Invalid "goto into protected scope" error

[llvmbot]

Bugzilla Link	10137
Resolution	INVALID
Resolved on	Jun 15, 2011 10:02
Version	trunk
OS	Linux
Reporter	LLVM Bugzilla Contributor
CC	@DougGregor,@efriedma-quic,@ehsan,@tkremenek

Extended Description

A small variation from the test is bug 10034 still fails:

void puts(const char*);

struct Y {};
struct X : Y {};

void exx(X) {}

int main(int argc, char **argv)
{
if (argc > 3)
goto end;

X x;
exx(x);

end:

for (int i = 0; i < argc; ++i)
{
puts(argv[i]);
}

return 0;
}

Are we OK with extending clang to accept it?

[tkremenek]

Why should we accept this? Because X is POD?

[llvmbot]

I don't know what the current policy is.
I just found that we decided to accept the case in bug 10034 and this one is a small variation of it. This is also accepted by gcc, so maybe we want to do the same.

[tkremenek]

I don't know what the current policy is.
I just found that we decided to accept the case in bug 10034 and this one is a
small variation of it. This is also accepted by gcc, so maybe we want to do the
same.

Being ideally compatible with GCC is not a good enough reason to do something that doesn't make sense. We should understand why we want to accept this.

[efriedma-quic]

Note that we do accept the given construct in C++0x mode; it might make sense to downgrade the error to a pedwarn for cases which are valid in C++0x, but invalid in C++03.

[llvmbot]

Being ideally compatible with GCC is not a good enough reason to do something
that doesn't make sense. We should understand why we want to accept this.

Having worked on gcc I couldn't agree more. I really don't know or have an opinion on if we should support this or not. It is just that there is no rationale in bug bug 10034, so I looked a good idea to check if the we could support this for the same reason we support that.

[DougGregor]

With bug 10034, the type was a POD type, so the code was well-formed according to the C++ standard and we have to accept it.

With this bug, the type is non-POD, so the code is ill-formed in C++98/03. Since GCC also rejects it, I see no reason to accept this ill-formed code.

In C++0x, the rules changed and this code because well-formed.

[llvmbot]

Just a correction, gcc (all versions up to 4.7) accept this code with -Wall -pedantic.

[DougGregor]

Apple GCC 4.2 correctly rejects this code:

Cove:clang dgregor$ g++ t3.cpp -Wall -pedantic
t3.cpp: In function ‘int main(int, char**)’:
t3.cpp:16: error: jump to label ‘end’
t3.cpp:11: error: from here
t3.cpp:13: error: enters scope of non-POD ‘X x’

I don't know when GCC started accepting it as an extension, but C++03 [stmt.decl]p3 is very clear that it's ill-formed.