https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84322
Bug ID: 84322
Summary: -fstrong-eval-order violated on assignment with
function call
Product: gcc
Version: 7.3.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
Assignee: unassigned at gcc dot gnu.org
Reporter: chortos at inbox dot lv
Target Milestone: ---
The following C++ code:
extern int a[10];
int g();
void f() {
a[a[1]] = g();
}
generates the following assembler code on amd64 Darwin with -O3 -std=c++17
(with or without an explicit -fstrong-eval-order, which should be implied if I
understand correctly):
movq _a@GOTPCREL(%rip), %rbp
movslq 4(%rbp), %rbx
call __Z1gv
movl %eax, 0(%rbp,%rbx,4)
and the following with -O0 -std=c++17:
movq _a@GOTPCREL(%rip), %rax
movl 4(%rax), %ebx
call __Z1gv
movl %eax, %ecx
movq _a@GOTPCREL(%rip), %rax
movslq %ebx, %rdx
movl %ecx, (%rax,%rdx,4)
In both cases, the a[1] is evaluated before the g() call, which contradicts the
strong right-to-left order of assignment operands guaranteed/required by C++17.
If g() modifies a[1], the program can observe the difference:
int a[10];
int g() {
return a[1] = 2;
}
void f();
int main() {
f();
__builtin_printf("%d %d\n", a[1], a[2]);
}
produces "2 0" when compiled with GCC 7 but "2 2" when compiled with Clang 5,
which (intentionally or not) evaluates the operands in the correct order.
Here are my gcc -v and uname -a, which I doubt will be useful but are requested
on https://gcc.gnu.org/bugs/:
$ g++-7 -v
Using built-in specs.
COLLECT_GCC=g++-7
COLLECT_LTO_WRAPPER=/usr/local/Cellar/gcc/7.3.0/libexec/gcc/x86_64-apple-darwin10.8.0/7.3.0/lto-wrapper
Target: x86_64-apple-darwin10.8.0
Configured with: ../configure --build=x86_64-apple-darwin10.8.0
--prefix=/usr/local/Cellar/gcc/7.3.0
--libdir=/usr/local/Cellar/gcc/7.3.0/lib/gcc/7
--enable-languages=c,c++,objc,obj-c++,fortran --program-suffix=-7
--with-gmp=/usr/local/opt/gmp --with-mpfr=/usr/local/opt/mpfr
--with-mpc=/usr/local/opt/libmpc --with-isl=/usr/local/opt/isl
--with-system-zlib --enable-checking=release --with-pkgversion='Homebrew GCC
7.3.0' --with-bugurl=https://github.com/Homebrew/homebrew-core/issues
--with-dwarf2 --disable-nls
Thread model: posix
gcc version 7.3.0 (Homebrew GCC 7.3.0)
$ uname -a
Darwin [hostname].local 10.8.0 Darwin Kernel Version 10.8.0: Tue Jun 7
16:33:36 PDT 2011; root:xnu-1504.15.3~1/RELEASE_I386 i386
For reference, I have also tried GCC 6.3.0 and 5.1.0 (which do not understand
-fstrong-eval-order but do understand -std=c++17), and they use the same order
as GCC 7.3.0.
