Changeset 1238

Timestamp:

2000-06-26T11:31:38+12:00 (24 years ago)

Author:

gwp

Message:

Additional explanation of hash problems in documentation.

File:

• trunk/gsdl/src/hashfile/hashfile.cpp (modified) (1 diff)

trunk/gsdl/src/hashfile/hashfile.cpp

@@ -159 +159 @@
 // and (z / 16 = 1) which means (z = 1).
 //
-// POSSIBLE CAUSE: Characters in the file are read one at a time; after each 
+// MAJOR REASON: suppose our prime number was 19.  Then, if we have a 
+// reasonably random distribution of numbers for which we are going to 
+// calculate (N % 19), we expect to get a roughly uniform distribution
+// of remainders where the possible values are 0, 1, 2, 3... 18.  The
+// problem is that 9 out of the 19 possible values (10,11... 18) start
+// with the digit 1.  Thus our hascode will start with "01".
+//
+// ANOTHER PROBLEM: Characters in the file are read one at a time; after each 
 // one is read it is prepended to the remainder, then the remainder is
-// recalculated on the string thus far seen.  I think the math here 
-// may be just plain wrong.
-//
-// ANOTHER POSSIBLE CAUSE (this appears to be a genuine bug): Each character 
-// from the file is read into  remainder at the most significant end, and
-// when that character is a zero you get a number like "01" which would
-// be considered larger than a number like "8" because it is longer (two
-// digits instead of 1).  
-//
-// These comments added by Gordon Paynter ([email protected]) on
-// June 23, 2000.  I didn't write this code, however.
+// recalculated on the string thus far seen.  I am sure the math here 
+// is wrong - if I try calculating (111 mod 7) by the same algorithm, it
+// simply does nor work.
+//
+// ANOTHER POSSIBLE PROLEM: Each character from the file is read into 
+// remainder at the most significant end, and when that character
+// is a zero you get a number like "01" which would be considered 
+// larger than a number like "8" because it is longer (two  digits
+// instead of 1).  
+//
+// These comments added by Gordon Paynter ([email protected]) in
+// June 2000.  I didn't write any code, however.
 
 char *hashfile (char *filename) {