union field access and compatible types

[nicolas.sitbon]

Hi all,
My question is simple but I can't find the answer in the standard (C99
TC3).
is this permitted?

union
{
const void * ro;
void * rw;
}
variant;

void * buf = strdup("bla");
/* check allocation success */
variant.rw = buf.
puts(variant.ro);

Thanks by advance.

 

[Ben Bacarisse]

My question is simple but I can't find the answer in the standard (C99
TC3).
is this permitted?

union
{
const void * ro;
void * rw;
}
variant;

void * buf = strdup("bla");
/* check allocation success */
variant.rw = buf.
puts(variant.ro);

Yes. Pointers to const and non-const versions of the same type have
the same alignment and representation (6.2.5 p27). I don't see any
other pitfalls but maybe you were worrying about some other aspect of
the code.

How did this come up? I'd prefer to use a cast if I needed to change
the qualifiers on the target of a pointer. It makes this significant
change so much more apparent in the code so I imagine that this effect
is not the main purpose of this code fragment.

 

[nicolas.sitbon]

Yes.  Pointers to const and non-const versions of the same type have
the same alignment and representation (6.2.5 p27).  I don't see any
other pitfalls but maybe you were worrying about some other aspect of
the code.

How did this come up?  I'd prefer to use a cast if I needed to change
the qualifiers on the target of a pointer.  It makes this significant
change so much more apparent in the code so I imagine that this effect
is not the main purpose of this code fragment.

thanks for your answer, in fact, I don't need to change the
qualifiers, rather I'm looking for a way to create a generic data
structure that can can be used with const generic pointer or non const
generic pointer, preventing the user from casting if I would hard code
the qualifier. Think of a generic linked list where you can put const
data or variable data without casting because I would code void * data
and the user want to put a const char *. I hope it is clear, english
is not my mother tongue.

 

[Ben Bacarisse]

"nicolas.sitbon" <[email protected]> writes:

How did this come up?

It is best to snip sigs unless you are commenting on them.

thanks for your answer, in fact, I don't need to change the
qualifiers, rather I'm looking for a way to create a generic data
structure that can can be used with const generic pointer or non const
generic pointer, preventing the user from casting if I would hard code
the qualifier.

If you hard-code the const, why does the user have to cast?

Think of a generic linked list where you can put const
data or variable data without casting because I would code void * data
and the user want to put a const char *.

As a rule, generic container structures don't know enough about the
data to do anything much with it (they may pass it to a function but
that is about it). That makes me wonder why you need the non-const
data variant.

I hope it is clear, english is not my mother tongue.

Yes, perfectly clear. Your English is excellent. While on the
subject of language, pointers are problematic. You talk about a
"const generic pointer or non const generic pointer" but in fact the
pointer is not const qualified in either case, it is the data pointed
*to* that is marked const. This is not a complaint -- everyone drops
into shorthands like this all the time -- but sometimes it matters and
it pays to be vigilant for those cases where they might be a
misunderstanding.

 

[nicolas.sitbon]

If you hard-code the const, why does the user have to cast?

If I hard-code the const and the user put a non const data, it makes
no problem, but if he wants to get back his data that is now stored in
a pointer to const, he is forced to cast.

As a rule, generic container structures don't know enough about the
data to do anything much with it (they may pass it to a function but
that is about it).  That makes me wonder why you need the non-const
data variant.

see above, when you want to get back your data in a pointer to non
const data, you are forced to cast.

Yes, perfectly clear.  Your English is excellent.  While on the
subject of language, pointers are problematic.  You talk about a
"const generic pointer or non const generic pointer" but in fact the
pointer is not const qualified in either case, it is the data pointed
*to* that is marked const.  This is not a complaint -- everyone drops
into shorthands like this all the time -- but sometimes it matters and
it pays to be vigilant for those cases where they might be a
misunderstanding.

You're right, it's a bad shorthand.

 

[Ben Bacarisse]

If I hard-code the const and the user put a non const data, it makes
no problem, but if he wants to get back his data that is now stored in
a pointer to const, he is forced to cast.

True, but I see that as an advantage. I'd like to see the end result
using your idea before deciding, but I /think/ it will obfuscate when
the data is and is not const. Do you have any working code yet?

<snip>

 

[Tim Rentsch]

Hi all,
My question is simple but I can't find the answer in the standard (C99
TC3).
is this permitted?

union
{
const void * ro;
void * rw;
}
variant;

void * buf = strdup("bla");
/* check allocation success */
variant.rw = buf.
puts(variant.ro);

The short answer is yes. The two members involved have types
that are guaranteed to have the same representation and alignment
requirements. This is covered in section 6.2.5 paragraph 27.