Cafe, Daniel Larsson noticed a correctness issue with the pureMD5 package. This issue would affect you if you built the value incrementally via the 'updateMD5' function (vs just using 'md5') and didn't provide 512 bit long bytestrings (an MD5 block of operation). As you can probably tell, I didn't invest enough into the non-performance aspects of pureMD5. Faced with actual users ;-), I have released version 0.2.0 which has the bug fix, a new API (type prevention from re-finalizing a digest), and a reasonable set of quickchecks (covering Show / Binary instances, known answer and incremented hashing). Oh, also the module name has changed to place it inline with 'Crypto' package naming while not colliding. Sorry if this causes anyone headaches. Daniel, Good catch, I hope this didn't consume much of your time. Thanks a bunch! TomMD On Fri, 2008-06-13 at 04:14 +0200, Daniel Larsson wrote:
Hi Thomas,
I was fiddling around with your pureMD5 package, and encountered some problems with using the md5Update/md5Finalize sequence. I tried to calculate the md5 of some scattered data, but it kept returning the wrong values. It seems that each md5Update must supply an exact blockSize number of bits, since the mdLeftOver part isn't taken into account in subsequent calls to md5Update.
I wrote a small patch, and a simple QuickCheck property, to support calculating md5 of scattered data, attached to this mail. Hopefully I didn't mess up something...
-- Daniel