read_huffsym(): Use 'unsigned' for 'entry' and 'key_bits'

[wimlib] / src / lzms-decompress.c

diff --git a/src/lzms-decompress.c b/src/lzms-decompress.c
index 532dfb59c9393f3b486c4c3661e07af283e6c951..7254091449c052442bae510444f93d7f92b9a9d0 100644 (file)
--- a/src/lzms-decompress.c
+++ b/src/lzms-decompress.c
@@ -160,18 +160,28 @@
  * have equal frequency.  Following that, each code must be rebuilt whenever a
  * certain number of symbols has been decoded with it.
  *
- * In general, multiple valid Huffman codes can be constructed from a set of
- * symbol frequencies.  Like other compression formats such as XPRESS, LZX, and
- * DEFLATE, the LZMS format solves this ambiguity by requiring that all Huffman
- * codes be constructed in canonical form.  This form requires that same-length
- * codewords be lexicographically ordered the same way as the corresponding
- * symbols and that all shorter codewords lexicographically precede longer
- * codewords.
+ * Like other compression formats such as XPRESS, LZX, and DEFLATE, the LZMS
+ * format requires that all Huffman codes be constructed in canonical form.
+ * This form requires that same-length codewords be lexicographically ordered
+ * the same way as the corresponding symbols and that all shorter codewords
+ * lexicographically precede longer codewords.  Such a code can be constructed
+ * directly from codeword lengths, although in LZMS this is not actually
+ * necessary because the codes are built using adaptive symbol frequencies.
  *
- * Codewords in all the LZMS Huffman codes are limited to 15 bits.  If the
- * canonical code for a given set of symbol frequencies has any codewords longer
- * than 15 bits, then all frequencies must be divided by 2, rounding up, and the
- * code construction must be attempted again.
+ * Even with the canonical code restriction, the same frequencies can be used to
+ * construct multiple valid Huffman codes.  Therefore, the decompressor needs to
+ * construct the right one.  Specifically, the LZMS format requires that the
+ * Huffman code be constructed as if the well-known priority queue algorithm is
+ * used and frequency ties are always broken in favor of leaf nodes.  See
+ * make_canonical_huffman_code() in compress_common.c for more information.
+ *
+ * Codewords in LZMS are guaranteed to not exceed 15 bits.  The format otherwise
+ * places no restrictions on codeword length.  Therefore, the Huffman code
+ * construction algorithm that a correct LZMS decompressor uses need not
+ * implement length-limited code construction.  But if it does (e.g. by virtue
+ * of being shared among multiple compression algorithms), the details of how it
+ * does so are unimportant, provided that the maximum codeword length parameter
+ * is set to at least 15 bits.
  *
  * An LZMS-compressed block seemingly cannot have a compressed size greater than
  * or equal to the uncompressed size.  In such cases the block must be stored
@@ -190,10 +200,10 @@
 #  include "config.h"
 #endif
 
-#include "wimlib.h"
 #include "wimlib/compress_common.h"
 #include "wimlib/decompressor_ops.h"
 #include "wimlib/decompress_common.h"
+#include "wimlib/error.h"
 #include "wimlib/lzms.h"
 #include "wimlib/util.h"
 
@@ -1002,16 +1012,6 @@ lzms_decompress(const void *compressed_data, size_t compressed_size,
        if (uncompressed_size == 0)
                return 0;
 
-       /* The x86 post-processor requires that the uncompressed length fit into
-        * a signed 32-bit integer.  Also, the position slot table cannot be
-        * searched for a position of INT32_MAX or greater.  */
-       if (uncompressed_size >= INT32_MAX) {
-               LZMS_DEBUG("Uncompressed length too large "
-                          "(got %zu, expected < INT32_MAX)",
-                          uncompressed_size);
-               return -1;
-       }
-
        /* Decode the literals and matches.  */
        if (lzms_decode_items(compressed_data, compressed_size,
                              uncompressed_data, uncompressed_size, ctx))
@@ -1034,12 +1034,16 @@ lzms_free_decompressor(void *_ctx)
 }
 
 static int
-lzms_create_decompressor(size_t max_block_size,
-                        const struct wimlib_decompressor_params_header *params,
-                        void **ctx_ret)
+lzms_create_decompressor(size_t max_block_size, void **ctx_ret)
 {
        struct lzms_decompressor *ctx;
 
+       /* The x86 post-processor requires that the uncompressed length fit into
+        * a signed 32-bit integer.  Also, the position slot table cannot be
+        * searched for a position of INT32_MAX or greater.  */
+       if (max_block_size >= INT32_MAX)
+               return WIMLIB_ERR_INVALID_PARAM;
+
        ctx = ALIGNED_MALLOC(sizeof(struct lzms_decompressor),
                             DECODE_TABLE_ALIGNMENT);
        if (ctx == NULL)