Пример #1
0
        /// <summary>
        /// A helper method to query and write metadata to the stream.
        /// </summary>
        /// <param name="writer">A binary writer, which if metadata exists for the
        /// indicated column the base stream will be positioned just past the end of
        /// the written metadata table of contents, and if metadata does not exist
        /// remains unchanged</param>
        /// <param name="schema">The schema to query for metadat</param>
        /// <param name="col">The column we are attempting to get metadata for</param>
        /// <param name="ch">The channel to which we write any diagnostic information</param>
        /// <returns>The offset of the metadata table of contents, or 0 if there was
        /// no metadata</returns>
        private long WriteMetadata(BinaryWriter writer, Schema schema, int col, IChannel ch)
        {
            _host.AssertValue(writer);
            _host.AssertValue(schema);
            _host.Assert(0 <= col && col < schema.Count);

            int count = 0;
            WriteMetadataCoreDelegate del = WriteMetadataCore <int>;
            MethodInfo methInfo           = del.GetMethodInfo().GetGenericMethodDefinition();

            object[] args = new object[] { writer.BaseStream, schema, col, null, null, null };

            List <long> offsets = new List <long>();

            offsets.Add(writer.BaseStream.Position);
            var metadataInfos = new List <Tuple <string, IValueCodec, CompressionKind> >();
            var kinds         = new HashSet <string>();

            // Write all metadata blocks for this column to the file, one after the other, keeping
            // track of the location and size of each for when we write the metadata table of contents.
            // (To be clear, this specific layout is not required by the format.)

            foreach (var metaColumn in schema[col].Metadata.Schema)
            {
                _host.Check(!string.IsNullOrEmpty(metaColumn.Name), "Metadata with null or empty kind detected, disallowed");
                _host.Check(metaColumn.Type != null, "Metadata with null type detected, disallowed");
                if (!kinds.Add(metaColumn.Name))
                {
                    throw _host.Except("Metadata with duplicate kind '{0}' encountered, disallowed", metaColumn.Name, schema[col].Name);
                }
                args[3] = metaColumn.Name;
                args[4] = metaColumn.Type;
                IValueCodec codec = (IValueCodec)methInfo.MakeGenericMethod(metaColumn.Type.RawType).Invoke(this, args);
                if (codec == null)
                {
                    // Nothing was written.
                    ch.Warning("Could not get codec for type {0}, dropping column '{1}' index {2} metadata kind '{3}'",
                               metaColumn.Type, schema[col].Name, col, metaColumn.Name);
                    continue;
                }
                offsets.Add(writer.BaseStream.Position);
                _host.CheckIO(offsets[offsets.Count - 1] > offsets[offsets.Count - 2], "Bad offsets detected during write");
                metadataInfos.Add(Tuple.Create(metaColumn.Name, codec, (CompressionKind)args[5]));
                count++;
            }
            if (metadataInfos.Count == 0)
            {
                _host.CheckIO(writer.BaseStream.Position == offsets[0], "unexpected offset after no writing of metadata");
                return(0);
            }
            // Write the metadata table of contents just past the end of the last metadata block.

            // *** Metadata TOC format ***
            // LEB128 int: Number of metadata TOC entries
            // Metadata TOC entries: As many of these as indicated by the count above

            long expectedPosition = offsets[metadataInfos.Count];

            writer.WriteLeb128Int((ulong)metadataInfos.Count);
            expectedPosition += Utils.Leb128IntLength((ulong)metadataInfos.Count);
            for (int i = 0; i < metadataInfos.Count; ++i)
            {
                // *** Metadata TOC entry format ***
                // string: metadata kind
                // codec definition: metadata codec
                // CompressionKind(byte): block compression strategy
                // long: Offset into the stream of the start of the metadata block
                // LEB128 int: Byte size of the metadata block in the file

                writer.Write(metadataInfos[i].Item1);
                int stringLen = Encoding.UTF8.GetByteCount(metadataInfos[i].Item1);
                expectedPosition += Utils.Leb128IntLength((ulong)stringLen) + stringLen;
                _host.CheckIO(writer.BaseStream.Position == expectedPosition, "unexpected offsets after metadata table of contents kind");

                expectedPosition += _factory.WriteCodec(writer.BaseStream, metadataInfos[i].Item2);
                _host.CheckIO(writer.BaseStream.Position == expectedPosition, "unexpected offsets after metadata table of contents type description");

                writer.Write((byte)metadataInfos[i].Item3);
                expectedPosition++;

                writer.Write(offsets[i]);
                expectedPosition += sizeof(long);

                long blockSize = offsets[i + 1] - offsets[i];
                writer.WriteLeb128Int((ulong)blockSize);
                expectedPosition += Utils.Leb128IntLength((ulong)blockSize);
                _host.CheckIO(writer.BaseStream.Position == expectedPosition, "unexpected offsets after metadata table of contents location");
            }
            _host.Assert(metadataInfos.Count == offsets.Count - 1);
            return(offsets[metadataInfos.Count]);
        }