No -Wempty-body warning for empty body with a prefix label

[chengniansun]

Bugzilla Link	18877
Version	trunk
OS	Windows NT
CC	@zygoloid,@zhendongsu

Extended Description

For the following statement：
if (i) a:;
Clang does not warn that the body is empty.

$: cat s.c
void f(int i) {
if (i) a:;
}
$: clang-trunk -Wempty-body -c s.c
$: gcc-trunk -Wempty-body -c s.c
s.c: In function ‘f’:
s.c:2:12: warning: suggest braces around empty body in an ‘if’ statement [-Wempty-body]
if (i) a:;
^
$: clang-trunk --version
clang version 3.5 (trunk 201498)
Target: x86_64-unknown-linux-gnu
Thread model: posix
$

[chengniansun]

assigned to @Weverything

[zygoloid]

Right. This warning is to detect the 'stray semicolon' bug:

if (foo);
blah;

The labeled semicolon case doesn't match this pattern.

[chengniansun]

Right. This warning is to detect the 'stray semicolon' bug:

if (foo);
blah;

The labelled semicolon case doesn't match this pattern.

Hi Richard,

Thanks. I understand why it is not warned, but I do not understand why the given test case with an empty body does not fit into -Wempty-body.

The pattern makes sense, but is it a little conservative? For the if statement "if (i) a:;", the programmer cannot easily find out that it is an empty statement due to the labelled semicolon, which is equally bad to the stray semicolon.

[zygoloid]

Why do you think this will detect bugs?

-Wempty-body is not intended to warn in any case where the 'if' doesn't contain any statements; it's only intended to catch the case where a semicolon was accidentally added after the 'if'. For instance, we also (deliberately) don't warn on these:

if (x)
;

if (x) {}

[chengniansun]

Why do you think this will detect bugs?

-Wempty-body is not intended to warn in any case where the 'if' doesn't
contain any statements; it's only intended to catch the case where a
semicolon was accidentally added after the 'if'. For instance, we also
(deliberately) don't warn on these:

if (x)
;

if (x) {}

As there is no documentation on -Wempty-body, I assumed -Wempty-body is the same as that of Gcc. However, if "it's only intended to catch the case where a semicolon was accidentally added after the 'if'", then I would say the reported case is not a bug as clang does what it intends to do.

My point here is still that the labelled semicolon makes the if statement look like having a non-empty body, but in fact it does not as the label is not a statement.

I never said that once the body was empty clang should warn on it. Of the two cases you mentioned,

1. the first case is the way clang suppresses the -Wempty-body warning
2. the second case is the way gcc suppresses the -Wempty-body warning.
   which is clearly indicated in their warning messages.

[chengniansun]

Clang only handles "if(exp);", but not "else;".

$: cat s.1.c
int f(int a) {
int i = 0;
if (a) {
i++;
} else;
i *= 2;
return i;
}
$: clang-trunk -Wempty-body -c s.1.c
$:
$: gcc-trunk -Wempty-body -c s.1.c
s.1.c: In function ‘f’:
s.1.c:5:9: warning: suggest braces around empty body in an ‘else’ statement [-Wempty-body]
} else;
^
$:

[zygoloid]

Clang only handles "if(exp);", but not "else;".

I'm pretty sure that's a bug. Richard, since you touched this last, do you agree?

[Weverything]

Clang only handles "if(exp);", but not "else;".

I'm pretty sure that's a bug. Richard, since you touched this last, do you
agree?

Yeah, I think it's pretty safe to extend -Wempty-body to the else case as well.