https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116909
Bug ID: 116909
Summary: Does g++ define the behavior for an "array" manually
created with a linker script?
Product: gcc
Version: 15.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: federico at kircheis dot it
Target Milestone: ---
Consider following program
main.cpp
----
using test_signature = void();
void test2(){std::puts("test2");}
[[gnu::used]] constexpr test_signature* const t1 [[gnu::section(".tests")]] =
+[]{std::puts("test1");};
[[gnu::used]] constexpr test_signature* const t2 [[gnu::section(".tests")]] =
test2;
[[gnu::used]] constexpr test_signature* const t3 [[gnu::section(".tests")]] =
test2;
// eventually many other
auto get_tests() noexcept {
extern const test_signature* const tests_begin[];
extern const test_signature* const tests_end[];
return std::span<test_signature* const>(tests_begin, tests_end);
}
int main(){
auto funcs = get_tests();
for(const auto& v : funcs){
v();
}
}
----
linkerscript.ld
----
SECTIONS
{
.tests : {
PROVIDE(tests_begin = .);
KEEP(*(.tests))
PROVIDE(tests_end = .);
}
}
INSERT AFTER .text;
----
And compile it with "g++ --std=c++20 -Wl,-Tlinkerscript.ld main.cpp"
The code works as expected (it calls the "registered functions" and prints
whatever is passed to std::puts), at least on linux, tested with different
optimizations flags (even with -flto and -fwhole-program), memory sanitizer and
valgrind.
I believe that the code triggers UB, because there is no array containing all
functions pointers, thus one is not allowed to do pointer arithmetic.
Since tests_begin and tests_end are created outside of C++, I guess there is
little for the compiler to do except assume that there is an array, and thus I
would like to claim that this code is portable between versions of gcc/ldd and
platforms (please correct me if I am wrong. any suggestion on how to improve it
or make more robust are welcome).
Assuming my linker script is correct, does g++ specify the behavior of the
program in this case?
If not, does it make sense to define it?
As the language does know nothing about "things" created outside of c++, it
cannot specify the behavior, but the toolchain should be able to do so.
I made the question for c++ with a c++ snippet, but the same question holds for
C too.