Optimization bug in g++ 3.4?

[Thorsten Raasch]

Hi,

the following C++ program seems to yield a segmentation fault
in g++ 3.4 when compiled with any of the optimization options -O*,
e.g. with "g++ -O2 bug.cpp -o bug".

class TestClass {
public:
TestClass() {
c[0] = 1;
c[1] = 2;
}

int f(const unsigned int i) { // segfault during optimization
// int f(unsigned int i) { // works fine
return c[(i==1) ? 1 : 0];
}

int c[2];
};

int main()
{
TestClass c;
int x = c.f(1);

return 0;
}

I tested the code with g++ 3.4 under Debian, CentOS and SunOS, resulting
in a segmentation fault in each case. Can anyone confirm the bug?
First, the point seems to be that the argument of the method f is a
"const unsigned int". Changing the argument either into a "const int" or
an "unsigned int" indeed yields a compileable code.
Second point: if the return value is just (i==1) ? 1 : 0, the code also
compiles fine.

Best,
Thorsten Raasch

[Victor Bazarov]

Thorsten Raasch wrote:

the following C++ program seems to yield a segmentation fault
in g++ 3.4 when compiled with any of the optimization options -O*,
e.g. with "g++ -O2 bug.cpp -o bug".
[...]
I tested the code with g++ 3.4 under Debian, CentOS and SunOS, resulting
in a segmentation fault in each case. Can anyone confirm the bug?
Please post to one of gnu.g++.* newsgroups. Specific compiler behaviour
is off-topic here, especially if there is a newsgroup dedicated to that
compiler.
[...]

[Thorsten Raasch]

Victor Bazarov wrote:

(...)
Please post to one of gnu.g++.* newsgroups. Specific compiler behaviour
is off-topic here, especially if there is a newsgroup dedicated to that
compiler.

Sorry for being OT. Especially the gnu.g++.bug newsgroup is full of
spam, so I hesitated posting there. I did try now, hopefully someone can
help me there.

[John Ratliff]

Thorsten Raasch wrote:

Hi,

the following C++ program seems to yield a segmentation fault
in g++ 3.4 when compiled with any of the optimization options -O*,
e.g. with "g++ -O2 bug.cpp -o bug".

class TestClass {
public:
TestClass() {
c[0] = 1;
c[1] = 2;
}

int f(const unsigned int i) { // segfault during optimization
// int f(unsigned int i) { // works fine
return c[(i==1) ? 1 : 0];
}

int c[2];
};

int main()
{
TestClass c;
int x = c.f(1);

return 0;
}

I tested the code with g++ 3.4 under Debian, CentOS and SunOS, resulting
in a segmentation fault in each case. Can anyone confirm the bug?
First, the point seems to be that the argument of the method f is a
"const unsigned int". Changing the argument either into a "const int" or
an "unsigned int" indeed yields a compileable code.
Second point: if the return value is just (i==1) ? 1 : 0, the code also
compiles fine.

Best,
Thorsten Raasch

I get no segfault with mingw/g++ 3.4.2 under Windows XP/msys.

I would post to the gcc-help mailing list if you're getting problems
with g++.

--John Ratliff

[Branimir Maksimovic]

"John Ratliff" <us**@example.n et> wrote in message
news:GdS5f.4502 21$_o.205227@at tbi_s71...

Thorsten Raasch wrote:

Hi,

the following C++ program seems to yield a segmentation fault
in g++ 3.4 when compiled with any of the optimization options -O*,
e.g. with "g++ -O2 bug.cpp -o bug".

class TestClass {
public:
TestClass() {
c[0] = 1;
c[1] = 2;
}

int f(const unsigned int i) { // segfault during optimization
// int f(unsigned int i) { // works fine
return c[(i==1) ? 1 : 0];
}

int c[2];
};

int main()
{
TestClass c;
int x = c.f(1);

return 0;
}

I get no segfault with mingw/g++ 3.4.2 under Windows XP/msys.

You didn't get segfault because compiler just silently fails and
does not produce object file.

Greetings, Bane.

[John Ratliff]

Branimir Maksimovic wrote:

"John Ratliff" <us**@example.n et> wrote in message
news:GdS5f.4502 21$_o.205227@at tbi_s71...

Thorsten Raasch wrote:

Hi,

the following C++ program seems to yield a segmentation fault
in g++ 3.4 when compiled with any of the optimization options -O*,
e.g. with "g++ -O2 bug.cpp -o bug".

class TestClass {
public:
TestClass() {
c[0] = 1;
c[1] = 2;
}

int f(const unsigned int i) { // segfault during optimization
// int f(unsigned int i) { // works fine
return c[(i==1) ? 1 : 0];
}

int c[2];
};

int main()
{
TestClass c;
int x = c.f(1);

return 0;
}

I get no segfault with mingw/g++ 3.4.2 under Windows XP/msys.

You didn't get segfault because compiler just silently fails and
does not produce object file.

Greetings, Bane.

You're right. I didn't notice that. I compiled one without optimization
first, and that's what I was running later with optimization turned on.

--John Ratliff

[Thorsten Raasch]

John Ratliff wrote:

Branimir Maksimovic wrote:

"John Ratliff" <us**@example.n et> wrote in message
news:GdS5f.4502 21$_o.205227@at tbi_s71...
(...)

I get no segfault with mingw/g++ 3.4.2 under Windows XP/msys.

You didn't get segfault because compiler just silently fails and
does not produce object file.

Greetings, Bane.

You're right. I didn't notice that. I compiled one without optimization
first, and that's what I was running later with optimization turned on.

--John Ratliff

So can you both confirm the segfault? The exact compiler versions I
tested the code with were 3.4.3 (under CentOS) and 3.4.4 (under SunOS).
Under g++ 4.0.2, the code compiles fine.
I'm currently reporting the bug to bugzilla.

By the way, does anyone know whether the gnu.g++.bug and gnu.g++.help
newsgroups are working at all? I posted my problem there yesterday, but
the posts did not arrive so far.

Best,
Thorsten Raasch