Update (#749)

Update:

 * Bazel: fix MSVC configuration
 * C: common: extended documentation and helpers around distance codes
 * C: common: enable BROTLI_DCHECK in "debug" builds
 * C: common: fix implicit trailing zero in `kPrefixSuffix`
 * C: dec: fix possible bit reader discharge for "large-window" mode
 * C: dec: simplify distance decoding via lookup table
 * C: dec: reuse decoder state members memory via union with lookup table
 * C: dec: add decoder state diagram
 * C: enc: clarify access to static dictionary
 * C: enc: improve static dictionary hash
 * C: enc: add "stream offset" parameter for parallel encoding
 * C: enc: reorganize hasher; now Q2-Q3 require exactly 256KiB
           to avoid global TCMalloc lock
 * C: enc: fix rare access to uninitialized data in ring-buffer
 * C: enc: reorganize logging / checks in `write_bits.h`
 * Java: dec: add "large-window" support
 * Java: dec: improve speed
 * Java: dec: debug and 32-bit mode are now activated via system properties
 * Java: dec: demystify some state variables (use better names)
 * Dictionary generator: add single input mode
 * Java: dec: modernize tests
 * Bazel: js: pick working commit for closure rules

3 files changed, 124 insertions, 4 deletions

diff --git a/c/common/constants.h b/c/common/constants.h
index d1b88d1..f6e44dc 100644
--- a/c/common/constants.h
+++ b/c/common/constants.h
@@ -4,9 +4,17 @@
    See file LICENSE for detail or copy at https://opensource.org/licenses/MIT
 */
 
+/**
+ * @file
+ * Common constants used in decoder and encoder API.
+ */
+
 #ifndef BROTLI_COMMON_CONSTANTS_H_
 #define BROTLI_COMMON_CONSTANTS_H_
 
+#include "./platform.h"
+#include <brotli/types.h>
+
 /* Specification: 7.3. Encoding of the context map */
 #define BROTLI_CONTEXT_MAP_MAX_RLE 16
 
@@ -29,12 +37,31 @@
 #define BROTLI_INITIAL_REPEATED_CODE_LENGTH 8
 
 /* "Large Window Brotli" */
+
+/**
+ * The theoretical maximum number of distance bits specified for large window
+ * brotli, for 64-bit encoders and decoders. Even when in practice 32-bit
+ * encoders and decoders only support up to 30 max distance bits, the value is
+ * set to 62 because it affects the large window brotli file format.
+ * Specifically, it affects the encoding of simple huffman tree for distances,
+ * see Specification RFC 7932 chapter 3.4.
+ */
 #define BROTLI_LARGE_MAX_DISTANCE_BITS 62U
 #define BROTLI_LARGE_MIN_WBITS 10
+/**
+ * The maximum supported large brotli window bits by the encoder and decoder.
+ * Large window brotli allows up to 62 bits, however the current encoder and
+ * decoder, designed for 32-bit integers, only support up to 30 bits maximum.
+ */
 #define BROTLI_LARGE_MAX_WBITS 30
 
 /* Specification: 4. Encoding of distances */
 #define BROTLI_NUM_DISTANCE_SHORT_CODES 16
+/**
+ * Maximal number of "postfix" bits.
+ *
+ * Number of "postfix" bits is stored as 2 bits in meta-block header.
+ */
 #define BROTLI_MAX_NPOSTFIX 3
 #define BROTLI_MAX_NDIRECT 120
 #define BROTLI_MAX_DISTANCE_BITS 24U
@@ -45,7 +72,16 @@
 #define BROTLI_NUM_DISTANCE_SYMBOLS \
     BROTLI_DISTANCE_ALPHABET_SIZE(  \
         BROTLI_MAX_NDIRECT, BROTLI_MAX_NPOSTFIX, BROTLI_LARGE_MAX_DISTANCE_BITS)
+
+/* ((1 << 26) - 4) is the maximal distance that can be expressed in RFC 7932
+   brotli stream using NPOSTFIX = 0 and NDIRECT = 0. With other NPOSTFIX and
+   NDIRECT values distances up to ((1 << 29) + 88) could be expressed. */
 #define BROTLI_MAX_DISTANCE 0x3FFFFFC
+
+/* ((1 << 31) - 4) is the safe distance limit. Using this number as a limit
+   allows safe distance calculation without overflows, given the distance
+   alphabet size is limited to corresponding size
+   (see kLargeWindowDistanceCodeLimits). */
 #define BROTLI_MAX_ALLOWED_DISTANCE 0x7FFFFFFC
 
 /* 7.1. Context modes and context ID lookup for literals */
@@ -61,4 +97,88 @@
 #define BROTLI_WINDOW_GAP 16
 #define BROTLI_MAX_BACKWARD_LIMIT(W) (((size_t)1 << (W)) - BROTLI_WINDOW_GAP)
 
+typedef struct BrotliDistanceCodeLimit {
+  uint32_t max_alphabet_size;
+  uint32_t max_distance;
+} BrotliDistanceCodeLimit;
+
+/* This function calculates maximal size of distance alphabet, such that the
+   distances greater than the given values can not be represented.
+
+   This limits are designed to support fast and safe 32-bit decoders.
+   "32-bit" means that signed integer values up to ((1 << 31) - 1) could be
+   safely expressed.
+
+   Brotli distance alphabet symbols do not represent consecutive distance
+   ranges. Each distance alphabet symbol (excluding direct distances and short
+   codes), represent interleaved (for NPOSTFIX > 0) range of distances.
+   A "group" of consecutive (1 << NPOSTFIX) symbols represent non-interleaved
+   range. Two consecutive groups require the same amount of "extra bits".
+
+   It is important that distance alphabet represents complete "groups".
+   To avoid complex logic on encoder side about interleaved ranges
+   it was decided to restrict both sides to complete distance code "groups".
+ */
+BROTLI_UNUSED_FUNCTION BrotliDistanceCodeLimit BrotliCalculateDistanceCodeLimit(
+    uint32_t max_distance, uint32_t npostfix, uint32_t ndirect) {
+  BrotliDistanceCodeLimit result;
+  /* Marking this function as unused, because not all files
+     including "constants.h" use it -> compiler warns about that. */
+  BROTLI_UNUSED(&BrotliCalculateDistanceCodeLimit);
+  if (max_distance <= ndirect) {
+    /* This case never happens / exists only for the sake of completeness. */
+    result.max_alphabet_size = max_distance + BROTLI_NUM_DISTANCE_SHORT_CODES;
+    result.max_distance = max_distance;
+    return result;
+  } else {
+    /* The first prohibited value. */
+    uint32_t forbidden_distance = max_distance + 1;
+    /* Subtract "directly" encoded region. */
+    uint32_t offset = forbidden_distance - ndirect - 1;
+    uint32_t ndistbits = 0;
+    uint32_t tmp;
+    uint32_t half;
+    uint32_t group;
+    /* Postfix for the last dcode in the group. */
+    uint32_t postfix = (1u << npostfix) - 1;
+    uint32_t extra;
+    uint32_t start;
+    /* Remove postfix and "head-start". */
+    offset = (offset >> npostfix) + 4;
+    /* Calculate the number of distance bits. */
+    tmp = offset / 2;
+    /* Poor-man's log2floor, to avoid extra dependencies. */
+    while (tmp != 0) {ndistbits++; tmp = tmp >> 1;}
+    /* One bit is covered with subrange addressing ("half"). */
+    ndistbits--;
+    /* Find subrange. */
+    half = (offset >> ndistbits) & 1;
+    /* Calculate the "group" part of dcode. */
+    group = ((ndistbits - 1) << 1) | half;
+    /* Calculated "group" covers the prohibited distance value. */
+    if (group == 0) {
+      /* This case is added for correctness; does not occur for limit > 128. */
+      result.max_alphabet_size = ndirect + BROTLI_NUM_DISTANCE_SHORT_CODES;
+      result.max_distance = ndirect;
+      return result;
+    }
+    /* Decrement "group", so it is the last permitted "group". */
+    group--;
+    /* After group was decremented, ndistbits and half must be recalculated. */
+    ndistbits = (group >> 1) + 1;
+    /* The last available distance in the subrange has all extra bits set. */
+    extra = (1u << ndistbits) - 1;
+    /* Calculate region start. NB: ndistbits >= 1. */
+    start = (1u << (ndistbits + 1)) - 4;
+    /* Move to subregion. */
+    start += (group & 1) << ndistbits;
+    /* Calculate the alphabet size. */
+    result.max_alphabet_size = ((group << npostfix) | postfix) + ndirect +
+        BROTLI_NUM_DISTANCE_SHORT_CODES + 1;
+    /* Calculate the maximal distance representable by alphabet. */
+    result.max_distance = ((start + extra) << npostfix) + postfix + ndirect + 1;
+    return result;
+  }
+}
+
 #endif  /* BROTLI_COMMON_CONSTANTS_H_ */
diff --git a/c/common/platform.h b/c/common/platform.h
index 84c448c..bf5f97b 100755
--- a/c/common/platform.h
+++ b/c/common/platform.h
@@ -466,20 +466,20 @@ static BROTLI_INLINE void BROTLI_UNALIGNED_STORE64LE(void* p, uint64_t v) {
 #endif
 
 #if defined(BROTLI_ENABLE_LOG)
-#define BROTLI_DCHECK(x) assert(x)
 #define BROTLI_LOG(x) printf x
 #else
-#define BROTLI_DCHECK(x)
 #define BROTLI_LOG(x)
 #endif
 
 #if defined(BROTLI_DEBUG) || defined(BROTLI_ENABLE_LOG)
+#define BROTLI_DCHECK(x) assert(x)
 static BROTLI_INLINE void BrotliDump(const char* f, int l, const char* fn) {
   fprintf(stderr, "%s:%d (%s)\n", f, l, fn);
   fflush(stderr);
 }
 #define BROTLI_DUMP() BrotliDump(__FILE__, __LINE__, __FUNCTION__)
 #else
+#define BROTLI_DCHECK(x)
 #define BROTLI_DUMP() (void)(0)
 #endif
 
diff --git a/c/common/transform.c b/c/common/transform.c
index c182053..c44f671 100755
--- a/c/common/transform.c
+++ b/c/common/transform.c
@@ -24,8 +24,8 @@ static const char kPrefixSuffix[217] =
 /* 8x  _0 _ _3    _8   _C _E _ _1     _7       _F */
       " not \3er \3al \4ful \4ive \5less \4es"
 /* Ax       _5   _9   _D    _2    _7     _D */
-      "t \4ize \2\xc2\xa0\4ous \5 the \2e \0";
-/* Cx    _2    _7___ ___ _A    _F     _5  _8 */
+      "t \4ize \2\xc2\xa0\4ous \5 the \2e "; /* \0 - implicit trailing zero. */
+/* Cx    _2    _7___ ___ _A    _F     _5        _8 */
 
 static const uint16_t kPrefixSuffixMap[50] = {
   0x00, 0x02, 0x05, 0x0E, 0x13, 0x16, 0x18, 0x1E, 0x23, 0x25,