1998-10-27   Gisle Aas <aas@sn.no>

   Release 1.99_57

   Rewritten most of the MD5 C code to make it real fast (especially
   on little-endian machines).  Compared to MD5-1.7 we can process
   files 4 times as fast and we digest small stuff in memory 7 times
   faster.  I came to these conclusions after these tests (gcc -O2,
   i586, Linux):

   First tested calculation of the digest of a 31 MB file, using
   perl -le 'print Digest::MD5->new->addfile(*STDIN)->hexdigest'
   and similar stuff:

      MD5-1.7:                 21.06s
      Digest::MD5-1.99_57:      5.23s
      md5sum (GNU textutils):   4.90s

   As you can see, we do nearly as good as the md5sum program.  I
   think the reason we don't beat md5sum is that perl always insist on
   loading extra modules like Config.pm, Carp.pm, strict.pm, vars.pm,
   AutoLoader.pm and DynaLoader.pm.  When I simply wrapped the MD5.xs
   hasher code in a C program I managed to process the file in 4.68s.

   Then we calculated the digest of the same 6 byte sting, 20000

      MD5-1.7:                 11.81s
      Digest::MD5-1.99_57:      1.68s

   Digest::MD5 benefit from making this into a plain procedure call
   instead of a static method call.

   Other changes in this release are:

   Documentation update

   Internal MD5.xs cleanup.

   $md5->digest will automatically reset now.

   Digest::HMAC methods add() and addfile() did not return the
   corret object.

   Added Digest.pm loading module.  I am not sure this is a good idea.

   Added Digest::SHA1 and Digest::HMAC_SHA1 module.  The Digest::SHA1
   module is just a wrapper around SHA.pm.  I hope to get the author
   of SHA.pm to move his module to the Digest:: category.

1998-10-25   Gisle Aas <aas@sn.no>

   Release 1.99_56

   Fix memcpy_byteswap() function in MD5.xs.  Must be careful with
   htovl() as it might evaluate its arguments more than once.

1998-10-25   Gisle Aas <aas@sn.no>

   Release 1.99_55

   Grahams HMAC_MD5.pm splitted into two modules.  Digest::HMAC and
   Digest::HMAC_MD5.  Also provide functional interface.  Documentation
   is still lacking.

   Included RFC 2202 based test for HMAC-MD5.

1998-10-24   Gisle Aas <aas@sn.no>

   Release 1.99_54

   Included HMAC_MD5.pm, contributed by Graham Barr <gbarr@ti.com>.

   I have a hard time to make up my mind :-)  md5_bin() renamed back
   to md5().   Functions are not exported by default any more.

   Try to Encode/Decode with memcpy_byteswap for 32-bit big-endian

1998-10-23   Gisle Aas <aas@sn.no>

   Release 1.99_53

   Renamed core module as Digest::MD5.  Leave a MD5.pm stub for
   legacy code.

   The md5() function renamed as md5_bin().

   The constructor, Digest::MD5->new, no longer takes any extra

   Added some new tests.

   Updated the documentation.

   $md5->b64digest implemented with same base64 encoder as md5_base64.

1998-10-23   Gisle Aas <aas@sn.no>

   Release 1.99_52

   Patch from Graham Barr which make it work for big-endian machines

1998-10-22   Gisle Aas <aas@sn.no>

   Release 1.99_51

   The MD5 class is now subclassable.

   The add() and addfile() methods now return $self.

   The reset() method is just an alias for new().

   The constructor (MD5->new) now takes optional arguments which are
   automatically added.  It means that we can now write:


   New $md5->b64digest method.

   New functions that are exported on request: md5, md5_hex, md5_base64

   Included RFC 1321

   Barely started to update the documentation.

1998-10-22   Gisle Aas <aas@sn.no>

   Release 1.99_50

   Much better performance (more than twice as fast now).  Mostly
   because we use Copy/Zero instead of the original MD5_memcpy and
   MD5_memset functions.

   The addfile() and hexdigest() methods are now XS implemented.

   All RSA functions now included in MD5.xs and made static.

   Use perl's Copy/Zero.

   Random cleanup, simplifications and reformatting.
   Merged things better with the perl configuration.

Neil Winton's versions below:

*** 96/06/20 Version 1.7

MD5 is now completely 64-bit clean (I hope). The basic MD5 code uses
32-bit quantities and requires a typedef UINT4 to be defined in
global.h. Perl configuration data (the value of BYTEORDER) is used to
determine if unsigned longs have 4 or 8 bytes. On 64-bit platforms (eg
DEC Alpha) then it assumes that "unsigned int" will be a 32-bit type.
If this is incorrect then adding -DUINT4_IS_LONG to the DEFINES line in
Makefile.PL will override this.

On some machines (at least Cray that I know of) there is no 32-bit
integer type. In this case defining TRUNCATE_UINT4 (which is done
automatically for a Cray) will ensure that 64-bit values are masked
down to 32 bits. I have done my best to test this but without easy
access to a true 64-bit machine I can not totally guarantee it (unless
anyone wants to lend me a spare Cray :-)

There is one remaining limitation for 64-bit enabled processors. The
amount of data passed to any single call to the underlying MD5
routines is limited to (2^32 - 1) bytes -- that's 4 gigabytes. I'm
sorry if that's a real problem for you ...

And finally, a minor complilation warning (unsigned char * used with
function having char * prototype) has also been eliminated.

*** 96/04/09 Version 1.6

Re-generated module framework using h2xs to pick up the latest module
conventions for versions etc. You can now say "use MD5 1.6;" and things
should work correctly. MD5.pod has been integrated into MD5.pm and
CHANGES renamed to Changes. There is a fairly comprehensive test.pl
which can be invoked via "make test". There are no functional changes
to the MD5 routines themselves.

*** 96/03/14 Version 1.5.3

Fixed addfile method to accept type-glob references for the file-handle
(eg \*STDOUT). This is more consistent with other routines and is now the
recommended way of passing file-handles. The documentation now gives more
examples as to how the routines might be used.

*** 96/03/12 Version 1.5.2

Minor fixes from Christopher J Madsen <madsen@computek.net> to provide
support for building on OS/2 (and to work arround a perl -w bug).

Remove warning about possible difference between add('foo', 'bar') and
add('foobar'). This is not true (it may have been true in the earliest
version of the module but is no longer the case).

*** 96/03/08 Version 1.5.1

Add CHANGES file to make it easier for people to figure out what has
been going on. (Meant to do this as part of 1.5)

*** 96/03/05 Version 1.5

Add hash() and hexhash() methods at the suggestion/request of Gary
Howland <gary@kampai.euronet.nl> before inclusion in a wider library
of cryptography modules.

*** 96/02/27 Version 1.4

Finally fixed the pesky Solaris dynamic loading bug. All kudos to Ken
Pizzini <kenp@spry.com>!

*** 95/11/29 Version 1.3.1

Add explanations of current known problems.

*** 95/06/02 Version 1.3

Fix problems with scope resolution in addfile() reported by
Jean-Claude Giese <Jean-Claude.Giese@loria.fr>. Basically ARGV is
always implicitly in package main while other filehandles aren't.
*** 95/05/23 Version 1.2.1

[Changes pre 1.2.1 not recorded]