NEWS

Release 2.0.5
=============

 * Fix SBCL-specific sequence copying implementation to use better
   suited internal interface.

Release 2.0.4
=============

 * Clarified public domain status for code authored by Pierre R. Mai
   through additional CC0 waiver.

Release 2.0.3
=============

 * Add support for md5sum-string for other implementations through
   through flexi-streams.  Patch supplied by Daniel Kochmanski.


Release 2.0.2
=============

 * This release adds support for LispWorks 7.0, reworking the fixes
   for 64-bit implementations of Lispworks in 2.0.1 and incorrect
   use of the sys:simple-int32-vector type.  Patch supplied by
   Martin Simmons from LispWorks.

 * Minor fix to prevent style-warnings for implementations with no
   support for md5sum-string.


Release 2.0.1
=============

 * This release fixes problems on 64-bit implementations of Lispworks,
   where sys:int32 arithmetic can overflow 32-bit values and hence
   yield wrong results.


Release 2.0.0
=============

 * This release consolidates the various versions of md5 that have
   been floating around, including versions with Lispworks
   optimizations, the version maintained by Kevin M. Rosenberg, the
   SBCL-optimized version, and various patches and fixes.

 * This release tries to separate out string-handling from actual
   byte-based MD5 hash generation.  Hence users who want to generate
   hashes from strings should use the new md5sum-string function,
   which will go through your implementation's external-format
   handling to generate the actual byte-array that is hashed.  Usage
   of md5sum-sequence for strings and md5sum-stream for character
   streams is deprecated, since it only ever worked correctly for 8bit
   characters in most implementations.

 * This release should work correctly on at least recentish versions
   of CMUCL, SBCL, ECL, CCL, Lispworks, AllegroCL and ABCL and
   generally most other conforming implementations.  It has been
   optimized for CMUCL, SBCL and Lispworks, and should work with
   adequate performance on most 64bit implementations.  Performance on
   other 32bit implementations is still sub-optimal if they don't
   support unboxed bit-operations on (unsigned-byte 32).