const and 'constant expression'

D

devdatta_clc

Hi C experts

I've a bunch of questions.

Consider this simplified piece of code.

const int a = 10;

int main () {
static int b = a;
return b;
}

~>gcc test.c
test.c: In function `main':
test.c:4: error: initializer element is not constant
~>

Question1
I understand that const is 'read-only' but I could not convince my
co-worker that its value can change.

6.7.8 Initialization

[...]

[#4] All the expressions in an initializer for an object
that has static storage duration shall be constant
expressions or string literals.

Can someone point me to text that says that 'const' is not 'constant
expression' and can in fact change?

He claims that the compiler may choose to substitute the value at
compile time.
I tried to explain how the value can change (but I had to use pointers
to const int) which he shot down saying the comparison of examples is
not valid (it's not initializer and it involves pointers).

Question2
Moreover, if we turn on optimization, (and assume that the real code
does not have any dead code like above), the compiler does not give any
error. Is the compiler correct after optimisation?
I think it is a bug in compiler but I need text in standard to prove
that.

Thanks in advance for your comments
-Dev
 
M

Marc Boyer

I've a bunch of questions.
Consider this simplified piece of code.

const int a = 10;
int main () {
static int b = a;
return b;
}

~>gcc test.c
test.c: In function `main':
test.c:4: error: initializer element is not constant
~>

Question1
I understand that const is 'read-only' but I could not convince my
co-worker that its value can change.

#include <stdlib.h>
#include <assert.h>
int main(int argc, char *argv []) {
assert( argc > 0 );
const int a= atoi(argv[1]);
static int b= a;
return b;
}

Marc Boyer
 
R

Richard Bos

const int a = 10;

int main () {
static int b = a;
return b;
}

~>gcc test.c
test.c: In function `main':
test.c:4: error: initializer element is not constant
~>

Question1
I understand that const is 'read-only' but I could not convince my
co-worker that its value can change.

Whether or not it can is of no import to that error. It is not a
constant, as defined by the Standard. That's all that matters.
6.7.8 Initialization

[...]

[#4] All the expressions in an initializer for an object
that has static storage duration shall be constant
expressions or string literals.

Can someone point me to text that says that 'const' is not 'constant
expression' and can in fact change?

Any object, whether const or not, is not a constant expression: 6.6.
"Can in fact change": immaterial.
He claims that the compiler may choose to substitute the value at
compile time.

He may claim whatever he wants. It remains a const object, not a
constant expression, and is therefore not a portable initaliser.
Moreover, if we turn on optimization, (and assume that the real code
does not have any dead code like above), the compiler does not give any
error. Is the compiler correct after optimisation?

By 6.6#10, yes. However, no other compiler (nor even the same compiler
the next time) is required to accept its extensions.

Richard
 
M

Marc Boyer

Le 19-04-2006 said:
Thanks Marc, point noted.

Now getting back to my example,

const int a = 10;

In this case, can value of 'a' change?

Not to my knowledge.
If so how?
If not, why can not variable 'a' be treated a compile time constant.

It 'can' in this case. But, just try to explain in wich case
it can and when it can not....

Marc Boyer
 
D

devdatta_clc

Thanks Marc, point noted.

Now getting back to my example,

const int a = 10;

In this case, can value of 'a' change? If so how?
If not, why can not variable 'a' be treated a compile time constant.

-Dev
 
D

devdatta_clc

If the compiler can treat this 'a' as a compile time constant then this
should also work

const int size = 5;
int array[size];

But seems like it does not compile even with optimizations.

This is causing confusion.

-Dev
 
K

Kenneth Brody

Marc said:
Not to my knowledge.

Isn't this legal?

==========

#include <stdio.h>

const int a = 10;

int main(int argc,char *argv[])
{
int *pt = (int *)&a;

printf("%d\n",a);

*pt = 20;

printf("%d\n",a);
}

==========

It fails here, at runtime, but only because "a" has been placed into
read-only memory on this platform. But it's perfectly "legal" C, as
far as I know. It compiles without even a warning about "pt" and "a",
even with maximum warnings turned on.

[...]

--
+-------------------------+--------------------+-----------------------------+
| Kenneth J. Brody | www.hvcomputer.com | |
| kenbrody/at\spamcop.net | www.fptech.com | #include <std_disclaimer.h> |
+-------------------------+--------------------+-----------------------------+
Don't e-mail me at: <mailto:[email protected]>
 
F

Flash Gordon

(e-mail address removed) wrote:

Please provide context when replying. Usenet is not Google Groups,
Google Groups is merely on of many Usenet service provides that provides
a poor interface. There is no guarantee that others have (or ever will)
see the post you are replying to.
If the compiler can treat this 'a' as a compile time constant then this
should also work

const int size = 5;
int array[size];

But seems like it does not compile even with optimizations.

This is causing confusion.

Basically, you can't do it because the standard says you can't. const in
C is a promise that you will not attempt to modify something, it does
not declare a real constant. So, for example, you can do:
volatile const int fred;
Then, you might be able to use some magic in your linker to map fred
over a hardware register and each time you read it you will get the
current contents of that register.

Note that C99 added variable length arrays, so in C99 you could do:
int main(void)
{
const int size = 5;
int array[size];
/* do stuff */
return 0;
}

However, in the above array is a variable length array where the length
is determined at run time, not a normal fixed length array.
 
F

Flash Gordon

Kenneth said:
Marc said:
Not to my knowledge.

Isn't this legal?

==========

#include <stdio.h>

const int a = 10;

int main(int argc,char *argv[])
{
int *pt = (int *)&a;

printf("%d\n",a);

*pt = 20;

printf("%d\n",a);
}

==========

It fails here, at runtime, but only because "a" has been placed into
read-only memory on this platform. But it's perfectly "legal" C, as
far as I know. It compiles without even a warning about "pt" and "a",
even with maximum warnings turned on.

It's legal C syntax and no diagnostic is required, but the attempt to
modify a const qualified object invokes undefined behaviour. The reason
for this is, I believe, specifically so that implementations *can* put
const qualified objects in to read only memory, so you implementation is
quite within its rights to do this. To quote from section 6.7.3 of
n1124.pdf:
| Semantics
| ...
| 5 If an attempt is made to modify an object defined with a
| const-qualified type through use of an lvalue with
| non-const-qualified type, the behavior is undefined...
 
D

Default User

Kenneth said:
Isn't this legal?

==========

#include <stdio.h>

const int a = 10;
int *pt = (int *)&a;
It fails here, at runtime, but only because "a" has been placed into
read-only memory on this platform. But it's perfectly "legal" C, as
far as I know. It compiles without even a warning about "pt" and "a",
even with maximum warnings turned on.

What do you mean by "legal"? It doesn't require a diagnostic, but it's
undefined behavior.



Brian
 
D

devdatta_clc

Marc said:
Not to my knowledge.


It 'can' in this case. But, just try to explain in wich case
it can and when it can not....


My bad. I did not include complete text.
Let me include enough context and repost.

~> cat test.c
const int a = 10;
int main () {
static int b = a;
return b;
}

If the compiler can treat this 'a' as a compile time constant then this
should also compile.

~> cat test2.c
const int size = 5;
int array[size];
int main() {
return 0;
}

But seems like it does not compile even with optimization turned on.
This is causing confusion. In one case, it compiles test.c with
optimisation.
In another case, it does not compile test2.c

Appreciate your patience,
-Dev
PS I'm aware of volatile const and VLA but that's a different topic.
I'm not using both here.
 
J

Jack Klein

Hi C experts

I've a bunch of questions.

Consider this simplified piece of code.

const int a = 10;

int main () {
static int b = a;
return b;
}

~>gcc test.c
test.c: In function `main':
test.c:4: error: initializer element is not constant
~>

Question1
I understand that const is 'read-only' but I could not convince my
co-worker that its value can change.

No strictly conforming program can change its value. Whether or not
it can change makes no difference.
6.7.8 Initialization

[...]

[#4] All the expressions in an initializer for an object
that has static storage duration shall be constant
expressions or string literals.

Can someone point me to text that says that 'const' is not 'constant
expression' and can in fact change?

Nothing in the standard states that a const object may change. But
there is a precise definition of what constitutes a constant
expression, in 6.6 Constant expressions. Nowhere in that section does
it include the value of an object, regardless of const qualification,
as a constant expression.
He claims that the compiler may choose to substitute the value at
compile time.
I tried to explain how the value can change (but I had to use pointers
to const int) which he shot down saying the comparison of examples is
not valid (it's not initializer and it involves pointers).

It is no longer a C program if you attempt to change an object defined
with the const qualifier.
Question2
Moreover, if we turn on optimization, (and assume that the real code
does not have any dead code like above), the compiler does not give any
error. Is the compiler correct after optimisation?

The compiler is neither correct nor incorrect, see below.
I think it is a bug in compiler but I need text in standard to prove
that.

No, it's not a bug in the compiler. But it is not portable behavior
either. Paragraph 10 of 6.6 consists of one sentence, which is "An
implementation may accept other forms of constant expressions."

So if your implementation happens to define the value of an object
defined with the const qualifier and an initializer, in scope, as a
constant expression, apparently the standard allows it to.
 
M

Marc Boyer

Le 19-04-2006 said:
My bad. I did not include complete text.
Let me include enough context and repost.

~> cat test.c
const int a = 10;
int main () {
static int b = a;
return b;
}

If the compiler can treat this 'a' as a compile time constant then this
should also compile.

It depend what you mean by 'the compiler can'. I think that, if
someone is able to write a C compiler, he would also be able to
write a augmented C compiler that, *in this special case*,
see that a is constant at compile time.

You should notice also that this is a valid C++ code,
but C++ does not require that static variable are initialized
by constant.

But in C, there is a standard that defines what must be accepted
by a C compiler. In this case, it may be detected because the value
of your const variable can be known at compile time By, in this case,
it should also accept
int main(){
int a=1;
static int b=a;
a--;
return a;
}

In this case, a is not any more a constant, but b is always
initialized with value 1.

And what about
const int a= random()?1:1;
or
const int a= random()?1:10/10;

What I mean is that, in the general case, it is hard (perhaps
undecidable in fact) to know when a const variable is a compile time
constant, and whan is is not. But, what is sure is that a constant
expression is a compile time constant. Then, the people who wrote
the standard have chosen to allow only initialisation of static
variable with constant expression.

This choice has the benefit of simplicity.

Marc Boyer
 

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

Forum statistics

Threads
473,731
Messages
2,569,432
Members
44,832
Latest member
GlennSmall

Latest Threads

Top