/* OPAQUE-HOLDER.hpp - buffer holding an object inline while hiding the concrete type Copyright (C) Lumiera.org 2009, Hermann Vosseler This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ /** @file opaque-holder.hpp ** Helper allowing type erasure while holding the actual object inline. ** Controlling the actual storage of objects usually binds us to commit ** to a specific type, thus ruling out polymorphism. But sometimes, when ** we are able to control the maximum storage for a family of classes, we ** can escape this dilemma by using the type erasure pattern combined with ** an inline buffer holding an object of the concrete subclass. Typically, ** this situation arises when dealing with functor objects. ** ** These templates help with building custom objects and wrappers based on ** this pattern: InPlaceAnyHolder provides an buffer for the target objects ** and controls access through a two-layer capsule; while the outer container ** exposes a neutral interface, the inner container keeps track of the actual ** type by means of a vtable. OpaqueHolder is built on top of InPlaceAnyHolder ** additionally to support a "common base interface" and re-access of the ** embedded object through this interface. For this to work, all of the ** stored types need to be derived from this common base interface. ** OpaqueHolder then may be even used like a smart-ptr, exposing this ** base interface. To the contrary, InPlaceAnyHolder has lesser requirements ** on the types to be stored within. It can be configured with policy classes ** to control the re-access; when using InPlaceAnyHolder_unrelatedTypes ** the individual types to be stored need not be related in any way, but ** of course this rules out anything beyond re-accessing the embedded object ** by knowing it's exact type. Generally speaking, re-accessing the concrete ** object requires knowledge of the actual type, similar to boost::any ** (but contrary to OpaqueHolder the latter uses heap storage). ** ** Using this approach is bound to specific stipulations regarding the ** properties of the contained object and the kind of access needed. ** When, to the contrary, the contained types are \em not related ** and you need to re-discover their concrete type, then maybe ** a visitor or variant record might be a better solution. ** ** @see opaque-holder-test.cpp ** @see function-erasure.hpp usage example ** @see variant.hpp */ #ifndef LIB_OPAQUE_HOLDER_H #define LIB_OPAQUE_HOLDER_H #include "lib/error.hpp" #include "lib/bool-checkable.hpp" #include "lib/access-casted.hpp" #include "lib/util.hpp" #include namespace lib { using lumiera::error::LUMIERA_ERROR_WRONG_TYPE; using util::isSameObject; using util::unConst; namespace { // implementation helpers... using boost::disable_if; using boost::is_convertible; bool validitySelfCheck (bool boolConvertible) { return boolConvertible; } template typename disable_if< is_convertible, bool >::type validitySelfCheck (X const&) { return true; // just pass if this type doesn't provide a validity check... } } /* ==== Policy classes controlling re-Access ==== */ /** * Standard policy for accessing the contents via * a common base class interface. Using this policy * causes static or dynamic casts or direct conversion * to be employed as appropriate. */ template struct InPlaceAnyHolder_useCommonBase { typedef BA Base; template static Base* convert2base (SUB& obj) { SUB* oPtr = &obj; BA* asBase = util::AccessCasted::access (oPtr); if (asBase) return asBase; throw lumiera::error::Logic ("Unable to convert concrete object to Base interface" , LUMIERA_ERROR_WRONG_TYPE ); } template static SUB* access (Base* asBase) { // Because we don't know anything about the involved types, // we need to exclude a brute force static cast // (which might slice or reinterpret or even cause SEGV) if (!util::use_static_downcast::value) { SUB* content = util::AccessCasted::access (asBase); return content; // might be NULL } else return 0; } }; /** * Alternative policy for accessing the contents without * a common interface; use this policy if the intention is * to use OpaqueHolder with a family of similar classes, * \em without requiring all of them to be derived from * a common base class. (E.g. tr1::function objects). * In this case, the "Base" type will be defined to void* * As a consequence, we loose all type information and * no conversions are possible on re-access. You need * to know the \em exact type to get back at the object. */ struct InPlaceAnyHolder_unrelatedTypes { typedef void Base; template static void* convert2base (SUB& obj) { return static_cast (&obj); } template static SUB* access (Base*) { return 0; } }; /** * Inline buffer holding and owning an object while concealing the * concrete type. The object is given either as ctor parameter or * by direct assignment; it is copy-constructed into the buffer. * It is necessary to specify the required buffer storage space * as a template parameter. InPlaceAnyHolder may be created empty * or cleared afterwards, and this #empty() state may be detected * at runtime. In a similar vein, when the stored object has a * \c bool validity check, this can be accessed though #isValid(). * Moreover \code !empty() && isValid() \endcode may be tested * as by \bool conversion of the Holder object. The whole compound * is copyable if and only if the contained object is copyable. * * @note assertion failure when trying to place an instance not * fitting into given size. * @note \em not threadsafe! */ template < size_t siz ///< maximum storage required for the targets to be held inline , class AccessPolicy = InPlaceAnyHolder_unrelatedTypes ///< how to access the contents via a common interface? > class InPlaceAnyHolder : public BoolCheckable > { typedef typename AccessPolicy::Base * BaseP; /** Inner capsule managing the contained object (interface) */ struct Buffer { char content_[siz]; void* ptr() { return &content_; } virtual ~Buffer() {} virtual bool isValid() const =0; virtual bool empty() const =0; virtual BaseP getBase() const =0; virtual void clone (void* targetStorage) const =0; }; /** special case: no stored object */ struct EmptyBuff : Buffer { virtual bool isValid() const { return false; } virtual bool empty() const { return true; } BaseP getBase() const { throw lumiera::error::Invalid("accessing empty holder"); } virtual void clone (void* targetStorage) const { new(targetStorage) EmptyBuff(); } }; /** concrete subclass managing a specific kind of contained object. * @note invariant: #content_ always contains a valid SUB object */ template struct Buff : Buffer { SUB& get() const ///< core operation: target is contained within the inline buffer { return *reinterpret_cast (unConst(this)->ptr()); } ~Buff() { get().~SUB(); } explicit Buff (SUB const& obj) { REQUIRE (siz >= sizeof(SUB)); new(Buffer::ptr()) SUB (obj); } Buff (Buff const& oBuff) { new(Buffer::ptr()) SUB (oBuff.get()); } Buff& operator= (Buff const& ref) ///< currently not used { if (&ref != this) get() = ref.get(); return *this; } /* == virtual access functions == */ virtual void clone (void* targetStorage) const { new(targetStorage) Buff(get()); } virtual BaseP getBase() const { return AccessPolicy::convert2base (get()); } virtual bool empty() const { return false; } virtual bool isValid() const { return validitySelfCheck (this->get()); } }; enum{ BUFFSIZE = sizeof(Buffer) }; /** embedded buffer actually holding the concrete Buff object, * which in turn holds and manages the target object. * @note Invariant: always contains a valid Buffer subclass */ char storage_[BUFFSIZE]; protected: /* === internal interface for managing the storage === */ Buffer& buff() { return *reinterpret_cast (&storage_); } const Buffer& buff() const { return *reinterpret_cast (&storage_); } void killBuffer() { buff().~Buffer(); } void make_emptyBuff() { new(&storage_) EmptyBuff(); } template void place_inBuff (SUB const& obj) { new(&storage_) Buff (obj); } void clone_inBuff (InPlaceAnyHolder const& ref) { ref.buff().clone (storage_); } public: ~InPlaceAnyHolder() { killBuffer(); } void clear () { killBuffer(); make_emptyBuff(); } InPlaceAnyHolder() { make_emptyBuff(); } template InPlaceAnyHolder(SUB const& obj) { place_inBuff (obj); } InPlaceAnyHolder (InPlaceAnyHolder const& ref) { clone_inBuff (ref); } InPlaceAnyHolder& operator= (InPlaceAnyHolder const& ref) { if (!isSameObject (*this, ref)) { killBuffer(); try { clone_inBuff (ref); } catch (...) { make_emptyBuff(); throw; } } return *this; } template InPlaceAnyHolder& operator= (SUB const& newContent) { if ( empty() || !isSameObject (*buff().getBase(), newContent) ) { killBuffer(); try { place_inBuff (newContent); } catch (...) { make_emptyBuff(); throw; } } return *this; } /** re-accessing the concrete contained object. * When the specified type is exactly the same * as used when storing the object, we can directly * re-access the buffer. Otherwise, a conversion might * be possible going through the Base type, depending * on the actual types involved and the AccessPolicy. * But, as we don't "know" the actual type of the object * in storage, a \em static upcast to any type \em between * the concrete object type and the base type has to be * ruled out for safety reasons. When the contained object * has RTTI, a \em dynamic cast can be performed in this * situation. You might consider using visitor.hpp instead * if this imposes a serious limitation. * @throws lumiera::error::Logic when conversion/access fails */ template SUB& get() const { typedef const Buffer* Iface; typedef const Buff * Actual; Iface interface = &buff(); Actual actual = dynamic_cast (interface); if (actual) return actual->get(); // second try: maybe we can perform a dynamic downcast // or direct conversion to the actual target type. BaseP asBase = buff().getBase(); ASSERT (asBase); SUB* content = AccessPolicy::template access (asBase); if (content) return *content; throw lumiera::error::Logic ("Attempt to access OpaqueHolder's contents " "specifying incompatible target type" , LUMIERA_ERROR_WRONG_TYPE ); } bool empty() const { return buff().empty(); } bool isValid() const { return buff().isValid(); } }; /** * Inline buffer holding and owning an object while concealing the * concrete type. Access to the contained object is similar to a * smart-pointer, but the object isn't heap allocated. OpaqueHolder * may be created empty, which can be checked by a bool test. * The whole compound is copyable if and only if the contained * object is copyable. * * \par using OpaqueHolder * OpaqueHolder instances are copyable value objects. They are created * either empty, by copy from an existing OpaqueHolder, or by directly * specifying the concrete object to embed. This target object will be * \em copy-constructed into the internal buffer. Additionally, you * may assign a new value, which causes the old value object to be * destroyed and a new one to be copy-constructed into the buffer. * Later on, the embedded value might be accessed * - using the smart-ptr-like access through the common base interface BA * - when knowing the exact type to access, the templated #get might be an option * - the empty state of the container and a \c isValid() on the target may be checked * - a combination of both is available as a \c bool check on the OpaqueHolder instance. * * For using OpaqueHolder, several \b assumptions need to be fulfilled * - any instance placed into OpaqueHolder is below the specified maximum size * - the caller cares for thread safety. No concurrent get calls while in mutation! */ template < class BA ///< the nominal Base/Interface class for a family of types , size_t siz = sizeof(BA) ///< maximum storage required for the targets to be held inline > class OpaqueHolder : public InPlaceAnyHolder > { typedef InPlaceAnyHolder > InPlaceHolder; public: OpaqueHolder() : InPlaceHolder() {} template OpaqueHolder(SUB const& obj) : InPlaceHolder(obj) {} template OpaqueHolder& operator= (SUB const& newContent) { static_cast(*this) = newContent; return *this; } // note: using standard copy operations /* === smart-ptr style access === */ BA& operator* () const { ASSERT (!InPlaceHolder::empty()); return *InPlaceHolder::buff().getBase(); } BA* operator-> () const { ASSERT (!InPlaceHolder::empty()); return InPlaceHolder::buff().getBase(); } }; /** * Variation of the concept realised by OpaqueHolder, but implemented here * with reduced security and lesser overhead. InPlaceBuffer is just a chunk of * storage, which can be accessed through a common base class interface and * allows to place new objects there. It has no way to keep track of the * actual object living currently in the buffer. Thus, using InPlaceBuffer * requires the placed class(es) themselves to maintain their lifecycle, * and especially it is mandatory for the base class to provide a * virtual dtor. On the other hand, just the (alignment rounded) * storage for the object(s) placed into the buffer is required. */ template < class BA ///< the nominal Base/Interface class for a family of types , size_t siz = sizeof(BA) ///< maximum storage required for the targets to be held inline , class DEFAULT = BA ///< the default instance to place initially > class InPlaceBuffer : boost::noncopyable { mutable char buf_[siz]; BA& getObj() const { return reinterpret_cast (buf_); } void placeDefault() { new(&buf_) DEFAULT(); } void destroy() { getObj().~BA(); } public: InPlaceBuffer () { placeDefault(); } ~InPlaceBuffer () { destroy(); } /** Abbreviation for placement new */ #define LIB_InPlaceBuffer_CTOR(_CTOR_CALL_) \ destroy(); \ try \ { \ REQUIRE (siz >= sizeof(TY)); \ return *new(&buf_) _CTOR_CALL_; \ } \ catch (...) \ { \ placeDefault(); \ throw; \ } template TY& create () { LIB_InPlaceBuffer_CTOR ( TY() ) } template TY& //___________________________________________ create (A1& a1) ///< place object of type TY, using 1-arg ctor { LIB_InPlaceBuffer_CTOR ( TY(a1) ) } template< class TY , typename A1 , typename A2 > TY& //___________________________________________ create (A1& a1, A2& a2) ///< place object of type TY, using 2-arg ctor { LIB_InPlaceBuffer_CTOR ( TY(a1,a2) ) } template< class TY , typename A1 , typename A2 , typename A3 > TY& //___________________________________________ create (A1& a1, A2& a2, A3& a3) ///< place object of type TY, using 3-arg ctor { LIB_InPlaceBuffer_CTOR ( TY(a1,a2,a3) ) } template< class TY , typename A1 , typename A2 , typename A3 , typename A4 > TY& //___________________________________________ create (A1& a1, A2& a2, A3& a3, A4& a4) ///< place object of type TY, using 4-arg ctor { LIB_InPlaceBuffer_CTOR ( TY(a1,a2,a3,a4) ) } template< class TY , typename A1 , typename A2 , typename A3 , typename A4 , typename A5 > TY& //___________________________________________ create (A1& a1, A2& a2, A3& a3, A4& a4, A5& a5) ///< place object of type TY, using 5-arg ctor { LIB_InPlaceBuffer_CTOR ( TY(a1,a2,a3,a4,a5) ) } /* === smart-ptr style access === */ BA& operator* () const { return getObj(); } BA* operator-> () const { return &getObj(); } template static SUB* access () { BA * asBase = &getObj(); SUB* content = util::AccessCasted::access (asBase); return content; } // NOTE: might be null. }; } // namespace lib #endif