Add new HTTP/2 and HPACK decoder in net/http2/.

net/http2/hpack/huffman/http2_hpack_huffman_decoder.h

Index: net/http2/hpack/huffman/http2_hpack_huffman_decoder.h
diff --git a/net/http2/hpack/huffman/http2_hpack_huffman_decoder.h b/net/http2/hpack/huffman/http2_hpack_huffman_decoder.h
new file mode 100644
index 0000000000000000000000000000000000000000..7b8edd10c6372decd8252ea97a8dfa220c15e941
--- /dev/null
+++ b/net/http2/hpack/huffman/http2_hpack_huffman_decoder.h
@@ -0,0 +1,149 @@
+// Copyright 2016 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef NET_HTTP2_HPACK_HUFFMAN_HTTP2_HPACK_HUFFMAN_DECODER_H_
+#define NET_HTTP2_HPACK_HUFFMAN_HTTP2_HPACK_HUFFMAN_DECODER_H_
+
+// HpackHuffmanDecoder is an incremental decoder of strings that have been
+// encoded using the Huffman table defined in the HPACK spec.
+// By incremental, we mean that the HpackHuffmanDecoder::Decode method does
+// not require the entire string to be provided, and can instead decode the
+// string as fragments of it become available (e.g. as HPACK block fragments
+// are received for decoding by HpackEntryDecoder).
+
+#include <stddef.h>
+
+#include <iosfwd>
+#include <string>
+
+#include "base/strings/string_piece.h"
+#include "net/base/net_export.h"
+
+namespace net {
+
+// HuffmanAccumulator is used to store bits during decoding, e.g. next N bits
+// that have not yet been decoded, but have been extracted from the encoded
+// string).  An advantage of using a uint64 for the accumulator
+// is that it has room for the bits of the longest code plus the bits of a full
+// byte; that means that when adding more bits to the accumulator, it can always
+// be done in whole bytes. For example, if we currently have 26 bits in the
+// accumulator, and need more to decode the current symbol, we can add a whole
+// byte to the accumulator, and not have to do juggling with adding 6 bits (to
+// reach 30), and then keep track of the last two bits we've not been able to
+// add to the accumulator.
+typedef uint64_t HuffmanAccumulator;
+typedef size_t HuffmanAccumulatorBitCount;
+
+// HuffmanBitBuffer stores the leading edge of bits to be decoded. The high
+// order bit of accumulator_ is the next bit to be decoded.
+class NET_EXPORT_PRIVATE HuffmanBitBuffer {
+ public:
+  HuffmanBitBuffer();
+
+  // Prepare for decoding a new Huffman encoded string.
+  void Reset();
+
+  // Add as many whole bytes to the accumulator (accumulator_) as possible,
+  // returning the number of bytes added.
+  size_t AppendBytes(base::StringPiece input);
+
+  // Get the bits of the accumulator.
+  HuffmanAccumulator value() const { return accumulator_; }
+
+  // Number of bits of the encoded string that are in the accumulator
+  // (accumulator_).
+  HuffmanAccumulatorBitCount count() const { return count_; }
+
+  // Are there no bits in the accumulator?
+  bool IsEmpty() const { return count_ == 0; }
+
+  // Number of additional bits that can be added to the accumulator.
+  HuffmanAccumulatorBitCount free_count() const;
+
+  // Consume the leading |code_length| bits of the accumulator.
+  void ConsumeBits(HuffmanAccumulatorBitCount code_length);
+
+  // Are the contents valid for the end of a Huffman encoded string? The RFC
+  // states that EOS (end-of-string) symbol must not be explicitly encoded in
+  // the bit stream, but any unused bits in the final byte must be set to the
+  // prefix of the EOS symbol, which is all 1 bits. So there can be at most 7
+  // such bits.
+  // Returns true if the bit buffer is empty, or contains at most 7 bits, all
+  // of them 1. Otherwise returns false.
+  bool InputProperlyTerminated() const;
+
+  std::string DebugString() const;
+
+ private:
+  HuffmanAccumulator accumulator_;
+  HuffmanAccumulatorBitCount count_;
+};
+
+inline std::ostream& operator<<(std::ostream& out, const HuffmanBitBuffer& v) {
+  return out << v.DebugString();
+}
+
+class NET_EXPORT_PRIVATE HpackHuffmanDecoder {
+ public:
+  HpackHuffmanDecoder();
+  ~HpackHuffmanDecoder();
+
+  // Prepare for decoding a new Huffman encoded string.
+  void Reset() { bit_buffer_.Reset(); }
+
+  // Decode the portion of a HPACK Huffman encoded string that is in |input|,
+  // appending the decoded symbols into |*output|, stopping when more bits are
+  // needed to determine the next symbol, which/ means that the input has been
+  // drained, and also that the bit_buffer_ is empty or that the bits that are
+  // in it are not a whole symbol.
+  // If |input| is the start of a string, the caller must first call Reset.
+  // If |input| includes the end of the encoded string, the caller must call
+  // InputProperlyTerminated after Decode has returned true in order to
+  // determine if the encoded string was properly terminated.
+  // Returns false if something went wrong (e.g. the encoding contains the code
+  // EOS symbol). Otherwise returns true, in which case input has been fully
+  // decoded or buffered; in particular, if the low-order bit of the final byte
+  // of the input is not the last bit of an encoded symbol, then bit_buffer_
+  // will contain the leading bits of the code for that symbol, but not the
+  // final bits of that code.
+  // Note that output should be empty, but that it is not cleared by Decode().
+  bool Decode(base::StringPiece input, std::string* output);
+
+  // Is what remains in the bit_buffer_ valid at the end of an encoded string?
+  // Call after passing the the final portion of a Huffman string to Decode,
+  // and getting true as the result.
+  bool InputProperlyTerminated() const {
+    return bit_buffer_.InputProperlyTerminated();
+  }
+
+  bool IsEmptyForTest() const { return bit_buffer_.IsEmpty(); }
+
+  std::string DebugString() const;
+
+  //////////////////////////////////////////////////////////////////////////////
+  // Alternate implementations of Decode:
+
+  // As above, implemented using a tree of if statements to determine the code
+  // length, etc., which are returned as a tree. See PrefixToInfo. This is the
+  // original implementation, current as of 2016-8-8.
+  bool DecodeWithIfTreeAndStruct(base::StringPiece input, std::string* output);
+
+  // Based on DecodeWithIfTreeAndStruct, but adds an optimization for the common
+  // case of short codes (5, 6 or 7), which make up a large fraction of the
+  // frequency distribution on which the HPACK table was based.
+  // TODO(jamessynge): Be precise about that fraction.
+  bool DecodeShortCodesFirst(base::StringPiece input, std::string* output);
+
+ private:
+  HuffmanBitBuffer bit_buffer_;
+};
+
+inline std::ostream& operator<<(std::ostream& out,
+                                const HpackHuffmanDecoder& v) {
+  return out << v.DebugString();
+}
+
+}  // namespace net
+
+#endif  // NET_HTTP2_HPACK_HUFFMAN_HTTP2_HPACK_HUFFMAN_DECODER_H_