GH-75: added a note.

morfologik-speller/src/main/java/morfologik/speller/Speller.java

@@ -435,8 +435,9 @@ private void findRepl(final int depth, final int node, final byte[] prevBytes, f
       byteBuffer.flip();
       decoder.reset();
       // FIXME: this isn't correct -- no checks for overflows, no decoder flush. I don't think this should be in here
-      // too, the decoder should run once on the input charsequence, then the decoded byte array should be used for 
-      // suggestions. Reallocating buffers and decoding over and over doesn't make much sense.
+      // too, the decoder should run once on accumulated temporary byte buffer (current path) only when there's
+      // a potential that this buffer can become a replacement candidate (isEndOfCandidate). Because we assume candidates
+      // are valid input strings (this is verified when building the dictionary), it's save a lot of conversions.
       final CoderResult c = decoder.decode(byteBuffer, charBuffer, true);
       if (c.isMalformed()) { // assume that only valid
         // encodings are there

morfologik-stemming/src/main/java/morfologik/stemming/BufferUtils.java

@@ -74,6 +74,13 @@ public static String toString(ByteBuffer buffer, Charset charset) {
     return new String(buf, charset);
   }
 
+  public static String toString(CharBuffer buffer) {
+    buffer = buffer.slice();
+    char [] buf = new char [buffer.remaining()];
+    buffer.get(buf);
+    return new String(buf);
+  }
+
   /**
    * @param buffer The buffer to read from.
    * @return Returns the remaining bytes from the buffer copied to an array.