loss of significance in difficulty (by lfm)

For instance any nBits compressed value from 0x1a44b800 thru
0x1a44b9ff will show as difficulty 244139.4816. This patch will
more accurately convert the nBits compressed values to the double
difficulty.

This will display any of the recent difficulty levels slightly
differently though. Early difficulties and testnet difficulties are
not large enough to trigger this bug.

None of the actual targets or compressed targets are changed, only
the conversion to the floating point difficulty is changed and afaik
it is only ever displayed, never converted back so the patch does not
effect the target calculations, binary files, databases nor the binary
protocol.

src/rpc.cpp

@@ -199,12 +199,26 @@ double GetDifficulty()
 {
     // Floating point number that is a multiple of the minimum difficulty,
     // minimum difficulty = 1.0.
+
     if (pindexBest == NULL)
         return 1.0;
-    int nShift = 256 - 32 - 31; // to fit in a uint
+    int nShift = (pindexBest->nBits >> 24) & 0xff;
-    double dMinimum = (CBigNum().SetCompact(bnProofOfWorkLimit.GetCompact()) >> nShift).getuint();
+
-    double dCurrently = (CBigNum().SetCompact(pindexBest->nBits) >> nShift).getuint();
+    double dDiff =
-    return dMinimum / dCurrently;
+        (double)0x0000ffff / (double)(pindexBest->nBits & 0x00ffffff);
+
+    while (nShift < 29)
+    {
+        dDiff *= 256.0;
+        nShift++;
+    }
+    while (nShift > 29)
+    {
+        dDiff /= 256.0;
+        nShift--;
+    }
+
+    return dDiff;
 }
 
 Value getdifficulty(const Array& params, bool fHelp)