X-Git-Url: http://mj.ucw.cz/gitweb/?a=blobdiff_plain;f=lib%2Flizard.c;h=6167fee043ba90c07befef746cd6bd917c5c21d7;hb=ff36b07f44efa12a78809ee05bd6d0c25fc60495;hp=ee2e245cd199992450ba7684de0cc08981a591a7;hpb=8bf50727d2f7427cc9d6d04284de4c564d9f9d36;p=libucw.git

diff --git a/lib/lizard.c b/lib/lizard.c
index ee2e245c..6167fee0 100644
--- a/lib/lizard.c
+++ b/lib/lizard.c
@@ -181,7 +181,7 @@ lizard_compress(byte *in, uns in_len, byte *out)
     uns len = find_match(hash_tab[hash], hash_rec, in, in_end, &best, head);
     if (len < 3)
 #if 0			// TODO: now, our routine does not detect matches of length 2
-      if (len == 2 && (in - best->string) < ((1<<10) + 1))
+      if (len == 2 && (in - best->string - 1) < (1<<10))
       { /* pass-thru */ }
       else
 #endif
@@ -212,26 +212,27 @@ literal:
 	goto dump_2sequence;
     } else
 #endif
-    if (len == 3 && is_in_copy_mode && shift >= (1<<11) && shift < (1<<11) + (1<<10))
+    /* now, len >= 3 */
+    if (shift < (1<<11) && len <= 8)
     {
-      /* optimisation for length-3 matches after a copy command */
-      shift -= 1<<11;
+      shift |= (len-3 + 2)<<11;
 dump_2sequence:
       if (copy_len)
 	out = flush_copy_command(bof, out, copy_start, copy_len);
       *out++ = (shift>>6) & ~3;			/* shift fits into 10 bits */
       *out++ = shift & 0xff;
     }
-    /* now, len >= 3 */
-    else if (shift < (1<<11) && len <= 8)
+    else if (len == 3 && is_in_copy_mode)
     {
-      shift |= (len-3 + 2)<<11;
-      goto dump_2sequence;			/* shift has 11 bits and contains also len */
+      if (shift < (1<<11) + (1<<10))		/* optimisation for length-3 matches after a copy command */
+      {
+	shift -= 1<<11;
+	goto dump_2sequence;			/* shift has 11 bits and contains also len */
+      }
+      else					/* avoid 3-sequence compressed to 3 sequence if it can simply be appended */
+	goto literal;
     }
     /* We have to use a 3-byte sequence.  */
-    else if (len == 3 && is_in_copy_mode)
-      /* avoid 3-sequence compressed to 3 sequence if it can simply be appended */
-      goto literal;
     else
     {
       if (copy_len)
@@ -414,18 +415,22 @@ perform_copy_command:
 Description of the LZO1X format :
 =================================
 
+The meaning of the commands depends on the current mode. It can be either
+the compressed mode or the copy mode. In some cases, the compressed mode
+also distinguishes whether we just left the copy mode or not.
+
 Beginning of file:
 ------------------
 
-If the first byte is 00010001, it means probably EOF (empty file), so switch
-to the compressed mode.  If it is bigger, subtract 17 and copy this number of
-the following characters to the ouput and switch to the compressed mode.  If
-it is smaller, go to the copy mode.
+Start in copy mode. If the first byte is 00010001, it means probably EOF (empty file),
+so switch to the compressed mode.  If it is bigger, subtract 17 and copy this number of
+the following characters to the output and switch to the compressed mode.
+If it is smaller, interpret it as a regular copy mode command.
 
-Compressed mode :
------------------
+Compressed mode:
+----------------
 
-Read the first byte of the sequence and determine the type of bit-encoding by
+Read the first byte of the sequence and determine the type of bit encoding by
 looking at the most significant bits.  The sequence is always at least 2 bytes
 long.  Decode sequences of these types until the EOF or END marker is read.
 
@@ -449,20 +454,18 @@ long.  Decode sequences of these types until the EOF or END marker is read.
 
 pattern					length		position
 
-0000ppCC 		 pppppppp	2		10 bits (*)
-0001pLLL L*	ppppppCC pppppppp	3..9 + extend	15 bits + EOF
+0000ppCC 		 pppppppp	2		10 bits		[default interpretation]
+0000ppCC 		 pppppppp	3		10 bits + 2048	[just after return from copy mode]
+0001pLLL L*	ppppppCC pppppppp	3..9 + extend	15 bits		[pos 0 interpreted as EOF]
 001LLLLL L*	ppppppCC pppppppp	3..33 + extend	14 bits
-01\
-10 \
-11  \
-LLLpppCC		 pppppppp	3..8		11 bits
+LLLpppCC		 pppppppp	3..8		11 bits		[LLL >= 010]
 
-Copy mode :
------------
+Copy mode:
+----------
 
 Read the first byte and, if the most significant bits are 0000, perform the
 following command, otherwise switch to the compressed mode (and evaluate the
-command).
+command there).
 
 pattern					length		position
 
@@ -471,9 +474,4 @@ pattern					length		position
   Copy L characters from the compressed text to the output.  The overhead for
   incompressible strings is only roughly 1/256 + epsilon.
 
-(*) After reading one copy command, switch to the compressed mode with the
-following "optimisation": the pattern 0000ppCC expands to length 3 instead of 2
-and 2048 is added to the position (now it is slightly more than 11 bits),
-because a sequence of length 2 would never be used.
-
 */