Legend:
	- = open bug
	* = fixed bug

3.1.0:
	* Error in final result computation of mpextgcd_w in certain circumstances.
	* PowerPC 64-bit assembler symbols aren't defined according to LSB 1.3.
	
3.0.0:
	- Can't seem to generate 64-bit shared libraries on AIX; use
	  --disable-shared on this platform for now.
	- Intel icc can't cope with gcj headers. There's also a problem in
	  combination with aio.h; solution should be to not test gcj when using
	  this compiler. As a workaround, you can specify --without-javaglue.
	- GCC 3.3 produces faster output for Blowfish on Pentium 4 than the
	  included assembler source; try coding two Blowfish rounds (without swap)
	  in C and compile to assembler to see how GCC accomplishes this.

2.3.0pre:
	* bug in certain instances of computing modular inverse.
	* incorrectly translated intel-style assembler version of x86 mp32odd
	  and mp32even routines into gnu assembler format.
	* base64 decoding of '+' and '/' characters is wrong.
	* incorrect testing of the result of the gcd operation in mp32prndconone.
	* Chinese Remainer Theorem for RSA private key operation doesn't work
	* incorrect header file inclusion when --disable-threads is selected or
	  when no multithreading is available.

2.2.0:
	[ Released by Jeff Johnson of RedHat, Inc.; lint & doxygen enhanced version
	  of 2.1.0. ]

2.1.0:
	* Entropy gathering from /dev/dsp contained error; size in words of entropy
	  data to gather was overwritten with audio sample size - which resulted in
	  much less entropy.

2.0.0:
	- Chinese Remainder Theorem for RSA private key operation doesn't work; it
	  needs fixing.

1.1.2:
	* SHA-1 i586 assembler code uses local variables below the current stack
	  frame.
	
1.1.1:
	* discrete logarithm domain parameter generator calculations contains bugs
	  - was detected in this release but present since the first release.

1.1.0:
	* The javaglue produces a NullPointerException when initializing with a
	  null IV; this should be treated correctly, i.e. as an all zero IV.

1.0.2:

1.0.1:
	* The Windows 2000 bug is still around

1.0.0:
	* On Windows 2000, the entropy system gets error WAVERR_BADFORMAT in
	  waveInOpen; So far I've been unable to determine why the system does this
	  for format WAVE_FORMAT_PCM. Suggestions to fix this problem are more than
	  welcome.
	* The assembler code for SHA-1 in gnu/fips180opt.gas.i586.s contains bugs.