This is a question about built-in variables and their uniqueness in several translation units. Consider the following code that separates a string literal in two translation units:
Hdr.h file:
inline const char* const str = "hello world";
File tu0.cpp:
#include <cstdio> #include "hdr.h" void foo() { printf("tu0: &str=%p, str=%p\n", &str, str); }
File tu1.cpp:
#include <cstdio> #include "hdr.h" extern void foo(); int main() { foo(); printf("tu1: &str=%p str=%p\n", &str, str); }
For completeness, here is how I compile them (on linux, 64-bit, binutils 2.28):
$ CXX -c -std=c++1z tu0.cpp -o tu0.o $ CXX -c -std=c++1z tu1.cpp -o tu1.o $ CXX -std=c++1z tu*.o -o a.out
With CXX = clang-4.0, both str and & str are identical:
tu0: &str=0x400608 str=0x4005f9 tu1: &str=0x400608 str=0x4005f9
But with g ++ versions 7.1 and 7.2, I get an unexpected result:
tu0: &str=0x4005d8 str=0x4005b4 tu1: &str=0x4005d8 str=0x4005e0
Ok, I really have a few questions (but they are all related):
Should str and & str have the same value in all translation units?
N4687 [dcl.inline] / 6 says that the built-in variable must have the same definition in each translation unit. Does "definition" refer to the source hello world token or to the value of the string literal pointer "hello world"?
How does GCC manage to get the same address for str, but different values depending on where this variable is accessed? And this leads to the last question:
Does the built-in variables take longer to execute (on every access) than the usual extern declarations approach with manual definition in one translation unit?
source share