我正在检测一些代码,并注意到有了C++ 14的功能,有两个新的delete运算符(来自http://en.cppreference.com/w/cpp/memory/new/operator_delete):
这些是5-6)如果提供了用户定义的替换,则调用而不是(1-2),除了它的实现定义在删除不完整类型和数组的对象时是否调用(1-2)或(5-6)非类和易破坏的类类型(自C++ 17以来).标准库实现与(1-2)相同.
我已经超载了这些并且想要专门调用这两个.当我用gcc重载这两个时,我没有问题.使用clang ++我得到一个未定义的引用operator delete(void*)
这是代码
void* operator new(long unsigned int howMuch) {
return reinterpret_cast<void*>(0xdeadbeef);
}
void* operator new[](long unsigned int howMuch) {
return reinterpret_cast<void*>(0xdeadbeef);
}
void operator delete(void* what, long unsigned int howmuch) {
if(what != reinterpret_cast<void*>(0xdeadbeef)) __builtin_trap();
if(howmuch != 1) __builtin_trap();
}
extern "C"
void _start() {
delete new char;
asm("syscall" : : "a"(60) : );
}
用gcc编译:g++ -ggdb -std=c++14 -nostdlib -fno-builtin -fno-exceptions 1.cc没有问题,运行正常.
用llvm/clang可以做到这一点吗？
您可以显式调用大小或非大小删除运算符，如下所示：
\n\nchar* ptr = new char;\ndelete ptr; // compiler selects which to call\noperator delete(ptr); // explicitly call the non-sized delete \noperator delete(ptr, 1); // explicitly call sized delete\n完整的例子：
\n\nvoid* operator new(long unsigned int howMuch) { \n return reinterpret_cast<void*>(0xdeadbeef);\n}\n\nvoid* operator new[](long unsigned int howMuch) {\n return reinterpret_cast<void*>(0xdeadbeef);\n}\n\nvoid operator delete(void* what) {\n if(what != reinterpret_cast<void*>(0xdeadbeef)) __builtin_trap();\n}\n\nvoid operator delete(void* what, long unsigned int howmuch) {\n if(what != reinterpret_cast<void*>(0xdeadbeef)) __builtin_trap();\n if(howmuch != 1) __builtin_trap();\n}\n\nextern "C"\nvoid _start() {\n char* ptr = new char;\n delete ptr;\n operator delete(ptr);\n operator delete(ptr, 1);\n asm("syscall" : : "a"(60) : ); \n}\n编译并查看生成的代码，可以清楚地知道在以下情况下调用哪些运算符：
\n\n$ clang++ -std=c++14 -nostdlib -fno-builtin -fno-exceptions -fsized-deallocation sized-deallocation.cpp -o sized-deallocation.bin && gdb -batch -ex \'file sized-deallocation.bin\' -ex \'disassemble _start\' | c++filt\nDump of assembler code for function _start:\n 0x0000000000401070 <+0>: push %rbp\n 0x0000000000401071 <+1>: mov %rsp,%rbp\n 0x0000000000401074 <+4>: sub $0x10,%rsp\n 0x0000000000401078 <+8>: mov $0x1,%eax\n 0x000000000040107d <+13>: mov %eax,%edi\n 0x000000000040107f <+15>: callq 0x401000 <operator new(unsigned long)>\n 0x0000000000401084 <+20>: mov %rax,-0x8(%rbp)\n 0x0000000000401088 <+24>: mov -0x8(%rbp),%rax\n 0x000000000040108c <+28>: cmp $0x0,%rax\n 0x0000000000401090 <+32>: mov %rax,-0x10(%rbp)\n 0x0000000000401094 <+36>: je 0x4010aa <_start+58>\n 0x000000000040109a <+42>: mov $0x1,%eax\n 0x000000000040109f <+47>: mov %eax,%esi\n 0x00000000004010a1 <+49>: mov -0x10(%rbp),%rdi\n 0x00000000004010a5 <+53>: callq 0x401040 <operator delete(void*, unsigned long)>\n 0x00000000004010aa <+58>: mov -0x8(%rbp),%rdi\n 0x00000000004010ae <+62>: callq 0x401020 <operator delete(void*)>\n 0x00000000004010b3 <+67>: mov $0x1,%eax\n 0x00000000004010b8 <+72>: mov %eax,%esi\n 0x00000000004010ba <+74>: mov -0x8(%rbp),%rdi\n 0x00000000004010be <+78>: callq 0x401040 <operator delete(void*, unsigned long)>\n 0x00000000004010c3 <+83>: mov $0x3c,%eax\n 0x00000000004010c8 <+88>: syscall \n 0x00000000004010ca <+90>: add $0x10,%rsp\n 0x00000000004010ce <+94>: pop %rbp\n 0x00000000004010cf <+95>: retq \nEnd of assembler dump.\n您使用 Clang 的实际原因undefined reference to `operator delete(void*)\'是（正如@TC所说）Clang 需要该-fsized-deallocation标志来启用C++14 大小的释放。
如果使用以下命令，您的示例编译不会出现错误：
\n\n$ clang++ -std=c++14 -nostdlib -fno-builtin -fno-exceptions -fsized-deallocation sized-deallocation.cpp -o sized-deallocation.bin && gdb -batch -ex \'file sized-deallocation.bin\' -ex \'disassemble _start\' | c++filt\nDump of assembler code for function _start:\n 0x0000000000401070 <+0>: push %rbp\n 0x0000000000401071 <+1>: mov %rsp,%rbp\n 0x0000000000401074 <+4>: sub $0x10,%rsp\n 0x0000000000401078 <+8>: mov $0x1,%eax\n 0x000000000040107d <+13>: mov %eax,%edi\n 0x000000000040107f <+15>: callq 0x401000 <operator new(unsigned long)>\n 0x0000000000401084 <+20>: mov %rax,-0x8(%rbp)\n 0x0000000000401088 <+24>: mov -0x8(%rbp),%rax\n 0x000000000040108c <+28>: cmp $0x0,%rax\n 0x0000000000401090 <+32>: mov %rax,-0x10(%rbp)\n 0x0000000000401094 <+36>: je 0x4010aa <_start+58>\n 0x000000000040109a <+42>: mov $0x1,%eax\n 0x000000000040109f <+47>: mov %eax,%esi\n 0x00000000004010a1 <+49>: mov -0x10(%rbp),%rdi\n 0x00000000004010a5 <+53>: callq 0x401040 <operator delete(void*, unsigned long)>\n 0x00000000004010aa <+58>: mov -0x8(%rbp),%rdi\n 0x00000000004010ae <+62>: callq 0x401020 <operator delete(void*)>\n 0x00000000004010b3 <+67>: mov $0x1,%eax\n 0x00000000004010b8 <+72>: mov %eax,%esi\n 0x00000000004010ba <+74>: mov -0x8(%rbp),%rdi\n 0x00000000004010be <+78>: callq 0x401040 <operator delete(void*, unsigned long)>\n 0x00000000004010c3 <+83>: mov $0x3c,%eax\n 0x00000000004010c8 <+88>: syscall \n 0x00000000004010ca <+90>: add $0x10,%rsp\n 0x00000000004010ce <+94>: pop %rbp\n 0x00000000004010cf <+95>: retq \nEnd of assembler dump.\n由于 Clang 3.7 C++14 大小的释放默认被禁用：
\n\nClang 中的 C++ 支持 > C++14 实现状态 > C++ 大小的释放 N3778
\n\n\n\n\n(7)：在 Clang 3.7 及更高版本中，仅当用户传递标志时才启用大小释放
\n-fsized-deallocation。用户必须通过显式提供或使用 C++ 标准库来提供大小释放函数的定义。libstdc++在5.0版本中添加了这些功能，libc++在3.7版本中添加了这些功能。
Clang 3.7 发行说明 > Clang 3.7 中有哪些新功能？> 新的编译器标志
\n\n\n\n\nC++14 的大小释放功能现在由标志控制
\n-fsized-deallocation。此功能依赖于尚未广泛部署的库支持，因此用户必须提供额外的标志才能获得额外的功能。
进行此更改的原因是当时广泛部署的标准库中缺少这些函数（2015-03-19）：
\n\n\n\n\n\n\n没有广泛部署的标准库提供大小释放函数，因此我们必须询问用户是否希望我们使用大小释放函数。将来，当部署此类库时，我们可以教驱动程序检测它们并启用此功能。
\n
可以在此处找到此选项的手动输入：
\n\nClang 命令行参数参考 > 编译标志 > 与目标无关的编译选项
\n\n\n\n\n\n
-fsized-deallocation,-fno-sized-deallocation启用 C++14 大小的全局释放函数
\n