simdjson 0.2.0 → 0.3.0

data/vendor/simdjson/dependencies/rapidjson/doc/schema.zh-cn.md

@@ -0,0 +1,237 @@
+# Schema
+
+（本功能于 v1.1.0 发布）
+
+JSON Schema 是描述 JSON 格式的一个标准草案。一个 schema 本身也是一个 JSON。使用 JSON Schema 去校验 JSON，可以让你的代码安全地访问 DOM，而无须检查类型或键值是否存在等。这也能确保输出的 JSON 是符合指定的 schema。
+
+RapidJSON 实现了一个 [JSON Schema Draft v4](http://json-schema.org/documentation.html) 的校验器。若你不熟悉 JSON Schema，可以参考 [Understanding JSON Schema](http://spacetelescope.github.io/understanding-json-schema/)。
+
+[TOC]
+
+# 基本用法 {#BasicUsage}
+
+首先，你要把 JSON Schema 解析成 `Document`，再把它编译成一个 `SchemaDocument`。
+
+然后，利用该 `SchemaDocument` 创建一个 `SchemaValidator`。它与 `Writer` 相似，都是能够处理 SAX 事件的。因此，你可以用 `document.Accept(validator)` 去校验一个 JSON，然后再获取校验结果。
+
+~~~cpp
+#include "rapidjson/schema.h"
+
+// ...
+
+Document sd;
+if (sd.Parse(schemaJson).HasParseError()) {
+    // 此 schema 不是合法的 JSON
+    // ...       
+}
+SchemaDocument schema(sd); // 把一个 Document 编译至 SchemaDocument
+// 之后不再需要 sd
+
+Document d;
+if (d.Parse(inputJson).HasParseError()) {
+    // 输入不是一个合法的 JSON
+    // ...       
+}
+
+SchemaValidator validator(schema);
+if (!d.Accept(validator)) {
+    // 输入的 JSON 不合乎 schema
+    // 打印诊断信息
+    StringBuffer sb;
+    validator.GetInvalidSchemaPointer().StringifyUriFragment(sb);
+    printf("Invalid schema: %s\n", sb.GetString());
+    printf("Invalid keyword: %s\n", validator.GetInvalidSchemaKeyword());
+    sb.Clear();
+    validator.GetInvalidDocumentPointer().StringifyUriFragment(sb);
+    printf("Invalid document: %s\n", sb.GetString());
+}
+~~~
+
+一些注意点：
+
+* 一个 `SchemaDocment` 能被多个 `SchemaValidator` 引用。它不会被 `SchemaValidator` 修改。
+* 可以重复使用一个 `SchemaValidator` 来校验多个文件。在校验其他文件前，须先调用 `validator.Reset()`。
+
+# 在解析／生成时进行校验 {#ParsingSerialization}
+
+与大部分 JSON Schema 校验器有所不同，RapidJSON 提供了一个基于 SAX 的 schema 校验器实现。因此，你可以在输入流解析 JSON 的同时进行校验。若校验器遇到一个与 schema 不符的值，就会立即终止解析。这设计对于解析大型 JSON 文件时特别有用。
+
+## DOM 解析 {#DomParsing}
+
+在使用 DOM 进行解析时，`Document` 除了接收 SAX 事件外，还需做一些准备及结束工作，因此，为了连接 `Reader`、`SchemaValidator` 和 `Document` 要做多一点事情。`SchemaValidatingReader` 是一个辅助类去做那些工作。
+
+~~~cpp
+#include "rapidjson/filereadstream.h"
+
+// ...
+SchemaDocument schema(sd); // 把一个 Document 编译至 SchemaDocument
+
+// 使用 reader 解析 JSON
+FILE* fp = fopen("big.json", "r");
+FileReadStream is(fp, buffer, sizeof(buffer));
+
+// 用 reader 解析 JSON，校验它的 SAX 事件，并存储至 d
+Document d;
+SchemaValidatingReader<kParseDefaultFlags, FileReadStream, UTF8<> > reader(is, schema);
+d.Populate(reader);
+
+if (!reader.GetParseResult()) {
+    // 不是一个合法的 JSON
+    // 当 reader.GetParseResult().Code() == kParseErrorTermination,
+    // 它可能是被以下原因中止：
+    // (1) 校验器发现 JSON 不合乎 schema；或
+    // (2) 输入流有 I/O 错误。
+
+    // 检查校验结果
+    if (!reader.IsValid()) {
+        // 输入的 JSON 不合乎 schema
+        // 打印诊断信息
+        StringBuffer sb;
+        reader.GetInvalidSchemaPointer().StringifyUriFragment(sb);
+        printf("Invalid schema: %s\n", sb.GetString());
+        printf("Invalid keyword: %s\n", reader.GetInvalidSchemaKeyword());
+        sb.Clear();
+        reader.GetInvalidDocumentPointer().StringifyUriFragment(sb);
+        printf("Invalid document: %s\n", sb.GetString());
+    }
+}
+~~~
+
+## SAX 解析 {#SaxParsing}
+
+使用 SAX 解析时，情况就简单得多。若只需要校验 JSON 而无需进一步处理，那么仅需要：
+
+~~~
+SchemaValidator validator(schema);
+Reader reader;
+if (!reader.Parse(stream, validator)) {
+    if (!validator.IsValid()) {
+        // ...    
+    }
+}
+~~~
+
+这种方式和 [schemavalidator](example/schemavalidator/schemavalidator.cpp) 例子完全相同。这带来的独特优势是，无论 JSON 多巨大，永远维持低内存用量（内存用量只与 Schema 的复杂度相关）。
+
+若你需要进一步处理 SAX 事件，便可使用模板类 `GenericSchemaValidator` 去设置校验器的输出 `Handler`：
+
+~~~
+MyHandler handler;
+GenericSchemaValidator<SchemaDocument, MyHandler> validator(schema, handler);
+Reader reader;
+if (!reader.Parse(ss, validator)) {
+    if (!validator.IsValid()) {
+        // ...    
+    }
+}
+~~~
+
+## 生成 {#Serialization}
+
+我们也可以在生成（serialization）的时候进行校验。这能确保输出的 JSON 符合一个 JSON Schema。
+
+~~~
+StringBuffer sb;
+Writer<StringBuffer> writer(sb);
+GenericSchemaValidator<SchemaDocument, Writer<StringBuffer> > validator(s, writer);
+if (!d.Accept(validator)) {
+    // Some problem during Accept(), it may be validation or encoding issues.
+    if (!validator.IsValid()) {
+        // ...
+    }
+}
+~~~
+
+当然，如果你的应用仅需要 SAX 风格的生成，那么只需要把 SAX 事件由原来发送到 `Writer`，改为发送到 `SchemaValidator`。
+
+# 远程 Schema {#RemoteSchema}
+
+JSON Schema 支持 [`$ref` 关键字](http://spacetelescope.github.io/understanding-json-schema/structuring.html)，它是一个 [JSON pointer](doc/pointer.zh-cn.md) 引用至一个本地（local）或远程（remote） schema。本地指针的首字符是 `#`，而远程指针是一个相对或绝对 URI。例如：
+
+~~~js
+{ "$ref": "definitions.json#/address" }
+~~~
+
+由于 `SchemaDocument` 并不知道如何处理那些 URI，它需要使用者提供一个 `IRemoteSchemaDocumentProvider` 的实例去处理。
+
+~~~
+class MyRemoteSchemaDocumentProvider : public IRemoteSchemaDocumentProvider {
+public:
+    virtual const SchemaDocument* GetRemoteDocument(const char* uri, SizeType length) {
+        // Resolve the uri and returns a pointer to that schema.
+    }
+};
+
+// ...
+
+MyRemoteSchemaDocumentProvider provider;
+SchemaDocument schema(sd, &provider);
+~~~
+
+# 标准的符合程度 {#Conformance}
+
+RapidJSON 通过了 [JSON Schema Test Suite](https://github.com/json-schema/JSON-Schema-Test-Suite) (Json Schema draft 4) 中 263 个测试的 262 个。
+
+没通过的测试是 `refRemote.json` 中的 "change resolution scope" - "changed scope ref invalid"。这是由于未实现 `id` schema 关键字及 URI 合并功能。
+
+除此以外，关于字符串类型的 `format` schema 关键字也会被忽略，因为标准中并没需求必须实现。
+
+## 正则表达式 {#RegEx}
+
+`pattern` 及 `patternProperties` 这两个 schema 关键字使用了正则表达式去匹配所需的模式。
+
+RapidJSON 实现了一个简单的 NFA 正则表达式引擎，并预设使用。它支持以下语法。
+
+|语法|描述|
+|------|-----------|
+|`ab`    | 串联 |
+|<code>a&#124;b</code>   | 交替 |
+|`a?`    | 零或一次 |
+|`a*`    | 零或多次 |
+|`a+`    | 一或多次 |
+|`a{3}`  | 刚好 3 次 |
+|`a{3,}` | 至少 3 次 |
+|`a{3,5}`| 3 至 5 次 |
+|`(ab)`  | 分组 |
+|`^a`    | 在开始处 |
+|`a$`    | 在结束处 |
+|`.`     | 任何字符 |
+|`[abc]` | 字符组 |
+|`[a-c]` | 字符组范围 |
+|`[a-z0-9_]` | 字符组组合 |
+|`[^abc]` | 字符组取反 |
+|`[^a-c]` | 字符组范围取反 |
+|`[\b]`   | 退格符 (U+0008) |
+|<code>\\&#124;</code>, `\\`, ...  | 转义字符 |
+|`\f` | 馈页 (U+000C) |
+|`\n` | 馈行 (U+000A) |
+|`\r` | 回车 (U+000D) |
+|`\t` | 制表 (U+0009) |
+|`\v` | 垂直制表 (U+000B) |
+
+对于使用 C++11 编译器的使用者，也可使用 `std::regex`，只需定义 `RAPIDJSON_SCHEMA_USE_INTERNALREGEX=0` 及 `RAPIDJSON_SCHEMA_USE_STDREGEX=1`。若你的 schema 无需使用 `pattern` 或 `patternProperties`，可以把两个宏都设为零，以禁用此功能，这样做可节省一些代码体积。
+
+# 性能 {#Performance}
+
+大部分 C++ JSON 库都未支持 JSON Schema。因此我们尝试按照 [json-schema-benchmark](https://github.com/ebdrup/json-schema-benchmark) 去评估 RapidJSON 的 JSON Schema 校验器。该评测测试了 11 个运行在 node.js 上的 JavaScript 库。
+
+该评测校验 [JSON Schema Test Suite](https://github.com/json-schema/JSON-Schema-Test-Suite) 中的测试，当中排除了一些测试套件及个别测试。我们在 [`schematest.cpp`](test/perftest/schematest.cpp) 实现了相同的评测。
+
+在 MacBook Pro (2.8 GHz Intel Core i7) 上收集到以下结果。
+
+|校验器|相对速度|每秒执行的测试数目|
+|---------|:------------:|:----------------------------:|
+|RapidJSON|155%|30682|
+|[`ajv`](https://github.com/epoberezkin/ajv)|100%|19770 (± 1.31%)|
+|[`is-my-json-valid`](https://github.com/mafintosh/is-my-json-valid)|70%|13835 (± 2.84%)|
+|[`jsen`](https://github.com/bugventure/jsen)|57.7%|11411 (± 1.27%)|
+|[`schemasaurus`](https://github.com/AlexeyGrishin/schemasaurus)|26%|5145 (± 1.62%)|
+|[`themis`](https://github.com/playlyfe/themis)|19.9%|3935 (± 2.69%)|
+|[`z-schema`](https://github.com/zaggino/z-schema)|7%|1388 (± 0.84%)|
+|[`jsck`](https://github.com/pandastrike/jsck#readme)|3.1%|606 (± 2.84%)|
+|[`jsonschema`](https://github.com/tdegrunt/jsonschema#readme)|0.9%|185 (± 1.01%)|
+|[`skeemas`](https://github.com/Prestaul/skeemas#readme)|0.8%|154 (± 0.79%)|
+|tv4|0.5%|93 (± 0.94%)|
+|[`jayschema`](https://github.com/natesilva/jayschema)|0.1%|21 (± 1.14%)|
+
+换言之，RapidJSON 比最快的 JavaScript 库（ajv）快约 1.5x。比最慢的快 1400x。

data/vendor/simdjson/dependencies/rapidjson/doc/stream.md

@@ -0,0 +1,429 @@
+# Stream
+
+In RapidJSON, `rapidjson::Stream` is a concept for reading/writing JSON. Here we first show how to use streams provided. And then see how to create a custom stream.
+
+[TOC]
+
+# Memory Streams {#MemoryStreams}
+
+Memory streams store JSON in memory.
+
+## StringStream (Input) {#StringStream}
+
+`StringStream` is the most basic input stream. It represents a complete, read-only JSON stored in memory. It is defined in `rapidjson/rapidjson.h`.
+
+~~~~~~~~~~cpp
+#include "rapidjson/document.h" // will include "rapidjson/rapidjson.h"
+
+using namespace rapidjson;
+
+// ...
+const char json[] = "[1, 2, 3, 4]";
+StringStream s(json);
+
+Document d;
+d.ParseStream(s);
+~~~~~~~~~~
+
+Since this is very common usage, `Document::Parse(const char*)` is provided to do exactly the same as above:
+
+~~~~~~~~~~cpp
+// ...
+const char json[] = "[1, 2, 3, 4]";
+Document d;
+d.Parse(json);
+~~~~~~~~~~
+
+Note that, `StringStream` is a typedef of `GenericStringStream<UTF8<> >`, user may use another encodings to represent the character set of the stream.
+
+## StringBuffer (Output) {#StringBuffer}
+
+`StringBuffer` is a simple output stream. It allocates a memory buffer for writing the whole JSON. Use `GetString()` to obtain the buffer.
+
+~~~~~~~~~~cpp
+#include "rapidjson/stringbuffer.h"
+#include <rapidjson/writer.h>
+
+StringBuffer buffer;
+Writer<StringBuffer> writer(buffer);
+d.Accept(writer);
+
+const char* output = buffer.GetString();
+~~~~~~~~~~
+
+When the buffer is full, it will increases the capacity automatically. The default capacity is 256 characters (256 bytes for UTF8, 512 bytes for UTF16, etc.). User can provide an allocator and a initial capacity.
+
+~~~~~~~~~~cpp
+StringBuffer buffer1(0, 1024); // Use its allocator, initial size = 1024
+StringBuffer buffer2(allocator, 1024);
+~~~~~~~~~~
+
+By default, `StringBuffer` will instantiate an internal allocator.
+
+Similarly, `StringBuffer` is a typedef of `GenericStringBuffer<UTF8<> >`.
+
+# File Streams {#FileStreams}
+
+When parsing a JSON from file, you may read the whole JSON into memory and use ``StringStream`` above.
+
+However, if the JSON is big, or memory is limited, you can use `FileReadStream`. It only read a part of JSON from file into buffer, and then let the part be parsed. If it runs out of characters in the buffer, it will read the next part from file.
+
+## FileReadStream (Input) {#FileReadStream}
+
+`FileReadStream` reads the file via a `FILE` pointer. And user need to provide a buffer.
+
+~~~~~~~~~~cpp
+#include "rapidjson/filereadstream.h"
+#include <cstdio>
+
+using namespace rapidjson;
+
+FILE* fp = fopen("big.json", "rb"); // non-Windows use "r"
+
+char readBuffer[65536];
+FileReadStream is(fp, readBuffer, sizeof(readBuffer));
+
+Document d;
+d.ParseStream(is);
+
+fclose(fp);
+~~~~~~~~~~
+
+Different from string streams, `FileReadStream` is byte stream. It does not handle encodings. If the file is not UTF-8, the byte stream can be wrapped in a `EncodedInputStream`. It will be discussed very soon.
+
+Apart from reading file, user can also use `FileReadStream` to read `stdin`.
+
+## FileWriteStream (Output) {#FileWriteStream}
+
+`FileWriteStream` is buffered output stream. Its usage is very similar to `FileReadStream`.
+
+~~~~~~~~~~cpp
+#include "rapidjson/filewritestream.h"
+#include <rapidjson/writer.h>
+#include <cstdio>
+
+using namespace rapidjson;
+
+Document d;
+d.Parse(json);
+// ...
+
+FILE* fp = fopen("output.json", "wb"); // non-Windows use "w"
+
+char writeBuffer[65536];
+FileWriteStream os(fp, writeBuffer, sizeof(writeBuffer));
+
+Writer<FileWriteStream> writer(os);
+d.Accept(writer);
+
+fclose(fp);
+~~~~~~~~~~
+
+It can also directs the output to `stdout`.
+
+# iostream Wrapper {#iostreamWrapper}
+
+Due to users' requests, RapidJSON provided official wrappers for `std::basic_istream` and `std::basic_ostream`. However, please note that the performance will be much lower than the other streams above.
+
+## IStreamWrapper {#IStreamWrapper}
+
+`IStreamWrapper` wraps any class drived from `std::istream`, such as `std::istringstream`, `std::stringstream`, `std::ifstream`, `std::fstream`, into RapidJSON's input stream.
+
+~~~cpp
+#include <rapidjson/document.h>
+#include <rapidjson/istreamwrapper.h>
+#include <fstream>
+
+using namespace rapidjson;
+using namespace std;
+
+ifstream ifs("test.json");
+IStreamWrapper isw(ifs);
+
+Document d;
+d.ParseStream(isw);
+~~~
+
+For classes derived from `std::wistream`, use `WIStreamWrapper`.
+
+## OStreamWrapper {#OStreamWrapper}
+
+Similarly, `OStreamWrapper` wraps any class derived from `std::ostream`, such as `std::ostringstream`, `std::stringstream`, `std::ofstream`, `std::fstream`, into RapidJSON's input stream.
+
+~~~cpp
+#include <rapidjson/document.h>
+#include <rapidjson/ostreamwrapper.h>
+#include <rapidjson/writer.h>
+#include <fstream>
+
+using namespace rapidjson;
+using namespace std;
+
+Document d;
+d.Parse(json);
+
+// ...
+
+ofstream ofs("output.json");
+OStreamWrapper osw(ofs);
+
+Writer<OStreamWrapper> writer(osw);
+d.Accept(writer);
+~~~
+
+For classes derived from `std::wostream`, use `WOStreamWrapper`.
+
+# Encoded Streams {#EncodedStreams}
+
+Encoded streams do not contain JSON itself, but they wrap byte streams to provide basic encoding/decoding function.
+
+As mentioned above, UTF-8 byte streams can be read directly. However, UTF-16 and UTF-32 have endian issue. To handle endian correctly, it needs to convert bytes into characters (e.g. `wchar_t` for UTF-16) while reading, and characters into bytes while writing.
+
+Besides, it also need to handle [byte order mark (BOM)](http://en.wikipedia.org/wiki/Byte_order_mark). When reading from a byte stream, it is needed to detect or just consume the BOM if exists. When writing to a byte stream, it can optionally write BOM.
+
+If the encoding of stream is known in compile-time, you may use `EncodedInputStream` and `EncodedOutputStream`. If the stream can be UTF-8, UTF-16LE, UTF-16BE, UTF-32LE, UTF-32BE JSON, and it is only known in runtime, you may use `AutoUTFInputStream` and `AutoUTFOutputStream`. These streams are defined in `rapidjson/encodedstream.h`.
+
+Note that, these encoded streams can be applied to streams other than file. For example, you may have a file in memory, or a custom byte stream, be wrapped in encoded streams.
+
+## EncodedInputStream {#EncodedInputStream}
+
+`EncodedInputStream` has two template parameters. The first one is a `Encoding` class, such as `UTF8`, `UTF16LE`, defined in `rapidjson/encodings.h`. The second one is the class of stream to be wrapped.
+
+~~~~~~~~~~cpp
+#include "rapidjson/document.h"
+#include "rapidjson/filereadstream.h"   // FileReadStream
+#include "rapidjson/encodedstream.h"    // EncodedInputStream
+#include <cstdio>
+
+using namespace rapidjson;
+
+FILE* fp = fopen("utf16le.json", "rb"); // non-Windows use "r"
+
+char readBuffer[256];
+FileReadStream bis(fp, readBuffer, sizeof(readBuffer));
+
+EncodedInputStream<UTF16LE<>, FileReadStream> eis(bis);  // wraps bis into eis
+
+Document d; // Document is GenericDocument<UTF8<> > 
+d.ParseStream<0, UTF16LE<> >(eis);  // Parses UTF-16LE file into UTF-8 in memory
+
+fclose(fp);
+~~~~~~~~~~
+
+## EncodedOutputStream {#EncodedOutputStream}
+
+`EncodedOutputStream` is similar but it has a `bool putBOM` parameter in the constructor, controlling whether to write BOM into output byte stream.
+
+~~~~~~~~~~cpp
+#include "rapidjson/filewritestream.h"  // FileWriteStream
+#include "rapidjson/encodedstream.h"    // EncodedOutputStream
+#include <rapidjson/writer.h>
+#include <cstdio>
+
+Document d;         // Document is GenericDocument<UTF8<> > 
+// ...
+
+FILE* fp = fopen("output_utf32le.json", "wb"); // non-Windows use "w"
+
+char writeBuffer[256];
+FileWriteStream bos(fp, writeBuffer, sizeof(writeBuffer));
+
+typedef EncodedOutputStream<UTF32LE<>, FileWriteStream> OutputStream;
+OutputStream eos(bos, true);   // Write BOM
+
+Writer<OutputStream, UTF32LE<>, UTF8<>> writer(eos);
+d.Accept(writer);   // This generates UTF32-LE file from UTF-8 in memory
+
+fclose(fp);
+~~~~~~~~~~
+
+## AutoUTFInputStream {#AutoUTFInputStream}
+
+Sometimes an application may want to handle all supported JSON encoding. `AutoUTFInputStream` will detection encoding by BOM first. If BOM is unavailable, it will use  characteristics of valid JSON to make detection. If neither method success, it falls back to the UTF type provided in constructor.
+
+Since the characters (code units) may be 8-bit, 16-bit or 32-bit. `AutoUTFInputStream` requires a character type which can hold at least 32-bit. We may use `unsigned`, as in the template parameter:
+
+~~~~~~~~~~cpp
+#include "rapidjson/document.h"
+#include "rapidjson/filereadstream.h"   // FileReadStream
+#include "rapidjson/encodedstream.h"    // AutoUTFInputStream
+#include <cstdio>
+
+using namespace rapidjson;
+
+FILE* fp = fopen("any.json", "rb"); // non-Windows use "r"
+
+char readBuffer[256];
+FileReadStream bis(fp, readBuffer, sizeof(readBuffer));
+
+AutoUTFInputStream<unsigned, FileReadStream> eis(bis);  // wraps bis into eis
+
+Document d;         // Document is GenericDocument<UTF8<> > 
+d.ParseStream<0, AutoUTF<unsigned> >(eis); // This parses any UTF file into UTF-8 in memory
+
+fclose(fp);
+~~~~~~~~~~
+
+When specifying the encoding of stream, uses `AutoUTF<CharType>` as in `ParseStream()` above.
+
+You can obtain the type of UTF via `UTFType GetType()`. And check whether a BOM is found by `HasBOM()`
+
+## AutoUTFOutputStream {#AutoUTFOutputStream}
+
+Similarly, to choose encoding for output during runtime, we can use `AutoUTFOutputStream`. This class is not automatic *per se*. You need to specify the UTF type and whether to write BOM in runtime.
+
+~~~~~~~~~~cpp
+using namespace rapidjson;
+
+void WriteJSONFile(FILE* fp, UTFType type, bool putBOM, const Document& d) {
+    char writeBuffer[256];
+    FileWriteStream bos(fp, writeBuffer, sizeof(writeBuffer));
+
+    typedef AutoUTFOutputStream<unsigned, FileWriteStream> OutputStream;
+    OutputStream eos(bos, type, putBOM);
+    
+    Writer<OutputStream, UTF8<>, AutoUTF<> > writer;
+    d.Accept(writer);
+}
+~~~~~~~~~~
+
+`AutoUTFInputStream` and `AutoUTFOutputStream` is more convenient than `EncodedInputStream` and `EncodedOutputStream`. They just incur a little bit runtime overheads.
+
+# Custom Stream {#CustomStream}
+
+In addition to memory/file streams, user can create their own stream classes which fits RapidJSON's API. For example, you may create network stream, stream from compressed file, etc.
+
+RapidJSON combines different types using templates. A class containing all required interface can be a stream. The Stream interface is defined in comments of `rapidjson/rapidjson.h`:
+
+~~~~~~~~~~cpp
+concept Stream {
+    typename Ch;    //!< Character type of the stream.
+
+    //! Read the current character from stream without moving the read cursor.
+    Ch Peek() const;
+
+    //! Read the current character from stream and moving the read cursor to next character.
+    Ch Take();
+
+    //! Get the current read cursor.
+    //! \return Number of characters read from start.
+    size_t Tell();
+
+    //! Begin writing operation at the current read pointer.
+    //! \return The begin writer pointer.
+    Ch* PutBegin();
+
+    //! Write a character.
+    void Put(Ch c);
+
+    //! Flush the buffer.
+    void Flush();
+
+    //! End the writing operation.
+    //! \param begin The begin write pointer returned by PutBegin().
+    //! \return Number of characters written.
+    size_t PutEnd(Ch* begin);
+}
+~~~~~~~~~~
+
+For input stream, they must implement `Peek()`, `Take()` and `Tell()`.
+For output stream, they must implement `Put()` and `Flush()`. 
+There are two special interface, `PutBegin()` and `PutEnd()`, which are only for *in situ* parsing. Normal streams do not implement them. However, if the interface is not needed for a particular stream, it is still need to a dummy implementation, otherwise will generate compilation error.
+
+## Example: istream wrapper {#ExampleIStreamWrapper}
+
+The following example is a simple wrapper of `std::istream`, which only implements 3 functions.
+
+~~~~~~~~~~cpp
+class MyIStreamWrapper {
+public:
+    typedef char Ch;
+
+    MyIStreamWrapper(std::istream& is) : is_(is) {
+    }
+
+    Ch Peek() const { // 1
+        int c = is_.peek();
+        return c == std::char_traits<char>::eof() ? '\0' : (Ch)c;
+    }
+
+    Ch Take() { // 2
+        int c = is_.get();
+        return c == std::char_traits<char>::eof() ? '\0' : (Ch)c;
+    }
+
+    size_t Tell() const { return (size_t)is_.tellg(); } // 3
+
+    Ch* PutBegin() { assert(false); return 0; }
+    void Put(Ch) { assert(false); }
+    void Flush() { assert(false); }
+    size_t PutEnd(Ch*) { assert(false); return 0; }
+
+private:
+    MyIStreamWrapper(const MyIStreamWrapper&);
+    MyIStreamWrapper& operator=(const MyIStreamWrapper&);
+
+    std::istream& is_;
+};
+~~~~~~~~~~
+
+User can use it to wrap instances of `std::stringstream`, `std::ifstream`.
+
+~~~~~~~~~~cpp
+const char* json = "[1,2,3,4]";
+std::stringstream ss(json);
+MyIStreamWrapper is(ss);
+
+Document d;
+d.ParseStream(is);
+~~~~~~~~~~
+
+Note that, this implementation may not be as efficient as RapidJSON's memory or file streams, due to internal overheads of the standard library.
+
+## Example: ostream wrapper {#ExampleOStreamWrapper}
+
+The following example is a simple wrapper of `std::istream`, which only implements 2 functions.
+
+~~~~~~~~~~cpp
+class MyOStreamWrapper {
+public:
+    typedef char Ch;
+
+    MyOStreamWrapper(std::ostream& os) : os_(os) {
+    }
+
+    Ch Peek() const { assert(false); return '\0'; }
+    Ch Take() { assert(false); return '\0'; }
+    size_t Tell() const {  }
+
+    Ch* PutBegin() { assert(false); return 0; }
+    void Put(Ch c) { os_.put(c); }                  // 1
+    void Flush() { os_.flush(); }                   // 2
+    size_t PutEnd(Ch*) { assert(false); return 0; }
+
+private:
+    MyOStreamWrapper(const MyOStreamWrapper&);
+    MyOStreamWrapper& operator=(const MyOStreamWrapper&);
+
+    std::ostream& os_;
+};
+~~~~~~~~~~
+
+User can use it to wrap instances of `std::stringstream`, `std::ofstream`.
+
+~~~~~~~~~~cpp
+Document d;
+// ...
+
+std::stringstream ss;
+MyOStreamWrapper os(ss);
+
+Writer<MyOStreamWrapper> writer(os);
+d.Accept(writer);
+~~~~~~~~~~
+
+Note that, this implementation may not be as efficient as RapidJSON's memory or file streams, due to internal overheads of the standard library.
+
+# Summary {#Summary}
+
+This section describes stream classes available in RapidJSON. Memory streams are simple. File stream can reduce the memory required during JSON parsing and generation, if the JSON is stored in file system. Encoded streams converts between byte streams and character streams. Finally, user may create custom streams using a simple interface.