has an off-by-some problem (see attached index.patch).

vu_rshift_bits is broken. First it shifts n1 by nShift / 32 digits, 
result in n2, then puts n1 bitshifted by nShift %= 32 into n2.
Similarly in vu_lshift_bits.

ALL bit and digit shifting functions are lacking requirements, which is 
why I'm answering this thread... and the reason why I didn't bother 
fixing for now.

vu_lshift_digits is not zeroing n2's leading digits in case n1 is too 
short (this may be just a missing requirement).

/--/

Additionally, have a look at BigNum::lshift_digits(size_t n):
It depends on alloca zeroing the allocated memory (which I suspect it 
does, but failed to investigate). If this is the case, the rmemset calls 
in vu_mul (and I think I spotted this somewhere else...) are superfluous.

/--/

Finally, constructing an nvec from a BigNum should result in a constant 
since it borrows the BigNum's innerds. Also the global nvec 0's and 1's 
representation should be constant. See attached noclobber.patch.

Thomas
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: index.patch
Url: http://www.coyotos.org/pipermail/coyotos-dev/attachments/20071008/ca95751d/attachment.pl 
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: noclobber.patch
Url: http://www.coyotos.org/pipermail/coyotos-dev/attachments/20071008/ca95751d/attachment-0001.pl