C99: variable declarations inside switch statements

N

Neil Zanella

Hello,

Unlike in pre-C99 versions of C where variables can only be defined at the
beginning of blocks, C99 allows variables to be defined in arbitrary
places inside blocks. However, gcc 3.2.2 seems to reveal that there
are still places where this is illegal. For instance, consider the
following code snippet.

int main(void) {
int i = 0;
switch (i) {
case 0:
char ch;
break;
}
}

Here the compiler complains with the error:

hello.c: In function `main':
hello.c:6: parse error before "char"

It seems like declaring a variable inside a switch statement is illegal.
Anyone know why this is the case. After all, the following compiles fine
with gcc 3.2.2:

int main(void) {
int i = 0;
switch (i) {
case 0: {
char ch;
break;
}
}
}

The only difference is the introduction of the block within the C
statement. Anyone know of similar cases where it is illegal to declare
variable in places where we would expect it to be legal in C99? Also,
can anyone explain the reason for the compiler complaint?

Thanks,

Neil
 
S

Sheldon Simms

Unlike in pre-C99 versions of C where variables can only be defined at the
beginning of blocks, C99 allows variables to be defined in arbitrary
places inside blocks. However, gcc 3.2.2 seems to reveal that there
are still places where this is illegal. For instance, consider the
following code snippet.

You can't count on conformance of current versions of GCC
to the C99 standard. With that out of the way I'll continue.
int main(void) {
int i = 0;
switch (i) {
case 0:
char ch;
break;
}
}

Here the compiler complains with the error:

hello.c: In function `main':
hello.c:6: parse error before "char"

can anyone explain the reason for the compiler complaint?

The answer is in the C99 grammar. Declarations can appear at
any place within a block because of these productions:

A.2.3 Statements

(6.8.2) compound-statement: { block-item-list? }
(6.8.2) block-item-list: block-item
block-item-list block-item
(6.8.2) block-item: declaration
statement

But you can't put a declaration after a case construct because
of these productions:

A.2.3 Statements

(6.8) statement: labeled-statement
(6.8.1) labeled-statement: case constant-expression : statement

IOW, only a statement can follow a case.

-Sheldon
 
C

Christian Bau

Neil Zanella said:
Hello,

Unlike in pre-C99 versions of C where variables can only be defined at the
beginning of blocks, C99 allows variables to be defined in arbitrary
places inside blocks. However, gcc 3.2.2 seems to reveal that there
are still places where this is illegal. For instance, consider the
following code snippet.

int main(void) {
int i = 0;
switch (i) {
case 0:
char ch;
break;
}
}

Here the compiler complains with the error:

hello.c: In function `main':
hello.c:6: parse error before "char"

It seems like declaring a variable inside a switch statement is illegal.
Anyone know why this is the case. After all, the following compiles fine
with gcc 3.2.2:

No, declaring a variable inside a switch statement is not illegal.
However, the syntax is slightly different then you think it is: When you
write

switch (i) {
<body of switch statement>
}

the body of the switch statement can contain the following in any order:

Unlabeled statements
Labeled statements
Declarations

case 0: char ch;

would be a labeled declaration, and they are not allowed. Write

case 0: ;
char ch;
break;

instead.
 
N

Neil Zanella

Hello,

Quite interesting. I wonder whether adding a rule to the C99 grammar for
accepting declarations following labeled statements would cause the
resulting grammar to be ambiguous (I doubt it), or whether there
would be some shift/reduce, reduce/shift, reduce/reduce conflicts
of some sort (more likely). How would such conflicts arise in
such a situation?

Thanks,

Neil
 
S

Sheldon Simms

top-posting corrected

Quite interesting. I wonder whether adding a rule to the C99 grammar for
accepting declarations following labeled statements would cause the
resulting grammar to be ambiguous (I doubt it), or whether there
would be some shift/reduce, reduce/shift, reduce/reduce conflicts
of some sort (more likely). How would such conflicts arise in
such a situation?

I don't know if the C standards committee worries too much about
those kinds of problems when writing the grammar. I wouldn't in any
case. The grammer in the standard isn't supposed to be directly
usable in bison.

FWIW C++ *does* allow declarations after case and other labels with
these productions:

ISO/IEC 14882:1998(E)
A.5
statement: labeled-statement
declaration-statement

labeled-statement: case constant-expression : statement

declaration-statement : block-declaration

where block-declaration is the kind of declaration that can occur
inside a block (i.e. - no function definitions, template declaration,
etc.)

-Sheldon
 
C

Christian Bau

Hello,

Quite interesting. I wonder whether adding a rule to the C99 grammar for
accepting declarations following labeled statements would cause the
resulting grammar to be ambiguous (I doubt it), or whether there
would be some shift/reduce, reduce/shift, reduce/reduce conflicts
of some sort (more likely). How would such conflicts arise in
such a situation?

Unlikely that this could introduce any conflicts.

We assume that there is no conflict between unlabeled statement and
declaration (otherwise C has a big problem anyway). Labels are quite
trivial to recognise and to distinguish from anything else. Once the
label is recognised, you are left again with the task of distinguishing
between unlabeled statement and declaration.
 

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments. After that, you can post your question and our members will help you out.

Ask a Question

Members online

No members online now.

Forum statistics

Threads
474,091
Messages
2,570,605
Members
47,225
Latest member
DarrinWhit

Latest Threads

Top