mpz_invert behavior -addendum

[Torbjorn Granlund]

Benjamin Eltzner <b.eltzner at gmx.de> writes:

  To make my discomfort clearer: note that all of the following are "true"
  
  a) for all n: 0 * n = 0 mod 1    (rop = indef, return = 0)
  b) for all n: 0 * n = 1 mod 1    (rop = 0, return != 0)
  c) for all n: 1 * n = 0 mod 1    (rop = indef, return = 0)
  d) for all n: 1 * n = 1 mod 1    (rop = 1, return != 0)
  
I don't follow how to relate the current GMP behaviour to your table.
To me the example a and b are the same just like c and d are the same.
(The number after the = sign is not input to GMP, so how you can find
different behaviour for these cases is a mystery.)

I believe we always return 0 as the result of any operation in the zero
ring.  If you disagree, please show a concrete counter-example.  In C
code.
  
  I can imagine there has been a lot of discussion around this and you
  are probably tired of the issue, so I would just like to ask you to
  please update your documentation. (However, should you choose to
  reconsider, I would be very glad if you reverted the  behavior to case a).)
  
I believe the current behaviour is mathematically correct.  The
behaviour of older GMP was not mathematically correct in the zero ring.

I agree that there is a slight documentation problem for mpz_invert wrt
the zero ring.

Torbjörn
Please encrypt, key id 0xC8601622