Refactor serializer

Improvements:

* User-provided buffers are now type-erased and used directly, rather
    than being copied into a flat buffer. This enables efficient
    handling of buffer sequences with many elements.

* Chunk encoding now occurs within the main buffers, allowing
    serialization of multiple chunks. This also improves buffer
    consumption efficiency by reducing the total number of buffers.

* The number of internal states maintained by the serializer has been
    reduced. This lowers memory overhead and simplifies reasoning about
    the implementation.

Author: ashtum

doc/local-playbook.yml

@@ -14,7 +14,7 @@ content:
 
 ui:
   bundle:
-    url: https://github.com/boostorg/website-v2-docs/releases/download/ui-master/ui-bundle.zip
+    url: https://github.com/boostorg/website-v2-docs/releases/download/ui-develop/ui-bundle.zip
     snapshot: true
 
 antora:

include/boost/http_proto/detail/impl/array_of_buffers.hpp

@@ -124,8 +124,6 @@ reset(std::size_t n)
     BOOST_ASSERT(n <= capacity());
     p_ = o_;
     n_ = n;
-    for( auto p = p_; p < p_ + n; ++p )
-        *p = value_type();
 }
 
 } // detail

include/boost/http_proto/serializer.hpp

@@ -17,24 +17,22 @@
 #include <boost/http_proto/detail/header.hpp>
 #include <boost/http_proto/detail/workspace.hpp>
 #include <boost/http_proto/source.hpp>
+
 #include <boost/buffers/circular_buffer.hpp>
 #include <boost/buffers/const_buffer_span.hpp>
 #include <boost/buffers/range.hpp>
 #include <boost/buffers/type_traits.hpp>
 #include <boost/system/result.hpp>
-#include <cstdint>
+
 #include <memory>
+#include <numeric>
 #include <type_traits>
 #include <utility>
 
 namespace boost {
 namespace http_proto {
 
 #ifndef BOOST_HTTP_PROTO_DOCS
-class request;
-class response;
-class request_view;
-class response_view;
 class message_view_base;
 namespace detail {
 class filter;
@@ -219,36 +217,14 @@ class serializer
     void
     consume(std::size_t n);
 
-    /** Applies deflate compression to the current message
-
-        After @ref reset is called, compression is not
-        applied to the next message.
-
-        Must be called before any calls to @ref start.
-    */
-    BOOST_HTTP_PROTO_DECL
-    void
-    use_deflate_encoding();
-
-    /** Applies gzip compression to the current message
+private:
+    class const_buf_gen_base;
 
-        After @ref reset is called, compression is not
-        applied to the next message.
+    template<class>
+    class const_buf_gen;
 
-        Must be called before any calls to @ref start.
-    */
-    BOOST_HTTP_PROTO_DECL
-    void
-    use_gzip_encoding();
-
-private:
-    static void copy(
-        buffers::const_buffer*,
-        buffers::const_buffer const*,
-        std::size_t n) noexcept;
-    auto
-    make_array(std::size_t n) ->
-        detail::array_of_const_buffers;
+    detail::array_of_const_buffers
+    make_array(std::size_t n);
 
     template<
         class Source,
@@ -279,10 +255,24 @@ class serializer
             ws_, std::forward<Args>(args)...);
     }
 
-    BOOST_HTTP_PROTO_DECL void start_init(message_view_base const&);
-    BOOST_HTTP_PROTO_DECL void start_empty(message_view_base const&);
-    BOOST_HTTP_PROTO_DECL void start_buffers(message_view_base const&);
-    BOOST_HTTP_PROTO_DECL void start_source(message_view_base const&, source*);
+    BOOST_HTTP_PROTO_DECL
+    void
+    start_init(
+        message_view_base const&);
+
+    BOOST_HTTP_PROTO_DECL
+    void
+    start_empty(
+        message_view_base const&);
+
+    BOOST_HTTP_PROTO_DECL
+    void start_buffers(
+        message_view_base const&);
+
+    BOOST_HTTP_PROTO_DECL
+    void
+    start_source(
+        message_view_base const&, source*);
 
     enum class style
     {
@@ -292,68 +282,25 @@ class serializer
         stream
     };
 
-    // chunked-body   = *chunk
-    //                  last-chunk
-    //                  trailer-section
-    //                  CRLF
-
-    static
-    constexpr
-    std::size_t
-    crlf_len_ = 2;
-
-    // chunk          = chunk-size [ chunk-ext ] CRLF
-    //                  chunk-data CRLF
-    static
-    constexpr
-    std::size_t
-    chunk_header_len_ =
-        16 + // 16 hex digits => 64 bit number
-        crlf_len_;
-
-    // last-chunk     = 1*("0") [ chunk-ext ] CRLF
-    static
-    constexpr
-    std::size_t
-    last_chunk_len_ =
-        1 + // "0"
-        crlf_len_ +
-        crlf_len_; // chunked-body termination requires an extra CRLF
-
-    static
-    constexpr
-    std::size_t
-    chunked_overhead_ =
-        chunk_header_len_ +
-        crlf_len_ + // closing chunk data
-        last_chunk_len_;
-
-    detail::workspace ws_;
-    detail::array_of_const_buffers buf_;
-    detail::filter* filter_ = nullptr;
-    source* src_;
     context& ctx_;
-    buffers::circular_buffer tmp0_;
-    buffers::circular_buffer tmp1_;
-    detail::array_of_const_buffers prepped_;
-
-    buffers::mutable_buffer chunk_header_;
-    buffers::mutable_buffer chunk_close_;
-    buffers::mutable_buffer last_chunk_;
+    detail::workspace ws_;
 
-    buffers::circular_buffer* in_ = nullptr;
-    buffers::circular_buffer* out_ = nullptr;
+    const_buf_gen_base* buf_gen_;
+    detail::filter* filter_;
+    source* source_;
 
-    buffers::const_buffer* hp_;  // header
+    buffers::circular_buffer cb0_;
+    buffers::circular_buffer cb1_;
+    detail::array_of_const_buffers prepped_;
+    buffers::const_buffer tmp_;
 
     style st_;
-    bool more_;
+    bool more_input_;
     bool is_done_;
     bool is_header_done_;
     bool is_chunked_;
-    bool is_expect_continue_;
-    bool is_compressed_ = false;
-    bool filter_done_ = false;
+    bool needs_exp100_continue_;
+    bool filter_done_;
 };
 
 //------------------------------------------------
@@ -422,17 +369,6 @@ struct serializer::stream
     std::size_t
     capacity() const noexcept;
 
-    /** Returns the number of octets serialized by this
-        stream.
-
-        The associated serializer stores stream output in its
-        internal buffers. The stream returns the size of this
-        output.
-    */
-    BOOST_HTTP_PROTO_DECL
-    std::size_t
-    size() const noexcept;
-
     /** Return true if the stream cannot currently hold
         additional output data.
 
@@ -503,28 +439,116 @@ struct serializer::stream
 
 //---------------------------------------------------------
 
+class serializer::const_buf_gen_base
+{
+public:
+    // Next non-empty buffer
+    virtual
+    buffers::const_buffer
+    next() = 0;
+
+    // Total size of remaining buffers
+    virtual
+    std::size_t
+    remaining_size() const = 0;
+
+    // Count of non-empty remaining buffers
+    virtual
+    std::size_t
+    remaining_count() const = 0;
+
+    // Returns true when there is no buffer or
+    // the remaining buffers are empty
+    virtual
+    bool
+    is_empty() const = 0;
+};
+
+template<class ConstBufferSequence>
+class serializer::const_buf_gen
+    : public const_buf_gen_base
+{
+    using it_t = decltype(buffers::begin(
+        std::declval<ConstBufferSequence>()));
+
+    using const_buffer =
+        buffers::const_buffer;
+
+    ConstBufferSequence cbs_;
+    it_t current_;
+public:
+    explicit
+    const_buf_gen(ConstBufferSequence cbs)
+        : cbs_(std::move(cbs))
+        , current_(buffers::begin(cbs_))
+    {
+    }
+
+    buffers::const_buffer
+    next() override
+    {
+        while(current_ != buffers::end(cbs_))
+        {
+            auto buf = static_cast<
+                const_buffer>(*current_++);
+            if(buf.size() != 0)
+                return buf;
+        }
+        return {};
+    }
+
+    std::size_t
+    remaining_size() const override
+    {
+        return std::accumulate(
+            current_,
+            buffers::end(cbs_),
+            std::size_t{},
+            [](std::size_t sum, const_buffer cb) {
+                return sum + cb.size(); });
+    }
+
+    std::size_t
+    remaining_count() const override
+    {
+        return std::count_if(
+            current_,
+            buffers::end(cbs_),
+            [](const_buffer cb) {
+                return cb.size() != 0; });
+    }
+
+    bool
+    is_empty() const override
+    {
+        return std::all_of(
+            current_,
+            buffers::end(cbs_),
+            [](const_buffer cb) {
+                return cb.size() == 0; });
+    }
+};
+
+//---------------------------------------------------------
+
 template<
     class ConstBufferSequence,
     class>
 void
 serializer::
 start(
     message_view_base const& m,
-    ConstBufferSequence&& body)
+    ConstBufferSequence&& cbs)
 {
-    start_init(m);
-    auto const& bs =
-        ws_.emplace<ConstBufferSequence>(
-            std::forward<ConstBufferSequence>(body));
-
-    std::size_t n = std::distance(
-        buffers::begin(bs),
-        buffers::end(bs));
+    static_assert(buffers::is_const_buffer_sequence<
+            ConstBufferSequence>::value,
+        "ConstBufferSequence type requirements not met");
 
-    buf_ = make_array(n);
-    auto p = buf_.data();
-    for(buffers::const_buffer b : buffers::range(bs))
-        *p++ = b;
+    start_init(m);
+    buf_gen_ = &ws_.emplace<
+        const_buf_gen<typename std::decay<
+            ConstBufferSequence>::type>>(
+            std::forward<ConstBufferSequence>(cbs));
 
     start_buffers(m);
 }
@@ -553,20 +577,6 @@ start(
     return src;
 }
 
-//------------------------------------------------
-
-inline
-auto
-serializer::
-make_array(std::size_t n) ->
-    detail::array_of_const_buffers
-{
-    return {
-        ws_.push_array(n,
-        buffers::const_buffer{}),
-        n };
-}
-
 } // http_proto
 } // boost
 

src/detail/impl/filter.hpp

@@ -41,7 +41,7 @@ process_impl(
     auto ib = *it_i++;
     for(;;)
     {
-        // empty buffers may be passed, and this is
+        // empty input buffer may be passed, and this is
         // intentional and valid.
         results rs = process_impl(ob, ib, more);
 
@@ -56,7 +56,7 @@ process_impl(
         ob = buffers::sans_prefix(ob, rs.out_bytes);
         ib = buffers::sans_prefix(ib, rs.in_bytes);
 
-        if( ob.size() == 0 )
+        while( ob.size() == 0 )
         {
             if( it_o == buffers::end(out) )
                 return rv;