Regarding critical fft bug

[Torbjorn Granlund]

Kate Minola <kminola at po0.glue.umd.edu> writes:

  Two questions regarding the critical bug in FFT multiply
  code and the recommendation that gmp be recompiled with
  --disable-fft.
  
  1. Is there a way to check if an installation of gmp has
  been built with --disable-fft?
  
You can use nm on the library file and look for the string "fft".

  2. What size of numbers does this bug affect?
  
That depends on the target.  Typically from 20000-50000 decimal
digits.

  For example, if I have been doing computations with numbers
  with 100 decimal digits, should I go back and redo the
  computations because the fft bug might have affected the
  computations?
  
No problems for such small operands.


We're actually no longer completely sure there is any bug.

The bug happened in the checked-in code until Kevin made a fix
recently.  It was not hard to trigger problems with that code
before the change.  Looking at 4.1.2, the same bug should affect
its version of the fft code.  But I have not been able to find
any operands that actually triggers problems with 4.1.2.

The author of this code has not responded yet.  Until he does, or
until Kevin or I have time to make a careful revision of the FFT
code, our recommendation will be that it is not used.

--
Torbjörn