/[svn]/libgig/trunk/src/Serialization.h
ViewVC logotype

Contents of /libgig/trunk/src/Serialization.h

Parent Directory Parent Directory | Revision Log Revision Log


Revision 3392 - (show annotations) (download) (as text)
Sun Dec 3 17:49:22 2017 UTC (2 years, 2 months ago) by schoenebeck
File MIME type: text/x-c++hdr
File size: 52105 byte(s)
* src/Serialization.cpp, src/Serialization.h:
  Hide pure internal declarations from header file to avoid numerous
  compiler warnings when building and linking against the public API.
* Bumped version (4.1.0.svn1).

1 /***************************************************************************
2 * *
3 * Copyright (C) 2017 Christian Schoenebeck *
4 * <cuse@users.sourceforge.net> *
5 * *
6 * This library is part of libgig. *
7 * *
8 * This library is free software; you can redistribute it and/or modify *
9 * it under the terms of the GNU General Public License as published by *
10 * the Free Software Foundation; either version 2 of the License, or *
11 * (at your option) any later version. *
12 * *
13 * This library is distributed in the hope that it will be useful, *
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of *
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the *
16 * GNU General Public License for more details. *
17 * *
18 * You should have received a copy of the GNU General Public License *
19 * along with this library; if not, write to the Free Software *
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, *
21 * MA 02111-1307 USA *
22 ***************************************************************************/
23
24 #ifndef LIBGIG_SERIALIZATION_H
25 #define LIBGIG_SERIALIZATION_H
26
27 #ifdef HAVE_CONFIG_H
28 # include <config.h>
29 #endif
30
31 #include <stdint.h>
32 #include <stdio.h>
33 #include <typeinfo>
34 #include <string>
35 #include <vector>
36 #include <map>
37 #include <time.h>
38 #include <stdarg.h>
39
40 #ifndef __has_extension
41 # define __has_extension(x) 0
42 #endif
43
44 #ifndef HAS_BUILTIN_TYPE_TRAITS
45 # if __cplusplus >= 201103L
46 # define HAS_BUILTIN_TYPE_TRAITS 1
47 # elif ( __has_extension(is_class) && __has_extension(is_enum) )
48 # define HAS_BUILTIN_TYPE_TRAITS 1
49 # elif ( __GNUC__ > 4 || ( __GNUC__ == 4 && __GNUC_MINOR__ >= 3 ) )
50 # define HAS_BUILTIN_TYPE_TRAITS 1
51 # elif _MSC_VER >= 1400 /* MS Visual C++ 8.0 (Visual Studio 2005) */
52 # define HAS_BUILTIN_TYPE_TRAITS 1
53 # elif __INTEL_COMPILER >= 1100
54 # define HAS_BUILTIN_TYPE_TRAITS 1
55 # else
56 # define HAS_BUILTIN_TYPE_TRAITS 0
57 # endif
58 #endif
59
60 #if !HAS_BUILTIN_TYPE_TRAITS
61 # include <tr1/type_traits>
62 # define LIBGIG_IS_CLASS(type) std::tr1::__is_union_or_class<type>::value //NOTE: without compiler support we cannot distinguish union from class
63 #else
64 # define LIBGIG_IS_CLASS(type) __is_class(type)
65 #endif
66
67 /** @brief Serialization / deserialization framework.
68 *
69 * See class Archive as starting point for how to implement serialization and
70 * deserialization with your application.
71 *
72 * The classes in this namespace allow to serialize and deserialize native
73 * C++ objects in a portable, easy and flexible way. Serialization is a
74 * technique that allows to transform the current state and data of native
75 * (in this case C++) objects into a data stream (including all other objects
76 * the "serialized" objects relate to); the data stream may then be sent over
77 * "wire" (for example via network connection to another computer, which might
78 * also have a different OS, CPU architecture, native memory word size and
79 * endian type); and finally the data stream would be "deserialized" on that
80 * receiver side, that is transformed again to modify all objects and data
81 * structures on receiver side to resemble the objects' state and data as it
82 * was originally on sender side.
83 *
84 * In contrast to many other already existing serialization frameworks, this
85 * implementation has a strong focus on robustness regarding long-term changes
86 * to the serialized C++ classes of the serialized objects. So even if sender
87 * and receiver are using different versions of their serialized/deserialized
88 * C++ classes, structures and data types (thus having different data structure
89 * layout to a certain extent), this framework aims trying to automatically
90 * adapt its serialization and deserialization process in that case so that
91 * the deserialized objects on receiver side would still reflect the overall
92 * expected states and overall data as intended by the sender. For being able to
93 * do so, this framework stores all kind of additional information about each
94 * serialized object and each data structure member (for example name of each
95 * data structure member, but also the offset of each member within its
96 * containing data structure, precise data types, and more).
97 *
98 * Like most other serialization frameworks, this frameworks does not require a
99 * tree-structured layout of the serialized data structures. So it automatically
100 * handles also cyclic dependencies between serialized data structures
101 * correctly, without i.e. causing endless recursion or redundancy.
102 *
103 * Additionally this framework also allows partial deserialization. Which means
104 * the receiver side may for example decide that it wants to restrict
105 * deserialization so that it would only modify certain objects or certain
106 * members by the deserialization process, leaving all other ones untouched.
107 * So this partial deserialization technique for example allows to implement
108 * flexible preset features for applications in a powerful and easy way.
109 */
110 namespace Serialization {
111
112 // just symbol prototyping
113 class DataType;
114 class Object;
115 class Member;
116 class Archive;
117 class ObjectPool;
118 class Exception;
119
120 typedef std::string String;
121
122 /** @brief Raw data stream of serialized C++ objects.
123 *
124 * This data type is used for the data stream as a result of serializing
125 * your C++ objects with Archive::serialize(), and for native raw data
126 * representation of individual serialized C/C++ objects, members and variables.
127 *
128 * @see Archive::rawData(), Object::rawData()
129 */
130 typedef std::vector<uint8_t> RawData;
131
132 /** @brief Abstract identifier for serialized C++ objects.
133 *
134 * This data type is used for identifying serialized C++ objects and members
135 * of your C++ objects. It is important to know that such an ID might not
136 * necessarily be unique. For example the ID of one C++ object might often
137 * be identical to the ID of the first member of that particular C++ object.
138 * That's why there is additionally the concept of an UID in this framework.
139 *
140 * @see UID
141 */
142 typedef void* ID;
143
144 /** @brief Version number data type.
145 *
146 * This data type is used for maintaining version number information of
147 * your C++ class implementations.
148 *
149 * @see Archive::setVersion() and Archive::setMinVersion()
150 */
151 typedef uint32_t Version;
152
153 /** @brief To which time zone a certain timing information relates to.
154 *
155 * The constants in this enum type are used to define to which precise time
156 * zone a time stamp relates to.
157 */
158 enum time_base_t {
159 LOCAL_TIME, ///< The time stamp relates to the machine's local time zone. Request a time stamp in local time if you want to present that time stamp to the end user.
160 UTC_TIME ///< The time stamp relates to "Greenwhich Mean Time" zone, also known as "Coordinated Universal Time". Request time stamp with UTC if you want to compare that time stamp with other time stamps.
161 };
162
163 /** @brief Check whether data is a C/C++ @c enum type.
164 *
165 * Returns true if the supplied C++ variable or object is of a C/C++ @c enum
166 * type.
167 *
168 * @param data - the variable or object whose data type shall be checked
169 */
170 template<typename T>
171 bool IsEnum(const T& data) {
172 #if !HAS_BUILTIN_TYPE_TRAITS
173 return std::tr1::is_enum<T>::value;
174 #else
175 return __is_enum(T);
176 #endif
177 }
178
179 /** @brief Check whether data is a C++ @c union type.
180 *
181 * Returns true if the supplied C++ variable or object is of a C/C++ @c union
182 * type. Note that the result of this function is only reliable if the C++
183 * compiler you are using has support for built-in type traits. If your C++
184 * compiler does not have built-in type traits support, then this function
185 * will simply return @c false on all your calls.
186 *
187 * @param data - the variable or object whose data type shall be checked
188 */
189 template<typename T>
190 bool IsUnion(const T& data) {
191 #if !HAS_BUILTIN_TYPE_TRAITS
192 return false; // without compiler support we cannot distinguish union from class
193 #else
194 return __is_union(T);
195 #endif
196 }
197
198 /** @brief Check whether data is a C/C++ @c struct or C++ @c class type.
199 *
200 * Returns true if the supplied C++ variable or object is of C/C++ @c struct
201 * or C++ @c class type. Note that if you are using a C++ compiler which
202 * does have built-in type traits support, then this function will also
203 * return @c true on C/C++ @c union types.
204 *
205 * @param data - the variable or object whose data type shall be checked
206 */
207 template<typename T>
208 bool IsClass(const T& data) {
209 #if !HAS_BUILTIN_TYPE_TRAITS
210 return std::tr1::__is_union_or_class<T>::value; // without compiler support we cannot distinguish union from class
211 #else
212 return __is_class(T);
213 #endif
214 }
215
216 /*template<typename T>
217 bool IsTrivial(T data) {
218 return __is_trivial(T);
219 }*/
220
221 /*template<typename T>
222 bool IsPOD(T data) {
223 return __is_pod(T);
224 }*/
225
226 /** @brief Unique identifier referring to one specific native C++ object, member, fundamental variable, or any other native C++ data.
227 *
228 * Reflects a unique identifier for one specific serialized C++ data, i.e.
229 * C++ class instance, C/C++ struct instance, member, primitive pointer,
230 * fundamental variables, or any other native C/C++ data originally being
231 * serialized.
232 *
233 * A unique identifier is composed of an id (an identifier which is not
234 * necessarily unique) and a size. Since the underlying ID is derived from
235 * the original C++ object's memory location, such an ID is not sufficient
236 * to distinguish a particular C++ object from the first member of that C++
237 * object, since both typically share the same memory address. So
238 * additionally the memory size of the respective object or member is
239 * bundled with UID objects to make them unique and distinguishable.
240 */
241 class UID {
242 public:
243 ID id; ///< Abstract non-unique ID of the object or member in question.
244 size_t size; ///< Memory size of the object or member in question.
245
246 bool isValid() const;
247 operator bool() const { return isValid(); } ///< Same as calling isValid().
248 //bool operator()() const { return isValid(); }
249 bool operator==(const UID& other) const { return id == other.id && size == other.size; }
250 bool operator!=(const UID& other) const { return id != other.id || size != other.size; }
251 bool operator<(const UID& other) const { return id < other.id || (id == other.id && size < other.size); }
252 bool operator>(const UID& other) const { return id > other.id || (id == other.id && size > other.size); }
253
254 /** @brief Create an unique indentifier for a native C++ object/member/variable.
255 *
256 * Creates and returns an unique identifier for the passed native C++
257 * object, object member or variable. For the same C++ object/member/variable
258 * this function will always return the same UID. For all other ones,
259 * this function is guaranteed to return a different UID.
260 */
261 template<typename T>
262 static UID from(const T& obj) {
263 return Resolver<T>::resolve(obj);
264 }
265
266 protected:
267 // UID resolver for non-pointer types
268 template<typename T>
269 struct Resolver {
270 static UID resolve(const T& obj) {
271 const UID uid = { (ID) &obj, sizeof(obj) };
272 return uid;
273 }
274 };
275
276 // UID resolver for pointer types (of 1st degree)
277 template<typename T>
278 struct Resolver<T*> {
279 static UID resolve(const T* const & obj) {
280 const UID uid = { (ID) obj, sizeof(*obj) };
281 return uid;
282 }
283 };
284 };
285
286 /**
287 * Reflects an invalid UID and behaves similar to NULL as invalid value for
288 * pointer types. All UID objects are first initialized with this value,
289 * and it essentially an all zero object.
290 */
291 extern const UID NO_UID;
292
293 /** @brief Chain of UIDs.
294 *
295 * This data type is used for native C++ pointers. The first member of the
296 * UID chain is the unique identifier of the C++ pointer itself, then the
297 * following UIDs are the respective objects or variables the pointer is
298 * pointing to. The size (the amount of elements) of the UIDChain depends
299 * solely on the degree of the pointer type. For example the following C/C++
300 * pointer:
301 * @code
302 * int* pNumber;
303 * @endcode
304 * is an integer pointer of first degree. Such a pointer would have a
305 * UIDChain with 2 members: the first element would be the UID of the
306 * pointer itself, the second element of the chain would be the integer data
307 * that pointer is pointing to. In the following example:
308 * @code
309 * bool*** pppSomeFlag;
310 * @endcode
311 * That boolean pointer would be of third degree, and thus its UIDChain
312 * would have a size of 4 (elements).
313 *
314 * Accordingly a non pointer type like:
315 * @code
316 * float f;
317 * @endcode
318 * would yield in a UIDChain of size 1.
319 *
320 * Since however this serialization framework currently only supports
321 * pointers of first degree yet, all UIDChains are currently either of
322 * size 1 or 2, which might change in future though.
323 */
324 typedef std::vector<UID> UIDChain;
325
326 #if LIBGIG_SERIALIZATION_INTERNAL
327 // prototyping of private internal friend functions
328 static String _encodePrimitiveValue(const Object& obj);
329 static DataType _popDataTypeBlob(const char*& p, const char* end);
330 static Member _popMemberBlob(const char*& p, const char* end);
331 static Object _popObjectBlob(const char*& p, const char* end);
332 static void _popPrimitiveValue(const char*& p, const char* end, Object& obj);
333 static String _primitiveObjectValueToString(const Object& obj);
334 // |
335 template<typename T>
336 static T _primitiveObjectValueToNumber(const Object& obj);
337 #endif // LIBGIG_SERIALIZATION_INTERNAL
338
339 /** @brief Abstract reflection of a native C++ data type.
340 *
341 * Provides detailed information about a serialized C++ data type, whether
342 * it is a fundamental C/C++ data type (like @c int, @c float, @c char,
343 * etc.) or custom defined data types like a C++ @c class, C/C++ @c struct,
344 * @c enum, as well as other features of the respective data type like its
345 * native memory size and more.
346 *
347 * All informations provided by this class are retrieved from the
348 * respective individual C++ objects, their members and other data when
349 * they are serialized, and all those information are stored with the
350 * serialized archive and its resulting data stream. Due to the availability
351 * of these extensive data type information within serialized archives, this
352 * framework is capable to use them in order to adapt its deserialization
353 * process upon subsequent changes to your individual C++ classes.
354 */
355 class DataType {
356 public:
357 DataType();
358 size_t size() const { return m_size; } ///< Returns native memory size of the respective C++ object or variable.
359 bool isValid() const;
360 bool isPointer() const;
361 bool isClass() const;
362 bool isPrimitive() const;
363 bool isInteger() const;
364 bool isReal() const;
365 bool isBool() const;
366 bool isEnum() const;
367 bool isSigned() const;
368 operator bool() const { return isValid(); } ///< Same as calling isValid().
369 //bool operator()() const { return isValid(); }
370 bool operator==(const DataType& other) const;
371 bool operator!=(const DataType& other) const;
372 bool operator<(const DataType& other) const;
373 bool operator>(const DataType& other) const;
374 String asLongDescr() const;
375 String baseTypeName() const;
376 String customTypeName(bool demangle = false) const;
377
378 /** @brief Construct a DataType object for the given native C++ data.
379 *
380 * Use this function to create corresponding DataType objects for
381 * native C/C++ objects, members and variables.
382 *
383 * @param data - native C/C++ object/member/variable a DataType object
384 * shall be created for
385 * @returns corresponding DataType object for the supplied native C/C++
386 * object/member/variable
387 */
388 template<typename T>
389 static DataType dataTypeOf(const T& data) {
390 return Resolver<T>::resolve(data);
391 }
392
393 protected:
394 DataType(bool isPointer, int size, String baseType, String customType = "");
395
396 template<typename T, bool T_isPointer>
397 struct ResolverBase {
398 static DataType resolve(const T& data) {
399 const std::type_info& type = typeid(data);
400 const int sz = sizeof(data);
401
402 // for primitive types we are using our own type names instead of
403 // using std:::type_info::name(), because the precise output of the
404 // latter may vary between compilers
405 if (type == typeid(int8_t)) return DataType(T_isPointer, sz, "int8");
406 if (type == typeid(uint8_t)) return DataType(T_isPointer, sz, "uint8");
407 if (type == typeid(int16_t)) return DataType(T_isPointer, sz, "int16");
408 if (type == typeid(uint16_t)) return DataType(T_isPointer, sz, "uint16");
409 if (type == typeid(int32_t)) return DataType(T_isPointer, sz, "int32");
410 if (type == typeid(uint32_t)) return DataType(T_isPointer, sz, "uint32");
411 if (type == typeid(int64_t)) return DataType(T_isPointer, sz, "int64");
412 if (type == typeid(uint64_t)) return DataType(T_isPointer, sz, "uint64");
413 if (type == typeid(bool)) return DataType(T_isPointer, sz, "bool");
414 if (type == typeid(float)) return DataType(T_isPointer, sz, "real32");
415 if (type == typeid(double)) return DataType(T_isPointer, sz, "real64");
416
417 if (IsEnum(data)) return DataType(T_isPointer, sz, "enum", rawCppTypeNameOf(data));
418 if (IsUnion(data)) return DataType(T_isPointer, sz, "union", rawCppTypeNameOf(data));
419 if (IsClass(data)) return DataType(T_isPointer, sz, "class", rawCppTypeNameOf(data));
420
421 return DataType();
422 }
423 };
424
425 // DataType resolver for non-pointer types
426 template<typename T>
427 struct Resolver : ResolverBase<T,false> {
428 static DataType resolve(const T& data) {
429 return ResolverBase<T,false>::resolve(data);
430 }
431 };
432
433 // DataType resolver for pointer types (of 1st degree)
434 template<typename T>
435 struct Resolver<T*> : ResolverBase<T,true> {
436 static DataType resolve(const T*& data) {
437 return ResolverBase<T,true>::resolve(*data);
438 }
439 };
440
441 template<typename T>
442 static String rawCppTypeNameOf(const T& data) {
443 #if defined _MSC_VER // Microsoft compiler ...
444 # warning type_info::raw_name() demangling has not been tested yet with Microsoft compiler! Feedback appreciated!
445 String name = typeid(data).raw_name(); //NOTE: I haven't checked yet what MSC actually outputs here exactly
446 #else // i.e. especially GCC and clang ...
447 String name = typeid(data).name();
448 #endif
449 //while (!name.empty() && name[0] >= 0 && name[0] <= 9)
450 // name = name.substr(1);
451 return name;
452 }
453
454 private:
455 String m_baseTypeName;
456 String m_customTypeName;
457 int m_size;
458 bool m_isPointer;
459
460 #if LIBGIG_SERIALIZATION_INTERNAL
461 friend DataType _popDataTypeBlob(const char*& p, const char* end);
462 #endif
463 friend class Archive;
464 };
465
466 /** @brief Abstract reflection of a native C++ class/struct's member variable.
467 *
468 * Provides detailed information about a specific C++ member variable of
469 * serialized C++ object, like its C++ data type, offset of this member
470 * within its containing data structure/class, its C++ member variable name
471 * and more.
472 *
473 * Consider you defined the following user defined C/C++ @c struct type in
474 * your application:
475 * @code
476 * struct Foo {
477 * int a;
478 * bool b;
479 * double someValue;
480 * };
481 * @endcode
482 * Then @c a, @c b and @c someValue are "members" of @c struct @c Foo for
483 * instance. So that @c struct would have 3 members in the latter example.
484 *
485 * @see Object::members()
486 */
487 class Member {
488 public:
489 Member();
490 UID uid() const;
491 String name() const;
492 size_t offset() const;
493 const DataType& type() const;
494 bool isValid() const;
495 operator bool() const { return isValid(); } ///< Same as calling isValid().
496 //bool operator()() const { return isValid(); }
497 bool operator==(const Member& other) const;
498 bool operator!=(const Member& other) const;
499 bool operator<(const Member& other) const;
500 bool operator>(const Member& other) const;
501
502 protected:
503 Member(String name, UID uid, size_t offset, DataType type);
504 friend class Archive;
505
506 private:
507 UID m_uid;
508 size_t m_offset;
509 String m_name;
510 DataType m_type;
511
512 #if LIBGIG_SERIALIZATION_INTERNAL
513 friend Member _popMemberBlob(const char*& p, const char* end);
514 #endif
515 };
516
517 /** @brief Abstract reflection of some native serialized C/C++ data.
518 *
519 * When your native C++ objects are serialized, all native data is
520 * translated and reflected by such an Object reflection. So each instance
521 * of your serialized native C++ class objects become available as an
522 * Object, but also each member variable of your C++ objects is translated
523 * into an Object, and any other native C/C++ data. So essentially every
524 * native data is turned into its own Object and accessible by this API.
525 *
526 * For each one of those Object reflections, this class provides detailed
527 * information about their native origin. For example if an Object
528 * represents a native C++ class instante, then it provides access to its
529 * C++ class/struct name, to its C++ member variables, its native memory
530 * size and much more.
531 *
532 * Even though this framework allows you to adjust abstract Object instances
533 * to a certain extent, most of the methods of this Object class are
534 * read-only though and the actual modifyable methods are made available
535 * not as part of this Object class, but as part of the Archive class
536 * instead. This design decision was made for performance and safety
537 * reasons.
538 *
539 * @see Archive::setIntValue() as an example for modifying Object instances.
540 */
541 class Object {
542 public:
543 Object();
544 Object(UIDChain uidChain, DataType type);
545
546 UID uid(int index = 0) const;
547 const UIDChain& uidChain() const;
548 const DataType& type() const;
549 const RawData& rawData() const;
550 Version version() const;
551 Version minVersion() const;
552 bool isVersionCompatibleTo(const Object& other) const;
553 std::vector<Member>& members();
554 const std::vector<Member>& members() const;
555 Member memberNamed(String name) const;
556 Member memberByUID(const UID& uid) const;
557 std::vector<Member> membersOfType(const DataType& type) const;
558 int sequenceIndexOf(const Member& member) const;
559 bool isValid() const;
560 operator bool() const { return isValid(); } ///< Same as calling isValid().
561 //bool operator()() const { return isValid(); }
562 bool operator==(const Object& other) const;
563 bool operator!=(const Object& other) const;
564 bool operator<(const Object& other) const;
565 bool operator>(const Object& other) const;
566
567 protected:
568 void remove(const Member& member);
569 void setVersion(Version v);
570 void setMinVersion(Version v);
571
572 private:
573 DataType m_type;
574 UIDChain m_uid;
575 Version m_version;
576 Version m_minVersion;
577 RawData m_data;
578 std::vector<Member> m_members;
579
580 #if LIBGIG_SERIALIZATION_INTERNAL
581 friend String _encodePrimitiveValue(const Object& obj);
582 friend Object _popObjectBlob(const char*& p, const char* end);
583 friend void _popPrimitiveValue(const char*& p, const char* end, Object& obj);
584 friend String _primitiveObjectValueToString(const Object& obj);
585 // |
586 template<typename T>
587 friend T _primitiveObjectValueToNumber(const Object& obj);
588 #endif // LIBGIG_SERIALIZATION_INTERNAL
589
590 friend class Archive;
591 };
592
593 /** @brief Destination container for serialization, and source container for deserialization.
594 *
595 * This is the main class for implementing serialization and deserialization
596 * with your C++ application. This framework does not require a a tree
597 * structured layout of your C++ objects being serialized/deserialized, it
598 * uses a concept of a "root" object though. So to start serialization
599 * construct an empty Archive object and then instruct it to serialize your
600 * C++ objects by pointing it to your "root" object:
601 * @code
602 * Archive a;
603 * a.serialize(&myRootObject);
604 * @endcode
605 * Or if you prefer the look of operator based code:
606 * @code
607 * Archive a;
608 * a << myRootObject;
609 * @endcode
610 * The Archive object will then serialize all members of the passed C++
611 * object, and will recursively serialize all other C++ objects which it
612 * contains or points to. So the root object is the starting point for the
613 * overall serialization. After the serialize() method returned, you can
614 * then access the serialized data stream by calling rawData() and send that
615 * data stream over "wire", or store it on disk or whatever you may intend
616 * to do with it.
617 *
618 * Then on receiver side likewise, you create a new Archive object, pass the
619 * received data stream i.e. via constructor to the Archive object and call
620 * deserialize() by pointing it to the root object on receiver side:
621 * @code
622 * Archive a(rawDataStream);
623 * a.deserialize(&myRootObject);
624 * @endcode
625 * Or with operator instead:
626 * @code
627 * Archive a(rawDataStream);
628 * a >> myRootObject;
629 * @endcode
630 * Now this framework automatically handles serialization and
631 * deserialization of fundamental data types automatically for you (like
632 * i.e. char, int, long int, float, double, etc.). However for your own
633 * custom C++ classes and structs you must implement one method which
634 * defines which members of your class should actually be serialized and
635 * deserialized. That method to be added must have the following signature:
636 * @code
637 * void serialize(Serialization::Archive* archive);
638 * @endcode
639 * So let's say you have the following simple data structures:
640 * @code
641 * struct Foo {
642 * int a;
643 * bool b;
644 * double c;
645 * };
646 *
647 * struct Bar {
648 * char one;
649 * float two;
650 * Foo foo1;
651 * Foo* pFoo2;
652 * Foo* pFoo3DontTouchMe; // shall not be serialized/deserialized
653 * };
654 * @endcode
655 * So in order to be able to serialize and deserialize objects of those two
656 * structures you would first add the mentioned method to each struct
657 * definition (i.e. in your header file):
658 * @code
659 * struct Foo {
660 * int a;
661 * bool b;
662 * double c;
663 *
664 * void serialize(Serialization::Archive* archive);
665 * };
666 *
667 * struct Bar {
668 * char one;
669 * float two;
670 * Foo foo1;
671 * Foo* pFoo2;
672 * Foo* pFoo3DontTouchMe; // shall not be serialized/deserialized
673 *
674 * void serialize(Serialization::Archive* archive);
675 * };
676 * @endcode
677 * And then you would implement those two new methods like this (i.e. in
678 * your .cpp file):
679 * @code
680 * #define SRLZ(member) \
681 * archive->serializeMember(*this, member, #member);
682 *
683 * void Foo::serialize(Serialization::Archive* archive) {
684 * SRLZ(a);
685 * SRLZ(b);
686 * SRLZ(c);
687 * }
688 *
689 * void Bar::serialize(Serialization::Archive* archive) {
690 * SRLZ(one);
691 * SRLZ(two);
692 * SRLZ(foo1);
693 * SRLZ(pFoo2);
694 * // leaving out pFoo3DontTouchMe here
695 * }
696 * @endcode
697 * Now when you serialize such a Bar object, this framework will also
698 * automatically serialize the respective Foo object(s) accordingly, also
699 * for the pFoo2 pointer for instance (as long as it is not a NULL pointer
700 * that is).
701 *
702 * Note that there is only one method that you need to implement. So the
703 * respective serialize() method implementation of your classes/structs are
704 * both called for serialization, as well as for deserialization!
705 *
706 * In case you need to enforce backward incompatiblity for one of your C++
707 * classes, you can do so by setting a version and minimum version for your
708 * class (see @c setVersion() and @c setMinVersion() for details).
709 */
710 class Archive {
711 public:
712 Archive();
713 Archive(const RawData& data);
714 Archive(const uint8_t* data, size_t size);
715 virtual ~Archive();
716
717 /** @brief Initiate serialization.
718 *
719 * Initiates serialization of all native C++ objects, which means
720 * capturing and storing the current data of all your C++ objects as
721 * content of this Archive.
722 *
723 * This framework has a concept of a "root" object which you must pass
724 * to this method. The root object is the starting point for
725 * serialization of your C++ objects. The framework will then
726 * recursively serialize all members of that C++ object an continue to
727 * serialize all other C++ objects that it might contain or point to.
728 *
729 * After this method returned, you might traverse all serialized objects
730 * by walking them starting from the rootObject(). You might then modify
731 * that abstract reflection of your C++ objects and finally you might
732 * call rawData() to get an encoded raw data stream which you might use
733 * for sending it "over wire" to somewhere where it is going to be
734 * deserialized later on.
735 *
736 * Note that whenever you call this method, the previous content of this
737 * Archive will first be cleared.
738 *
739 * @param obj - native C++ root object where serialization shall start
740 * @see Archive::operator<<()
741 */
742 template<typename T>
743 void serialize(const T* obj) {
744 m_operation = OPERATION_SERIALIZE;
745 m_allObjects.clear();
746 m_rawData.clear();
747 m_root = UID::from(obj);
748 const_cast<T*>(obj)->serialize(this);
749 encode();
750 m_operation = OPERATION_NONE;
751 }
752
753 /** @brief Initiate deserialization.
754 *
755 * Initiates deserialization of all native C++ objects, which means all
756 * your C++ objects will be restored with the values contained in this
757 * Archive. So that also means calling deserialize() only makes sense if
758 * this a non-empty Archive, which i.e. is the case if you either called
759 * serialize() with this Archive object before or if you passed a
760 * previously serialized raw data stream to the constructor of this
761 * Archive object.
762 *
763 * This framework has a concept of a "root" object which you must pass
764 * to this method. The root object is the starting point for
765 * deserialization of your C++ objects. The framework will then
766 * recursively deserialize all members of that C++ object an continue to
767 * deserialize all other C++ objects that it might contain or point to,
768 * according to the values stored in this Archive.
769 *
770 * @param obj - native C++ root object where deserialization shall start
771 * @see Archive::operator>>()
772 *
773 * @throws Exception if the data stored in this Archive cannot be
774 * restored to the C++ objects passed to this method, i.e.
775 * because of version or type incompatibilities.
776 */
777 template<typename T>
778 void deserialize(T* obj) {
779 Archive a;
780 m_operation = OPERATION_DESERIALIZE;
781 obj->serialize(&a);
782 a.m_root = UID::from(obj);
783 Syncer s(a, *this);
784 m_operation = OPERATION_NONE;
785 }
786
787 /** @brief Initiate serialization of your C++ objects.
788 *
789 * Same as calling @c serialize(), this is just meant if you prefer
790 * to use operator based code instead, which you might find to be more
791 * intuitive.
792 *
793 * Example:
794 * @code
795 * Archive a;
796 * a << myRootObject;
797 * @endcode
798 *
799 * @see Archive::serialize() for more details.
800 */
801 template<typename T>
802 void operator<<(const T& obj) {
803 serialize(&obj);
804 }
805
806 /** @brief Initiate deserialization of your C++ objects.
807 *
808 * Same as calling @c deserialize(), this is just meant if you prefer
809 * to use operator based code instead, which you might find to be more
810 * intuitive.
811 *
812 * Example:
813 * @code
814 * Archive a(rawDataStream);
815 * a >> myRootObject;
816 * @endcode
817 *
818 * @throws Exception if the data stored in this Archive cannot be
819 * restored to the C++ objects passed to this method, i.e.
820 * because of version or type incompatibilities.
821 *
822 * @see Archive::deserialize() for more details.
823 */
824 template<typename T>
825 void operator>>(T& obj) {
826 deserialize(&obj);
827 }
828
829 const RawData& rawData();
830 virtual String rawDataFormat() const;
831
832 /** @brief Serialize a native C/C++ member variable.
833 *
834 * This method is usually called by the serialize() method
835 * implementation of your C/C++ structs and classes, for each of the
836 * member variables that shall be serialized and deserialized
837 * automatically with this framework. It is recommend that you are not
838 * using this method name directly, but rather define a short hand C
839 * macro in your .cpp file like:
840 * @code
841 * #define SRLZ(member) \
842 * archive->serializeMember(*this, member, #member);
843 *
844 * void Foo::serialize(Serialization::Archive* archive) {
845 * SRLZ(a);
846 * SRLZ(b);
847 * SRLZ(c);
848 * }
849 * @endcode
850 * As you can see, using such a macro makes your code more readable and
851 * less error prone.
852 *
853 * It is completely up to you to decide which ones of your member
854 * variables shall automatically be serialized and deserialized with
855 * this framework. Only those member variables which are registered by
856 * calling this method will be serialized and deserialized. It does not
857 * really matter in which order you register your individiual member
858 * variables by calling this method, but the sequence is actually stored
859 * as meta information with the resulting archive and the resulting raw
860 * data stream. That meta information might then be used by this
861 * framework to automatically correct and adapt deserializing that
862 * archive later on for a future (or older) and potentially heavily
863 * modified version of your software. So it is recommended, even though
864 * also not required, that you may retain the sequence of your
865 * serializeMember() calls for your individual C++ classes' members over
866 * all your software versions, to retain backward compatibility of older
867 * archives as much as possible.
868 *
869 * @param nativeObject - native C++ object to be registered for
870 * serialization / deserialization
871 * @param nativeMember - native C++ member variable of @a nativeObject
872 * to be registered for serialization /
873 * deserialization
874 * @param memberName - name of @a nativeMember to be stored with this
875 * archive
876 */
877 template<typename T_classType, typename T_memberType>
878 void serializeMember(const T_classType& nativeObject, const T_memberType& nativeMember, const char* memberName) {
879 const size_t offset =
880 ((const uint8_t*)(const void*)&nativeMember) -
881 ((const uint8_t*)(const void*)&nativeObject);
882 const UIDChain uids = UIDChainResolver<T_memberType>(nativeMember);
883 const DataType type = DataType::dataTypeOf(nativeMember);
884 const Member member(memberName, uids[0], offset, type);
885 const UID parentUID = UID::from(nativeObject);
886 Object& parent = m_allObjects[parentUID];
887 if (!parent) {
888 const UIDChain uids = UIDChainResolver<T_classType>(nativeObject);
889 const DataType type = DataType::dataTypeOf(nativeObject);
890 parent = Object(uids, type);
891 }
892 parent.members().push_back(member);
893 const Object obj(uids, type);
894 const bool bExistsAlready = m_allObjects.count(uids[0]);
895 const bool isValidObject = obj;
896 const bool bExistingObjectIsInvalid = !m_allObjects[uids[0]];
897 if (!bExistsAlready || (bExistingObjectIsInvalid && isValidObject)) {
898 m_allObjects[uids[0]] = obj;
899 // recurse serialization for all members of this member
900 // (only for struct/class types, noop for primitive types)
901 SerializationRecursion<T_memberType>::serializeObject(this, nativeMember);
902 }
903 }
904
905 /** @brief Set current version number for your C++ class.
906 *
907 * By calling this method you can define a version number for your
908 * current C++ class (that is a version for its current data structure
909 * layout and method implementations) that is going to be stored along
910 * with the serialized archive. Only call this method if you really want
911 * to constrain compatibility of your C++ class.
912 *
913 * Along with calling @c setMinVersion() this provides a way for you
914 * to constrain backward compatibility regarding serialization and
915 * deserialization of your C++ class which the Archive class will obey
916 * to. If required, then typically you might do so in your
917 * @c serialize() method implementation like:
918 * @code
919 * #define SRLZ(member) \
920 * archive->serializeMember(*this, member, #member);
921 *
922 * void Foo::serialize(Serialization::Archive* archive) {
923 * // when serializing: the current version of this class that is
924 * // going to be stored with the serialized archive
925 * archive->setVersion(*this, 6);
926 * // when deserializing: the minimum version this C++ class is
927 * // compatible with
928 * archive->setMinVersion(*this, 3);
929 * // actual data mebers to serialize / deserialize
930 * SRLZ(a);
931 * SRLZ(b);
932 * SRLZ(c);
933 * }
934 * @endcode
935 * In this example above, the C++ class "Foo" would be serialized along
936 * with the version number @c 6 and minimum version @c 3 as additional
937 * meta information in the resulting archive (and its raw data stream
938 * respectively).
939 *
940 * When deserializing archives with the example C++ class code above,
941 * the Archive object would check whether your originally serialized
942 * C++ "Foo" object had at least version number @c 3, if not the
943 * deserialization process would automatically be stopped with a
944 * @c Serialization::Exception, claiming that the classes are version
945 * incompatible.
946 *
947 * But also consider the other way around: you might have serialized
948 * your latest version of your C++ class, and might deserialize that
949 * archive with an older version of your C++ class. In that case it will
950 * likewise be checked whether the version of that old C++ class is at
951 * least as high as the minimum version set with the already seralized
952 * bleeding edge C++ class.
953 *
954 * Since this Serialization / deserialization framework is designed to
955 * be robust on changes to your C++ classes and aims trying to
956 * deserialize all your C++ objects correctly even if your C++ classes
957 * have seen substantial software changes in the meantime; you might
958 * sometimes see it as necessary to constrain backward compatibility
959 * this way. Because obviously there are certain things this framework
960 * can cope with, like for example that you renamed a data member while
961 * keeping the layout consistent, or that you have added new members to
962 * your C++ class or simply changed the order of your members in your
963 * C++ class. But what this framework cannot detect is for example if
964 * you changed the semantics of the values stored with your members, or
965 * even substantially changed the algorithms in your class methods such
966 * that they would not handle the data of your C++ members in the same
967 * and correct way anymore.
968 *
969 * @param nativeObject - your C++ object you want to set a version for
970 * @param v - the version number to set for your C++ class (by default,
971 * that is if you do not explicitly call this method, then
972 * your C++ object will be stored with version number @c 0 ).
973 */
974 template<typename T_classType>
975 void setVersion(const T_classType& nativeObject, Version v) {
976 const UID uid = UID::from(nativeObject);
977 Object& obj = m_allObjects[uid];
978 if (!obj) {
979 const UIDChain uids = UIDChainResolver<T_classType>(nativeObject);
980 const DataType type = DataType::dataTypeOf(nativeObject);
981 obj = Object(uids, type);
982 }
983 setVersion(obj, v);
984 }
985
986 /** @brief Set a minimum version number for your C++ class.
987 *
988 * Call this method to define a minimum version that your current C++
989 * class implementation would be compatible with when it comes to
990 * deserialization of an archive containing an object of your C++ class.
991 * Like the version information, the minimum version will also be stored
992 * for objects of your C++ class with the resulting archive (and its
993 * resulting raw data stream respectively).
994 *
995 * When you start to constrain version compatibility of your C++ class
996 * you usually start by using 1 as version and 1 as minimum version.
997 * So it is eligible to set the same number to both version and minimum
998 * version. However you must @b not set a minimum version higher than
999 * version. Doing so would not raise an exception, but the resulting
1000 * behavior would be undefined.
1001 *
1002 * It is not relevant whether you first set version and then minimum
1003 * version or vice versa. It is also not relevant when exactly you set
1004 * those two numbers, even though usually you would set both in your
1005 * serialize() method implementation.
1006 *
1007 * @see @c setVersion() for more details about this overall topic.
1008 *
1009 * @param nativeObject - your C++ object you want to set a version for
1010 * @param v - the minimum version you want to define for your C++ class
1011 * (by default, that is if you do not explicitly call this
1012 * method, then a minium version of @c 0 is assumed for your
1013 * C++ class instead).
1014 */
1015 template<typename T_classType>
1016 void setMinVersion(const T_classType& nativeObject, Version v) {
1017 const UID uid = UID::from(nativeObject);
1018 Object& obj = m_allObjects[uid];
1019 if (!obj) {
1020 const UIDChain uids = UIDChainResolver<T_classType>(nativeObject);
1021 const DataType type = DataType::dataTypeOf(nativeObject);
1022 obj = Object(uids, type);
1023 }
1024 setMinVersion(obj, v);
1025 }
1026
1027 virtual void decode(const RawData& data);
1028 virtual void decode(const uint8_t* data, size_t size);
1029 void clear();
1030 bool isModified() const;
1031 void removeMember(Object& parent, const Member& member);
1032 void remove(const Object& obj);
1033 Object& rootObject();
1034 Object& objectByUID(const UID& uid);
1035 void setAutoValue(Object& object, String value);
1036 void setIntValue(Object& object, int64_t value);
1037 void setRealValue(Object& object, double value);
1038 void setBoolValue(Object& object, bool value);
1039 void setEnumValue(Object& object, uint64_t value);
1040 String valueAsString(const Object& object);
1041 int64_t valueAsInt(const Object& object);
1042 double valueAsReal(const Object& object);
1043 bool valueAsBool(const Object& object);
1044 void setVersion(Object& object, Version v);
1045 void setMinVersion(Object& object, Version v);
1046 String name() const;
1047 void setName(String name);
1048 String comment() const;
1049 void setComment(String comment);
1050 time_t timeStampCreated() const;
1051 time_t timeStampModified() const;
1052 tm dateTimeCreated(time_base_t base = LOCAL_TIME) const;
1053 tm dateTimeModified(time_base_t base = LOCAL_TIME) const;
1054
1055 protected:
1056 // UID resolver for non-pointer types
1057 template<typename T>
1058 class UIDChainResolver {
1059 public:
1060 UIDChainResolver(const T& data) {
1061 m_uid.push_back(UID::from(data));
1062 }
1063
1064 operator UIDChain() const { return m_uid; }
1065 UIDChain operator()() const { return m_uid; }
1066 private:
1067 UIDChain m_uid;
1068 };
1069
1070 // UID resolver for pointer types (of 1st degree)
1071 template<typename T>
1072 class UIDChainResolver<T*> {
1073 public:
1074 UIDChainResolver(const T*& data) {
1075 const UID uids[2] = {
1076 { &data, sizeof(data) },
1077 { data, sizeof(*data) }
1078 };
1079 m_uid.push_back(uids[0]);
1080 m_uid.push_back(uids[1]);
1081 }
1082
1083 operator UIDChain() const { return m_uid; }
1084 UIDChain operator()() const { return m_uid; }
1085 private:
1086 UIDChain m_uid;
1087 };
1088
1089 // SerializationRecursion for non-pointer class/struct types.
1090 template<typename T, bool T_isRecursive>
1091 struct SerializationRecursionImpl {
1092 static void serializeObject(Archive* archive, const T& obj) {
1093 const_cast<T&>(obj).serialize(archive);
1094 }
1095 };
1096
1097 // SerializationRecursion for pointers (of 1st degree) to class/structs.
1098 template<typename T, bool T_isRecursive>
1099 struct SerializationRecursionImpl<T*,T_isRecursive> {
1100 static void serializeObject(Archive* archive, const T*& obj) {
1101 if (!obj) return;
1102 const_cast<T*&>(obj)->serialize(archive);
1103 }
1104 };
1105
1106 // NOOP SerializationRecursion for primitive types.
1107 template<typename T>
1108 struct SerializationRecursionImpl<T,false> {
1109 static void serializeObject(Archive* archive, const T& obj) {}
1110 };
1111
1112 // NOOP SerializationRecursion for pointers (of 1st degree) to primitive types.
1113 template<typename T>
1114 struct SerializationRecursionImpl<T*,false> {
1115 static void serializeObject(Archive* archive, const T*& obj) {}
1116 };
1117
1118 // Automatically handles recursion for class/struct types, while ignoring all primitive types.
1119 template<typename T>
1120 struct SerializationRecursion : SerializationRecursionImpl<T, LIBGIG_IS_CLASS(T)> {
1121 };
1122
1123 class ObjectPool : public std::map<UID,Object> {
1124 public:
1125 // prevent passing obvious invalid UID values from creating a new pair entry
1126 Object& operator[](const UID& k) {
1127 static Object invalid;
1128 if (!k.isValid()) {
1129 invalid = Object();
1130 return invalid;
1131 }
1132 return std::map<UID,Object>::operator[](k);
1133 }
1134 };
1135
1136 friend String _encode(const ObjectPool& objects);
1137
1138 private:
1139 String _encodeRootBlob();
1140 void _popRootBlob(const char*& p, const char* end);
1141 void _popObjectsBlob(const char*& p, const char* end);
1142
1143 protected:
1144 class Syncer {
1145 public:
1146 Syncer(Archive& dst, Archive& src);
1147 protected:
1148 void syncObject(const Object& dst, const Object& src);
1149 void syncPrimitive(const Object& dst, const Object& src);
1150 void syncPointer(const Object& dst, const Object& src);
1151 void syncMember(const Member& dstMember, const Member& srcMember);
1152 static Member dstMemberMatching(const Object& dstObj, const Object& srcObj, const Member& srcMember);
1153 private:
1154 Archive& m_dst;
1155 Archive& m_src;
1156 };
1157
1158 enum operation_t {
1159 OPERATION_NONE,
1160 OPERATION_SERIALIZE,
1161 OPERATION_DESERIALIZE
1162 };
1163
1164 virtual void encode();
1165
1166 ObjectPool m_allObjects;
1167 operation_t m_operation;
1168 UID m_root;
1169 RawData m_rawData;
1170 bool m_isModified;
1171 String m_name;
1172 String m_comment;
1173 time_t m_timeCreated;
1174 time_t m_timeModified;
1175 };
1176
1177 /**
1178 * Will be thrown whenever an error occurs during an serialization or
1179 * deserialization process.
1180 */
1181 class Exception {
1182 public:
1183 String Message;
1184
1185 Exception(String format, ...);
1186 Exception(String format, va_list arg);
1187 void PrintMessage();
1188 virtual ~Exception() {}
1189
1190 protected:
1191 Exception();
1192 static String assemble(String format, va_list arg);
1193 };
1194
1195 } // namespace Serialization
1196
1197 #endif // LIBGIG_SERIALIZATION_H

  ViewVC Help
Powered by ViewVC